Personalization – the use of machine learning models trained on individual-level data to tailor a product, prediction, or treatment to a specific person rather – is not a single phenomenon that artificial intelligence applies uniformly across the economy. Its success is governed by three measurable variables: 1. the cost of acquiring data, 2. the cost of a wrong prediction, and 3. the speed of the feedback loop between action and outcome.
Health care is simultaneously the sector with the strongest scientific justification for personalization and the sector most prone to expensive overpromising, since the same conditions that make personalization valuable also make it slow, costly, and difficult to validate.
Why Personalization by AI isn’t Fungible Across Sectors
The “personalization of everything” narrative incorrectly suggests that Netflix’s recommendation engine, Amazon’s product suggestions, and an AI-matched cancer therapy as points on the same continuum. Media and retail personalization succeeded quickly and cheaply because the underlying data is abundant and continuously regenerated by user behavior, a wrong recommendation costs the company almost nothing (the user simply scrolls past it), and the feedback loop between a suggestion and its outcome is measured in seconds. An industry like precision agriculture occupies a middle position: sensor and satellite data are becoming cheap, but the feedback loop is a full growing season, and an error costs real yield, which is why adoption has been steady rather than explosive.
Health care sits at the opposite pole: clinical and genomic data are expensive to generate, fragmented, and constrained by privacy regulation that has no analogue in ad-tech: a wrong prediction risks patient fatality rather than simply waste an advertising impression. In addition, the feedback loop connecting a treatment decision to its true outcome, survival, recurrence, or long-term toxicity, can run into years, far too slow to train and correct a model the way a recommendation algorithm corrects itself in real time.
Health care outcomes are surely one of the most impactful use for artificial intelligence, with personalization an important pathway. Earlier this summer, financial participants confirmed this view: in August, Moderna stock surged because its personalized mRNA cancer vaccine – AI helps identify the right targets for each patient – showed success in a pivotal trial. Of course, the story is larger than AI in medicine: investors cheered because it provided clinical validation that mRNA can work in cancer, not just infectious diseases.
Whether the optimism is warranted is yet to be seen, as real-world failures also do exist: between 2025 and 2022, IBM spent roughly $4 billion building Watson Health on the premise that a system trained the way Watson beat Jeopardy! could generalize to oncology. Internal documents later showed the system generating “unsafe and incorrect” treatment recommendations, with failures traced directly to training on a small number of synthetic cases that reflected one hospital’s practice patterns rather than diverse, real-world outcomes data. IBM sold off the unit’s data and analytics assets in 2022 for roughly $1 billion, a fraction of what had been invested.
Nonetheless, it’s becoming clear that artificial intelligence has the potential to personalize treatments and therefore significantly improve outcomes in a high impact sector like health care. Today’s Note identifies the companies and investment opportunities related to this theme.
Health Care Personalization: Who Does What
To evaluate “AI plus health care”, we immediately notice that a lower risk strategy is to own the “picks and shovels” (or infrastructure) of personalized medicine rather than the biotech companies that own the therapies. The table below highlights why this is the case, but put simply, biotech companies own the regulatory and clinical risk and commercial risk whereas the suppliers of the inputs to the final drug treatments, i.e. the picks and shovels, are relatively shielded from these unknowns.
Below, we separate the three layers that must each work for personalized medicine to reach a patient: the data that makes personalization possible, the software and diagnostics that turn that data into a decision, and the manufacturing capacity that actually produces a personalized product at scale.
1. Data Generation
When it comes to personalized medicine, the first step requires data generation companies that produce the raw genomic, single-cell, or spatial biology data, through sequencing instruments and sample-preparation tools, that all downstream personalization applications depend on. The clearest illustration of the importance of this layer occurred in August 2026, when Merck (MRK-US) and Moderna (MRNA-US) reported the first positive Phase 3 result for any individualized neoantigen therapy, in resected melanoma. The therapy is, in effect, one drug per patient: each dose requires that bioinformatics pipeline that selects which specific mutations to encode. This essential manufacturing input enables the therapy to exist and falls outside the category of discretionary diagnostic tests subject to payer discretion. Two companies in this space stand out:
- Illumina (ILMN-US) develops and manufactures life science tools and integrated systems for large-scale analysis of genetic variation and function. It is considered a mature, profitable incumbent whose sequencing platforms generate the raw genomic data every downstream application depends on. Second-quarter 2026 revenue grew 9.5% year-over-year, and the business is comfortably profitable.
- 10x Genomics (TXG-US) provides single-cell and spatial biology tools that capture the kind of granular data bulk sequencing cannot. Its 2026 revenue guidance implies roughly flat to modestly positive growth, but gross margin improved to 70% in the first quarter and net losses narrowed meaningfully. It is a smaller and more cyclical business than Illumina but one sitting directly upstream of the next generation of personalization applications: most of 10x’s workflows are ultimately sequenced on Illumina instruments, but Illumina has also built competing single-cell and spatial product lines.
2. Data-to-Decision
Data-to-decision companies turn raw data into an actionable clinical choice, such as which diagnostic test result, drug, or dose is right for a specific patient. This is the layer generating the most investor attention, partly because of its reliance on AI in specific products and partly because the category has adopted “AI” as a branding convention. The companies doing this work have divergent financial profiles, making for a tricky backdrop for investors:
- Natera (NTRA-US) is a diagnostics company, which engages in the discovery, development, and commercialization of genetic testing services. Second-quarter 2026 revenue grew 37%, management raised full-year revenue guidance and reiterated guidance for positive full-year cash flow, a rare combination of hypergrowth and a credible, near-term path to self-funding.
- Tempus AI (TEM-US) is a precision medicine company, historically centered on oncology, that runs two linked businesses: diagnostics and data. Diagnostics account for about three quarters of revenue, running genomic and molecular test on tumor samples. (about three-quarters of revenue). The data division keeps de-identified clinical and molecular patient records. It licenses that data, and the AI tools built on it, to pharmaceutical companies for drug discovery, trial design, and finding patients for trials. The company grew revenue 22% in the second quarter and posted a small profit, its first,
- Guardant Health (GH-US) is a “precision” oncology company, which provides treatment selection for cancer through the use of proprietary blood-based tests, vast data sets, and advanced analytics. In their most recent earnings results, they reported oncology test volume up 63% year-over-year, although net losses widened over the same period.
- Certara (CERT-US) provides software and technology-enabled services for drug developers including cell and gene therapy, clinical pharmacology, complex biologics, decision analytics and modeling. The company reported roughly flat second-quarter 2026 revenue and reported a small loss (compare to a profit gains in the previous quarters).
3. Manufacturing at Scale
Manufacturers physically produce personalized or small-batch biologic therapies (mRNA vaccines, cell and gene therapies). This is the bottleneck that determines whether personalized medicine can move beyond rare-disease niches into broader clinical use.
Bespoke, per-patient biologic manufactured in small batches is the structural opposite of the blockbuster-drug economics that built the modern pharmaceutical business model.
- Lonza (LONN-CH) is a Swiss company founded in 1897 that is a manufacturing partner for biopharmaceuticals. Historically, the company has “mass produced” popular treatments, but they now have a dedicated line of business develops, de-risk, commercializes, and scales emerging therapies. Sales growth was reported at over 20% in 2025, across mRNA, cell and gene therapy, and biologics manufacturing capacity.
The summary table above shows the large dispersion in the financial performance of the major actors in health care personalization. Revenue growth, gross margin, and profitability vary significantly; owning a basket of companies may be a more appropriate strategy than bearing single name risk.
With AI agents now roaming the virtual corridors of the services industry, it’s fair to ask how far this technology ultimately displaces workers across finance, legal, accounting, and HR functions at larger organizations. Agentic AI platforms let firms create and train agents in a way that effectively builds “digital twins” of entire services’ workforces. Where firms used to add headcount, often through temp staffing agencies, they can now nurture agents natively trained on their own context and workflows. In this not-so-utopian world, now at our doorstep, it’s fair to wonder whether temp staffing companies retain a durable role, and whether the sector’s historically controversial labor model becomes a smaller share of how large employers fill gaps.
The most rigorous data point on employer intent comes from the Adecco Group’s Workforce Trends 2026 Report, a survey of roughly 2,000 C-suite executives and 37,500 workers across 13 countries. Forty-five percent of business leaders expect AI agents to be fully integrated into workflows within twelve months, yet only 36% say their talent strategy clearly communicates that AI will create opportunities for employees rather than displace them, and just 22% are confident they are building the digital capabilities their workforce actually needs.
The gap cuts the other way too: 70% of non-managerial employees say they would be comfortable collaborating with AI agents, compared with only 39% of leaders who believe their staff feel that way. The implication is that AI agent execution risk sits with management confidence and change management, not worker resistance. PwC’s AI agent survey corroborates this pattern: 75% of executives report confidence in their AI agent strategy, but describe the near-term effect as redeploying staff to new tasks rather than eliminating headcount outright. Census Bureau firm-level tracking adds a useful reality check: among companies currently using AI, 87% report it displaces only a small number of tasks, with large-scale task replacement still confined to under 1% of firms.
Taken together, the evidence does not support a clean substitution thesis in which AI agents are displacing temporary or contingent labor at scale today. Instead, it points to a slower, uneven redesign of work in which the binding constraint is organizational trust and skills investment, not the technology’s raw capability to replace tasks currently filled by temp staffing.
Build vs. Buy, and the “Enshittification” Risk
Outsourcing to temp staffing versus building in-house AI agents is a classic build-versus-buy tradeoff, with each option carrying a distinct risk profile rather than one being categorically superior. Temp staffing converts labor into a variable cost, transfer screening, compliance, and turnover risk to the agency, and, critically, carries a network effect: a temp worker arrives having already seen how dozens of other employers solve the same problem, so each placement imports cross-company best practice and gives the employer an external market check on quality. A temp’s output is benchmarked against a market rate and a market skill pool, and underperformance is cheap to terminate.
In-house AI agents have no equivalent mechanism. They are trained and reinforced almost entirely on one company’s own data, KPIs, and management preferences, so they tend to optimize for pleasing the existing hierarchy rather than for ground truth. The upfront cost is largely sunk (data infrastructure, fine-tuning, integration, governance), but the marginal cost of each additional task approaches zero once built, which is the entire investment case for agentic automation. The catch is that there is no natural market discipline, forcing a reset the way a temp worker’s next assignment does. An agent fine-tuned on a company’s own historical decisions and success metrics will tend to reproduce and amplify that company’s existing blind spots and internal politics.
For investors evaluating a company’s AI agent buildout, the diligence question is not simply how much capex is going into it, but whether implementing agents can occur while avoiding enshittification – the idea that your employees start drinking the corporate kool aid a bit too much, i.e., whether the company has built independent evaluation benchmarks, external data grounding, and adversarial red teaming to substitute for the cross-employer network effect that in-house agents structurally lack.
ESG: The Temp Staffing Risks that AI Agents Sidestep
Temp staffing practices typically split legal employment (the agency) from operational control (the client), which creates accountability gaps. For instance, temp staffing companies have been historically accused of precarious pay and hours, wage theft, discrimination, weak safety training for high-risk placements, and even recruitment-fee debt bondage or forced labor among migrant workers, alongside erosion of union bargaining power. These risks translate financially into litigation exposure, joint-employer liability, regulatory fines, higher insurance costs, and lost contracts, especially as supply-chain due-diligence rules tighten globally. By deploying AI agents in lieu of temp workers, a company can sidestep this entire risk category.
How Big is the Temp Market, Really?
Temp employment peaked at 2.5% of total U.S. non-farm employment in 2020 and now stands at roughly 1.5% of the employed workforce. With job openings now back near 2019 levels, the current low level of temp staffing may be explained as much by the temporary sluggishness of the broader job market as by any AI-driven substitution effect.
Looking at the past six months, BLS data offers a more nuanced picture than the popular “AI is gutting white-collar jobs” narrative suggests: temp staffing employment has in fact increased by 34,800 jobs (+1.4%).
That said, there is considerable dispersion across industries. Temp staffing for financial activities and information services show genuine, broadening contraction, down 57,000 (-0.6%) and 35,000 (-1.2%) respectively over six months. This contraction is concentrated in insurance carriers, credit intermediation, and computer systems design, precisely the automatable back-office and technical functions linked to the fastest AI adoption rates. Legal services and accounting, by contrast, show no aggregate displacement at all: legal payrolls are up 11,300 (+0.9%) and accounting is roughly flat over the same window. Healthcare remains the only sector still driving overall job growth (+2.9% YoY through March 2026, across all healthcare employment, not just temp placements), while temp staffing specifically has stabilized after last year’s slide, both supporting the case that clinical staffing demand is more insulated from AI substitution than white-collar temp placement. The Dallas Federal Reserve and Anthropic research suggest the likely explanation is that AI’s labor impact so far shows up as a freeze in entry-level hiring, associates, junior auditors, rather than incumbent layoffs, a dynamic that is invisible in aggregate industry headcount data but very real for the cohort it affects.
Agents Are Priced In
Publicly listed temp staffing companies have derated substantially over the past year and underperformed the broader indices. Thin profit margins, flattening revenue growth, and a slower overall job market have taken the air out of the post-Covid staffing bubble.
Within the group, Robert Half looks most vulnerable to agentic AI, given its strong exposure to back-office financial functions, precisely the roles most likely to be automated within the next five years. AMN Healthcare, by contrast, provides staff that remain far harder to replace given the clinical nature of the work, which should make it more resilient to AI displacement, at least until humanoid robotics make meaningful inroads into care settings.
From an operational standpoint, most staffing firms now have an explicit AI strategy, often centered on using AI tools to streamline their own HR and placement processes. In addition, they will increasingly compete for client budgets with management consultants and IT integrators, which are also pitching “build your agentic workforce” engagements to the same corporate buyers, as the strategies highlighted in the table below suggest.
Equity market participants have reacted quickly to the disruption narrative. Short interest in Robert Half has reached 24%, reflecting a deep-seated conviction that AI will permanently erode demand for white-collar placement services, even as domestic staffing trends are inflecting toward recovery, a combination that could set up a short squeeze as sentiment catches up to improving operational realities. ManpowerGroup carries more moderate short interest of roughly 15%.
The Paradox at the Heart of the Grid
AI is both a source of the problem and the most credible solution to it. Global data center electricity consumption reached approximately 415 TWh in 2024, growing at 12% per year since 2017, and the IEA projects this figure could reach up to 1,200 TWh by 2030. A single AI-focused data center today consumes as much electricity as 100,000 households. At the same time, the electricity grid is increasingly unreliable: the North American Electric Reliability Corporation (NERC) flagged rising shortfall risk under both extreme and normal conditions in its latest Long-Term Reliability Assessment, and grid outages already cost U.S. businesses roughly $150 billion annually.
The irony is that the same computing paradigm generating this demand surge is also the most capable tool available to manage it. The question for investors is not whether AI will change how electricity is produced, distributed, and consumed. It will. The question is which companies in the stack are positioned to capture durable, recurring revenue from that shift.
Two Domains, Two Mechanisms
Grid Optimization
AI’s most direct contribution to grid efficiency operates at the level of real-time sensing, forecasting, and control. Smart meters and IoT sensors generate continuous operational data; AI algorithms process that data to build a live picture of grid state and issue control signals in milliseconds, far faster than human operators can respond.
The IEA’s 2025 “Energy and AI” report identifies AI-based fault detection as one of the highest-impact near-term applications, with documented reductions in outage durations of 30-50%. Given the $150 billion annual cost of outages, a 30% improvement represents tens of billions in recoverable economic value, even before accounting for carbon savings from avoided backup generation.
Virtual Power Plants (VPPs) represent the next frontier. These systems aggregate small-scale distributed resources, including behind-the-meter solar, batteries, demand-response loads, and EV chargers, and coordinate them through AI to respond to grid signals in real time. The U.S. Department of Energy estimates that tripling VPP scale in America alone could reduce grid costs by $10 billion per year, primarily by deferring expensive peak capacity additions. As the share of intermittent renewables grows, the economic case for AI-driven aggregation and dispatch becomes progressively stronger.
Industrial Process Optimization
Industry and commercial buildings together account for roughly 64% of U.S. electricity demand, making them the largest addressable market for AI efficiency tools. Here, the applications are more granular but no less material. We see 4 key industrial sectors where AI could markedly optimize production and reduce energy costs:
Steel: Steel is an instructive case. ArcelorMittal (MT-US) partnered with energy AI specialist Energiency to deploy machine learning models that predict and control energy consumption continuously across its smelting operations, achieving 2-5% annual energy savings on an energy bill that represents roughly one quarter of total production costs. A separate deployment using C3.ai’s energy management platform at a different major steel producer generated $14 million in annual energy cost savings at a single hot roll mill, alongside a 40 MW reduction in monthly utility demand charges.
Cement : Cement, which another thermodynamically constrained industry, can also achieve more efficiency using AI. A 2026 peer-reviewed analysis found 3-9% net energy and CO2 reductions from AI-optimized kiln controls, after explicitly subtracting the energy cost of running the AI itself. For an industry running near thermodynamic limits, 5% is financially significant and competes favorably with far more expensive physical decarbonization options.
Cooling : For cooling and facility management, the IEA independently cites a 23% cooling energy saving from AI-optimized controls as a representative industrial deployment benchmark. Across optimizable industrial subsystems more broadly, documented savings commonly reach 20-30% of the targeted system’s energy consumption, though total plant-level savings are typically 5-15% given that no single subsystem represents the entire load.
Chemicals: Chemical production offers one of the highest AI optimization ceilings in heavy industry. Energy costs represent 40-60% of variable production costs at specialty chemical and petrochemical facilities, and process complexity (multi-variable reactor management, distillation train sequencing, heat integration across unit operations) creates exactly the conditions where narrow ML models identify non-obvious savings that rule-based controls miss. AI-optimized process control at chemical plants has demonstrated sustained energy reductions of 3-8% on targeted units, with the strongest results on distillation columns where temperature and pressure gradients interact in ways that respond well to continuous model retraining. At industrial scale, a 5% energy reduction at a large petrochemical complex running 200-400 GWh annually translates to $3-8 million per year at current industrial power prices, with no capital expenditure on physical equipment.
A useful way to frame AI’s efficiency contribution is to ask what energy is consumed by the AI relative to what it saves. The answer varies dramatically depending on the type of AI deployed, and investors who treat all AI as equivalent in energy intensity will systematically misread this opportunity.
Industrial process optimization does not rely on large language models. It uses narrow, purpose-built machine learning models, often small enough to run on edge hardware or modest on-premise servers. A factory consuming 100 GWh per year that deploys a continuous ML optimization model on local hardware might consume 0.01-0.1 GWh annually on inference. If that model saves 5% of plant energy, or 5 GWh, the implied leverage ratio is 50:1 to 500:1 in energy saved per unit of AI energy consumed. This is not marginal: it is strongly net positive, and it is precisely why the IEA’s analysis of well-scoped industrial applications concludes that AI is energy-additive rather than energy-neutral.
The energy paradox narrative applies to large language models and generative AI, where a single query consumes 3-30 times the energy of a typical web search, and training runs consume electricity equivalent to small towns. But these architectures are not what drives a cement kiln or optimizes a steel furnace. The AI that matters most for industrial efficiency is lean, specialized, and deployed close to the process. The companies building and selling that software are not competing against data center scale-up; they are competing against the baseline of unoptimized industrial operations, and winning.
Where the Investment Opportunity Sits
The most defensible long-term positions across this theme sit with the large-cap infrastructure companies in the coverage universe. Their combination of certified field service networks, multi-decade customer relationships, and systems integration depth creates a moat that pure software entrants cannot replicate quickly.
Schneider Electric (SU-FR) is the most fully integrated play in the coverage universe. Its EcoStruxure platform spans building management, grid automation, and industrial control as a unified software stack, and recurring software and services revenue is growing as a share of EBITA. The company’s dominant position in data center power distribution gives it direct exposure to the AI infrastructure build-out, independent of any efficiency software thesis.
GE Vernova (GEV-US) sells grid software through a customer base it already serves with physical grid hardware and services contracts, which substantially lowers sales friction. GridOS, GEV’s advanced distribution management system, addresses dispatch optimization, fault detection, and VPP orchestration directly, covering the three highest-value AI grid applications identified by the IEA.
Eaton (ETN-US) is the most hardware-proximate name in the group, but its power management franchise is increasingly software-defined. Its fastest-growing segments (data center and industrial power quality) are exactly where AI-driven efficiency generates the clearest and most measurable ROI for end customers.
ABB (ABBN-CH) brings the deepest industrial automation capability in the group. Its ABB Ability digital suite is most concentrated in process industries (chemicals, mining, pulp and paper) where energy optimization ROI is highest and willingness to pay for proven software is well established. ABB’s multi-year portfolio restructuring has improved capital returns without eroding its technical position.
C3.ai (AI-US) is the primary enterprise software plays with revenue concentrated in industrial energy management. C3.ai’s documented results in steel energy forecasting are among the most quantified in the public domain.
Itron (ITRI-US) and Landis+Gyr (LAND-CH) offer small-cap exposure to the smart metering and grid sensing layer, which is foundational infrastructure without which AI optimization algorithms have no data to work with. Landis+Gyr is exclusively a smart grid company; Itron earns approximately 67% of revenues from smart grid products. Both carry higher execution risk than the large-cap names, but provide concentrated exposure to grid digitization spending.
Vertiv (VRT-US) manufactures the thermal management systems and power distribution infrastructure that sit inside data centers and determine how efficiently electricity is converted into compute. As AI chips generate five to ten times more heat per rack than conventional servers, the hardware required to cool them becomes more complex and more expensive, expanding Vertiv’s addressable market with every GPU cluster installed.
The YTD performance dispersion within the coverage universe is instructive. Vertiv’s 75% gain reflects the market’s pricing of several years of AI cooling demand; the thesis is sound but execution risk is now fully reflected in the price. The large-cap infrastructure names (Schneider, GE Vernova, Eaton, and ABB) have been relative laggards and offer better risk-adjusted entry given backlog-supported earnings visibility and growing recurring revenue. The pure software layer (AI-US, AZPN-US) requires higher conviction on near-term revenue inflection; C3.ai in particular has not yet demonstrated a durable path to profitability despite strong individual customer case studies.
Sustainability-oriented equity mandates have spent a decade building around a practical constraint: ESG data for small cap companies is less available and less reliable than its large cap counterparts. Third-party ratings agencies often concentrate their analytical resources on large caps, where disclosure is richer and the revenue-per-company economics make coverage worthwhile. The consequence has been a structural underweight in small caps across ESG portfolios that is a data artifact rather than a deliberate investment judgment.
Over the twelve months ending June 30, 2026, that underweight cost money. The Russell 2000 returned 40% and in the first half of 2026 alone, the Russell 2000 outperformed the S&P 500 by more than 12 percentage points, its best start to a year since 1991.
The deeper issue for sustainability mandates is that low scores and missing scores are treated as equivalent signals of ESG risk, when in most cases they reflect nothing more than the disclosure practices of a company with less staff dedicated to the investor relations function.
We propose a different entry point: rather than relying on ratings coverage that does not exist, we identify small cap companies where observable, verifiable sustainability events have recently occurred or are imminent – first ESG reports, CDP initiation, SBTi validation, board-level governance upgrades, etc. These triggers do not require proprietary data infrastructure. They require reading company filings, sustainability reports, and proxy statements, which is what active small cap managers already do. The re-rating that follows is systematic and documented: research published in 2025 found that MSCI ESG rating changes drive abnormal returns over multiple months after the change, with institutional investor repositioning as the primary transmission mechanism.
The implication is actionable: in small caps, identify the companies that are publishing the required data and own them before the agencies react.
The Investment Thesis
When a company improves its sustainability posture in a material, verifiable way, the re-rating tends to follow a predictable sequence. First, the company appears in sustainability-focused screening databases (MSCI ESG, Sustainalytics, ISS) with an upgraded score. This typically takes 12 to 24 months from the triggering event. Second, institutional mandates with ESG minimum screens begin to include the name in their eligible universe, generating incremental demand that is largely independent of traditional fundamental catalysts. Third, coverage by sustainability-focused analysts follows, compounding the effect.
Companies know that improving their sustainability posture is good for their stock price; the survey results showed in the chart below highlight that companies believe the main opportunities related to ESG ratings are acquiring and retaining interest in their stock.
It is logical that the re-rating premium would be amplified in small caps relative to large caps because the starting ownership base is dominated by generalist managers who have not priced sustainability improvement into their valuation. There are some practical signals that investors can track to anticipate a re-rating:
- First publication of a sustainability report (particularly if structured around SASB materiality topics)
- Adoption of science-based emissions targets or commitment to SBTi validation
- Board composition changes (addition of an independent director with sustainability credentials; board-level committee dedicated to ESG oversight)
- First participation in CDP disclosure (Carbon Disclosure Project questionnaire responses carry significant weight in downstream rating models)
- Emissions intensity reduction documented over two consecutive years (a single year is insufficient signal; agencies require a trend)
Where to Look
The most fertile ground for the improvers investment thesis sits in sectors where SASB-defined material sustainability topics are operationally manageable at scale for a small company, but where disclosure has historically been minimal. Industrial services, specialty chemicals, and building products are all areas where energy intensity, waste management, and worker health and safety are financially material (under SASB) and operationally addressable without the capital requirements that make transition difficult. A small industrial company that installs remote energy monitoring across its facilities, reduces its recordable incident rate, and publishes a first SASB-aligned sustainability report is doing all three things at once – and doing them in a way that is both financially rational and likely to trigger a meaningful ESG re-rating within two years.
The Counter-Argument
A skeptic may point to three things that undermine the arguments above:
- The re-rating timeline of 12 to 24 months is long, and the path is uncertain: a company can publish a sustainability report without triggering any rating agency review if the report is judged insufficiently substantive.
- The sustainability reporting landscape is not frictionless: first-time reporters routinely publish reports of uneven quality, and a poorly constructed first report can lock in a low score rather than improve it.
- The liquidity premium embedded in small caps is not always fully compensated: if a sustainability-driven re-rating coincides with a broad risk-off environment, the demand lift from ESG-screened mandates may be insufficient to offset market-wide selling.
These are real risks, but they can be mitigated by concentrating on companies whose underlying operational improvements are genuine and financially material, rather than those pursuing disclosure as a cosmetic exercise. For example, an emissions intensity trend that tracks with capital investment in energy efficiency is credible, while an emissions intensity trend that appears only in the sustainability report – without corresponding capex – is not.
Investment Implications
For small cap equity managers, a systematic scan for the leading indicators listed above (sustainability report publication, CDP participation, SBTi commitment, and board changes) applied across the portfolio universe can generates a watchlist of names likely to re-rate within 24 months.
In the section below, we highlight companies that qualify for a “sustainability re-rating” based on the criteria identified above.
Small Cap Sustainability Improvers Ripe for Re-Rating
- Onterris Inc (ONT-US)
Market cap: $750M | SASB sector: Waste Management / Environmental Services
Onterris, an environmental and engineering consulting firm, had its near-term SBTi target and long-term net-zero target formally validated by the Science Based Targets initiative in early 2025. SBTi validation is arguably the most credible sustainability event a company can execute. Ratings agencies like MSCI operate on annual review cycles, meaning a company that validates SBTi targets today is unlikely to see a formal rating upgrade until the next scheduled assessment, usually annually.
Onterris has published annual SASB-aligned sustainability reports for three consecutive years and participates in CDP. Revenue approached $900M in 2025 and 2026 guidance is above that level. The business is growing into the disclosure infrastructure rather than shrinking, which gives the rating upgrade a durable foundation.
It is also worth noting that until recently, the company operated under the name Montrose Environmental Group, but rebranded as Onterris in April 2026. Corporate identity resets of this kind often precede a broader ESG narrative refresh and tend to attract new analyst attention.
- Dexterra Group (DXT-CA)
Market cap: CAD $880M | SASB sector: Professional & Commercial Services / Industrial Support Services
Dexterra provides supports services for the creation, management and operation of infrastructure. The company published its first-ever comprehensive, GRI-aligned sustainability report in March 2025, including a quantified Scope 1 and Scope 2 baseline following the GHG Protocol Corporate Standard. Prior reports were informal communications without a quantified emissions baseline.. Importantly, the business is financially robust: Q1 2026 revenue grew 15% year on year, and earnings beat estimates by approximately 40%. This removes the risk that sustainability commitments get deprioritized because of financial stress.
- John B. Sanfilippo & Son (JBSS-US)
Market cap: ~$990M | SASB sector: Processed Food
JBSS, a leading American processor of tree nuts and related snacks, has quietly built one of the more comprehensive environmental disclosure programs among U.S. food small caps without having published a formal consolidated sustainability report. The company participates in CDP across all three disclosure streams simultaneously (climate change, water security, and forests), which is rare at this market cap. The company discloses scope 1, 2 and 3 emissions. In fiscal year 2025, total emissions intensity declined 8.3%. The company updated its Environmental Policy in May 2024, conducts water risk analysis using WRI’s Aqueduct Atlas across all six production facilities, and holds Energy Star Partner of the Year recognition. It reached 89.5% recyclable packaging by weight. SASB material topics for processed food (energy, water management, packaging waste, food safety) are each being actively addressed. The re-rating catalyst here is specific: MSCI and Sustainalytics currently rate JBSS without a published PDF sustainability report to anchor on, meaning the quantitative data sitting on its website is likely being partially discounted or missed. Publication of a first formal consolidated report would immediately elevate the company’s data quality score.
- Ag Growth International (AFN-CA)
Market cap: CAD $450M | SASB sector: Industrial Machinery & Goods
AGI makes grain handling, aeration, and storage equipment, placing it in a sustainability-adjacent commercial position (food security infrastructure) that creates natural demand for ESG transparency. Their 2024 Sustainability Report, released in June 2025, was the first to include a SASB Index and a standalone TCFD-aligned supplemental disclosure, neither of which appeared in the 2023 report. This is a meaningful escalation from narrative-only sustainability communication to framework-indexed disclosure, which is exactly what most ratings agencies require to move from qualitative to quantitative scoring. The Board operates a Governance and Sustainability Committee. On the quantitative side, AGI set a target in 2023 to reduce Scope 1 and 2 GHG emissions intensity by 25% by 2030 against a 2021 baseline; in 2024, it had already achieved a 16% reduction, keeping the trajectory on track. Safety performance improved sharply: lost-time incident rate down 46% and total recordable incident rate down 49% year on year, both SASB material metrics for this sector.
It is important to note that the names highlighted here are cited uniquely based on an expectation that financial market participants will latently give credit to their sustainability efforts; For example, AGI carries a highly leveraged balance sheet and has been under financial stress from agricultural equipment cycle headwinds: the sustainability re-rating is an independent catalyst, but the equity thesis depends on stabilization of the underlying business.
In the table below, we summarize the main sustainability re-rating triggers for the aforementioned companies, plus six more that we believe could benefit from the same dynamic.
The sustainable investment industry has spent two years debating whether defence is a sustainable investment. That is the wrong question. Whether Rheinmetall qualifies for an Article 9 fund misses what actually matters: whether investors have the analytical tools to hold defence companies to account for what they make, who their customers are, and what happens when their products reach the battlefield. Responsibility, rather than sustainability, is the appropriate standard, and it is a more demanding one than any taxonomy classification.
Data published this month records 65 active armed conflicts in 2025, the highest count since records began in 1946, with 244,600 people killed in organised violence, the second deadliest total since the Rwanda genocide in 1994. Eight of those conflicts are interstate wars, the most since World War II.
As a logical consequence, world military expenditure reached $2.8 trillion in 2025, up 41% since 2015, with forecasts pointing to $6.6 trillion by 2035. Europe’s ReArm Europe plan commits €800 billion; the United Kingdom has pledged 5% of GDP in defence spending by 2035. The capital flows are real, growing, and unlikely to reverse whether or not an Article 9 fund knows what to call them.
This note builds on the SMS framework introduced in our March 2025 note “The European Defense Offensive: From Peace Dividend to War Premium”, which identified five screening criteria (Controversial Weapons Exclusions, Allied Revenue Alignment, Government Independence, R&D Investment, and Strategic Defence Orientation) and a cohort of European specialists we termed the second wave.
Between July 2022 and the publication of our March 2025 European Defence Offensive, the SMS European Defence Specialists Basket delivered approximately 550% in total return. That performance continued for some time but has corrected heavily since the February 2026 peak. Over the fifteen months since publication, the basket, the S&P 500, and the American Defence Primes delivered 28%, 27%, and 25% respectively – effectively the same trade. The valuation re-rating of underowned European specialists as defence budgets re-accelerated is complete. The entry point that defined the first and second waves is gone.
The chart above is the investment case for this note: the opportunity that justified the first note has closed, and a new one has opened. What follows is where the next leg of returns will come from.
Red Lines That No Longer Hold
There was a time when a blanket defence exclusion was defensible. The sector consisted predominantly of identifiable end-products: tanks, artillery, fighter jets, submarines. Revenue streams were opaque but outputs were not. An investor that decided it did not want to own defence companies could draw a clean line around the category and walk away without much analytical strain.
That is no longer the case. The reason is dual-use, a property that once described finished technology products but now runs the entire length of the value chain, from raw materials to drones.
Upstream, the rare earth elements that power an EV motor, a wind turbine, or a smartphone are chemically identical to those in a drone motor, a missile guidance system, or an electronic warfare suite. Neodymium is neodymium. MP Materials (MP-US) is the only integrated rare earth mine and mill in the United States, indispensable to the domestic clean energy supply chain AND is simultaneously supplying the Pentagon under a ten-year offtake agreement backed by $400 million in government investment. An exclusionary policy that excludes MP Materials while retaining clean energy exposure would apply different standards to identical atoms depending on who writes the purchase order. Some investors will make that choice deliberately, as an expression of values. The point is that it is a values choice, not an analytical finding, and the clean line it tries to draw no longer maps cleanly onto the physical world.
Downstream, the same logic applies in reverse. The satellite constellation that provides broadband internet provides battlefield targeting. The AI platform that optimises logistics for a retailer optimises resupply for a military base. The same imaging system that monitors agricultural land monitors conflict zones. Revenue segmentation by end-use is rarely disclosed at the granularity a meaningful screen requires, and when it is, it rarely captures what happens to the technology once it leaves the company.
This is why the third wave which we discussed in this week’s note is different in character from the first and second.
The Rare Earth Elements (REE) Megatrend
Permanent magnets made from neodymium, praseodymium, dysprosium, and terbium are the enabling component in precision guidance systems, drone motors, missile actuators, electronic warfare equipment, and submarine propulsion. Demand for magnet rare earths has doubled since 2015, with defence among the structural growth drivers alongside EVs and wind. China processes approximately 85-95% of these materials at the midstream metallization and separation stage – a share the IEA projects will remain at 75% by 2035 even under an optimistic build-out scenario.

We have covered the OECD REE investment case in depth across three prior notes, beginning with “Critical Minerals, Critical Moves” (May 2025) and “Rare Opportunities in Rare Earths” (November 2025), and most recently in “From Transition Play to Arms Race: A Critical Materials Reset” (May 28, 2026) – published days before China’s June escalation. Readers seeking the full company-level analysis of the OECD REE basket should refer to these notes. What follows is specific to the defence demand driver and the regulatory timeline it has introduced.
That pressure reached a new peak just a couple weeks ago, in the week of June 15, 2026. The G7 set a target of no single country supplying more than 60% of each member’s rare-earth imports by 2030; the European Council proposed a formal de-risking mechanism; and China responded on June 22 with targeted export controls on US defence companies and rare-earth mining firms. Washington had already anticipated this: from January 1, 2027, a federal rule (DFARS) bans the Pentagon from procuring any system containing permanent magnets originating from China, a deadline China’s own controls have now made considerably harder to ignore. Active inventory depletion sharpens the timeline: the US has reportedly consumed approximately 45% of its Precision Strike Missile stockpile and 30% of its Tomahawk inventory, and restocking every unit requires REE-dependent components that cannot contain Chinese-origin material after that date.
The bottleneck is not mining but the midstream. A separation facility without downstream magnet makers has no local market, and magnet makers need committed offtake from defence, automotive, wind, and industrial customers simultaneously. The mine-to-magnet loop is only investable when all links close at once. The gap between stated policy ambition and physical processing capacity is precisely where the returns are.
The Investable OECD REE Basket
Most of the names below will be familiar to readers of our prior REE note series. Our conviction on the core positions is unchanged.
The three core positions remain MP Materials (MP-US), Lynas Rare Earths (LYC-AU), and Energy Fuels (UUUU-US). MP Materials is the only integrated US rare earth mine and mill, backed by a $400 million Pentagon investment, with REE separation commissioning at Mountain Pass by mid-2026. Lynas, the largest non-Chinese REE company at $13.5 billion, is completing its Texas REE separation facility with DoD funding. Energy Fuels received a $725 million conditional DoD loan commitment on June 18 and disclosed the $1.9 billion acquisition of Vacuumschmelze (VAC), the German precision magnet manufacturer. VAC holds existing qualification relationships with defence OEMs, bypassing the multi-year approval cycles new entrants face. The combined entity is the only OECD operation with a credible mine-to-magnet platform connecting US processing to European defence manufacturing. At $3.5 billion market cap, it looks underpriced relative to the strategic value of the VAC qualification book.
Two smaller names address specific gaps the core three do not cover. NEO Performance Materials (NEO-CA) is ramping its Estonia magnet factory to 2,000 tpa by end-2026, entirely outside Chinese processing influence. REalloys (ALOY-US) is building the largest REE metallization facility outside China in partnership with the Saskatchewan Research Council, with a 15-year offtake from the Tanbreez project in Greenland. It is important to note that REalloys came to market via reverse merger and is the highest-risk name in the basket. 5N Plus (VNP-CA) rounds out the basket with exposure to gallium, germanium, indium, and bismuth – the four materials subject to China’s August 2023 export controls and directly applied in compound semiconductors, radar, and night vision systems.
The REE basket has delivered; but unlike the European defence name re-rating, which converged to S&P 500 performance after March 2025, we do not believe this trade is over. The structural drivers – China’s escalating export control posture, a federal procurement ban on Chinese-origin magnets effective January 2027, and a supply chain buildout still measured in years rather than quarters – have not been resolved by the price action.
Third Defence Wave: Beyond the Primes
The REE basket addresses the supply side of the thesis. The names below address the demand side: platform specialists, systems integrators, and software-adjacent companies whose growth directly drives REE demand. What distinguishes them from the first and second wave is selectivity, not valuation re-rating – that trade is complete. Defence budgets are accelerating in unmanned systems, electronic warfare, naval domain awareness, and AI-assisted decision-making. The five names below are positioned where the spending is going and have not yet been fully absorbed by institutional capital.
- Hensoldt (HAG-DE) is Germany’s pure-play radar, electronic warfare, and optronics specialist. Its TRML-4D radar and Kalaetron electronic warfare suite are on order across multiple NATO allies. Revenue for FY2025 was approximately €2.1 billion against a backlog exceeding three times that figure.
- Renk Group (RENK-DE) is the sole supplier of tank transmissions for the Leopard 2 and propulsion systems for Germany’s Type 212 submarines, with no direct European competitor in naval propulsion at its scale. Revenue is approximately €1.1 billion with an order book accelerating across European navies and armoured forces. The demand trajectory is the strongest of any name in this section.
- Exail Technologies (EXA-FR) operates in the fastest-growing segment of naval spending: underwater domain awareness. Its DriX unmanned surface vessel and Gaps autonomous underwater vehicle serve both oceanographic and military customers, with NATO’s post-Ukraine emphasis on submarine detection providing a direct demand catalyst. The civilian revenue base diversifies against defence budget cyclicality.
- Kratos Defense (KTOS-US) is the closest pure-play on the affordable autonomous systems thesis: drones cheap enough to deploy in volume and lose in combat. This technology sits in a sub-sector where the US Air Force has no incumbent prime to protect. Revenue for FY2025 was approximately $1.2 billion.
- Booz Allen Hamilton (BAH-US) is the primary beneficiary of AI adoption within the US defence and intelligence establishment, bringing the cleared workforce and contract vehicles that pure technology platforms cannot replicate. Revenue for FY2025 was approximately $11.5 billion. The thesis is durable margin expansion as AI tooling reduces contract delivery costs. Government contractor risk is endemic to the category and priced into the current multiple.
Most of the basket has underperformed the MSCI World in the last year. That underperformance is the entry point, not a reason for caution.
The structural case has not deteriorated. It has strengthened. The Bundeswehr’s €100 billion fund is disbursing. NATO navies are ordering. The federal procurement ban on Chinese-origin rare earth magnets takes effect in seven months. Hensoldt and Renk enter this note’s coverage at a correction that their underlying order books do not justify. Booz Allen has been caught in the crossfire of US government IT spending reviews that are, by design, temporary.
The first and second wave re-rated on valuation. The third wave will re-rate on earnings, driven by a procurement cycle accelerating into the specific sub-sectors where these names face no credible competition from the large primes. The gap between the chart above and the order books behind these companies is where the next leg of returns is.
In just the last week, OpenAI (ChatGPT) has come under scrutiny for failing to report the Tumbler Ridge mass killer while Anthropic (Claude) was in a standoff with the U.S. government for refusing to remove restrictions on how the military can use its models; it may seem like the wild west in the AI industry.
To be sure, artificial intelligence capabilities are accelerating far faster than legal frameworks: in the U.S. and Canada, there are no federal regulations outlining any prohibitions on AI or surrounding the safety of the technology. The European Union passed the AI act in 2024, with most high-risk AI obligations only taking effect in 2027.
This leaves the private sector largely in charge of overseeing their own AI activities. While this may seem extremely imprudent, it creates a significant opportunity for the testing, inspection and certification (TIC) industry.
What is TIC?
The TIC sector consists of conformity assessment bodies who provide services ranging from auditing and inspection to testing, verification, quality assurance and certification. Many TIC companies were founded in the 18th and 19th century to provide information to shipping underwriters about the condition of ships and equipment. Today, TIC companies may be most associated with standards like ISO 9001, though their operations are far reaching and span across industries that are at the core of sustainability mandates: environmental protection, food, crop, health science, road safety, etc.
Indeed, we wrote about the sustainability merits of TIC companies in previous research (please see our note “Testing the (TIC) Market”, June 6, 2024). Given that TIC companies are paid to ensure the quality of others, it should not be a surprise that the household names in this industry have impeccable ESG records.
TIC companies – the majors include Bureau Veritas (BVI-FR), SGS (SGSN-CH) and Intertek (ITRK-GB) – are considered to operate in staid business lines; their stock prices tend to be highly correlated and have a beta to the global equity market of well under one. This is no doubt because many of the companies derive significant revenue from the public sector and because it is difficult to reduce expenditure on testing and assurance, even during economic downturns. These companies operate with high moats: relationships and reputation are crucial.
The TIC industry typically grows at a reliable 3 – 4% per year, avoiding major dips, but rarely experiencing excessive growth spurts; most third-party market insight reports forecast 3% CAGR for the 2026 – 2030 period. We believe these underestimate the role of AI, not least through novel new business lines like AI assurance.
TIC and AI: An Underpriced Opportunity
Historically, the TIC industry has thrived by monetizing both regulation complexity and safety requirements. AI is likely to massively increase both of these things as governments will eventually require proof that AI is safe, non-biased, cybersecure, and explainable. This creates a new industry involving the auditing of algorithms: AI assurance.
There are already new standards emerging for AI, such as ISO 42001, which is an ISO standard for AI management systems. Many TIC companies have received accreditation to certify companies under ISO 42001 thus enabling global AI audits.
Since the standard is relatively new – officially launched in December 2023 – no information is yet available on how many companies have implemented the standard. However, there are over 1.3 million companies that use ISO 9001, the global benchmark in quality management systems. The market size for AI management is likely to be equivalent.
TIC companies are just getting started when it comes to capitalizing on AI certification opportunities. SGS acquired AI certification specialist CertX in late 2024 specifically to build digital trust capabilities. They then joined NVIDIA’s Halos AI Systems Inspection Lab ecosystem, which is an inspection framework for physical AI systems (autonomous vehicles, robotics, industrial AI). The lab is intended to integrate functional safety, cybersecurity and AI assurance, provide standardized inspection reports and accelerate certification for AI systems globally,
Since NVIDIA cannot certify every AI system themselves without losing neutrality, independent certification via companies like SGS provides regulatory legitimacy, liability protection, and standardized compliance pathways.
SGS’ role is to verify that AI systems using NVIDIA platforms are safe enough for deployment: NVIDIA builds the AI engine and SGS certifies it is safe to operate in society.
In this sense, the potential role for TIC companies can be viewed as the equivalent of the big four auditors for accounting rules. Just like accountants, AI auditors will be needed across industries and sectors: AI deployment will likely require third-party verification across regulated sectors such as autonomous driving, medical AI, industrial robotics, defense systems, smart infrastructure, data centers, etc.
It is likely that established TIC companies will dominate the third-party verification (auditing) market because they already have accreditation authority, regulator relationships and global audit networks; they already have the credibility and reputation to provide this trust layer.
TIC Upside
Investors currently price TIC companies as slow industrial service providers, reflecting stable dividends and modest growth assumptions. But AI is opening new streams of revenue including recurring AI certification and embedded partnerships with hyperscalers like NVIDIA: TIC entities may just become mandatory infrastructure for AI deployment and regulation.
Indeed, once government regulations inevitably fall into place (e.g. the EU AI Act, automotive AI safety, etc.) demand for AI assurance services is likely to rise exponentially.
In this environment, a re-rating of TIC companies is possible with several possible triggers including:
- EU AI Act enforcement requiring third-party audits.
- Hyperscaler partnerships like the one announced between SGS-NVIDIA
- Segment reporting whereby TIC companies begin breaking out revenue tied to AI trust.
- M&A wave as TIC firms acquire AI governance startups like SGS’ acquisition of CertX
Note that so far, our bullish case for TIC and AI outlined above only hinges on the possibility of new AI revenue streams becoming available. TIC companies present solid use cases for artificial intelligence given that most aspects of testing, certification and inspection involves processing large amounts of data; machine learning and computer vision are already automating high-precision inspections, predictive analytics are catching defects earlier and generative AI can streamline report generation and anomaly detection. In other words, we believe that AI assurance is additive to TIC revenues while traditional business lines will also benefit from productivity improvements.
Our thesis that AI will bring new revenue streams to TIC companies is not currently reflected in stock prices. The table below shows financial details of the three major TIC incumbents: SGS, Bureau Veritas and Intertek. Note that in all three cases, P/Es remain at the low end of their historic range – there has been no re-rating to date.
To be sure, these companies are just beginning to discuss the upside to their businesses. Below are highlights from recent earnings calls:
SGS: The company has a “digital trust” unit where management sees “demand for digital trust services particularly strong in the U.S., with significant requirements around cyber resilience, AI algorithm validation for bias, and compliance testing. SGS is investing in building new service offerings in digital trust through both organic development and strategic acquisitions, with this area representing a key component of the megatrends driving future growth”.
Bureau Veritas: the company achieved 30% organic growth year-on-year in data center activities and continues to benefit from the structural need for data center expansion driven by AI infrastructure buildup. “Bureau Veritas is building AI assurance capabilities to help customers verify and validate that AI models are aligned with their values and policies, compliant with legal frameworks, and responsive to stakeholder expectations, especially as the regulatory landscape around AI assurance evolves”.
Intertek: The company has launched an AI lab to support global teams to develop the right AI solutions for the business. The company believes “there is a significant opportunity to help clients manage risk associated with investment in AI, and has launched AI², an independent end-to-end AI assurance program to help customers operate smarter and safer with the right trust in AI algorithms”.
Note the recent stock performance of TIC companies in the chart below. Over the past two years, the companies that now seem destined to become the auditors of AI have massively lagged the AI producers and first users (tech companies). We expect a catch-up phase is ahead as investors recognize the role TIC companies will play in the AI supply chain. Note that Intertek looks like a particularly interesting buying opportunity; the company announced earnings earlier this week and investors did not like that the strengthening sterling against major currencies significantly impacted results. On a day when the FTSE was down sharply due to geopolitical turmoil (Iran), Intertek stock took an exaggerated dive which we believe makes an alluring buying opportunity.
For decades, logistics was the invisible infrastructure of globalization – cheap, frictionless, and engineered for cost minimization rather than resiliency. COVID exposed those assumptions, and the system never fully recovered. As manufacturing re-shores to North America and Europe, trade policy becomes more protectionist, and corporates confront the limits of net-zero commitments, logistics has shifted from a background function to a material sustainability and financial risk factor.
When production sat in low-cost geographies, moving goods thousands of miles was still cheaper than manufacturing locally – making inefficiencies tolerable. Re-shoring changes the equation. The two charts below make the re-shoring trend unmistakable: manufacturing construction spending and re-shoring job announcements have both accelerated dramatically since COVID. If one assumes that the proposed Trump-era tariff regime materializes (a big IF), the next few years could see further re-shoring as firms hedge against heightened trade friction.
More volume must now be processed closer to end markets, last-mile and returns networks must tighten, and transportation costs re-enter the foreground. This is not “sustainability as virtue signaling”: for many companies, improving logistics efficiency is simply good business.
From a regulatory perspective, Europe’s Carbon Border Adjustment Mechanism (CBAM) is reshaping cross-border cost structures by attaching carbon pricing to imports. At the same time, mandatory disclosure frameworks – most notably the Corporate Sustainability Reporting Directive (CSRD) – require companies to measure and report Scope 3 emissions across their full value chain, placing scrutiny on the largest share of their footprint. With most corporates having already captured the low-hanging Scope 1 and 2 reductions, the decarbonization frontier has shifted upstream and downstream to suppliers.
The chart below shows how sharply the emissions profile has tilted. Operational emissions (Scopes 1 and 2) have fallen by roughly 15% from their 2019 peak, but this progress has driven Scope 3 to more than 90% of total emissions among the world’s 500 largest companies – up from about 75% in 2014.
The final catalyst is that logistics networks built for 2010 demand patterns are misaligned with 2025 realities. E-commerce growth, SKU proliferation, labor shortages, congested ports, and high empties (nearly 33% of U.S. truck miles still run unloaded) are pushing the system to its limits. At the same time, cutting-edge technologies like automation, robotics, AI-driven routing, and end-to-end visibility platforms are finally mature enough to produce step-change efficiency gains. The timing is consequential: structural stress is colliding with tools capable of materially reducing it.
Three Pillars of the Sustainable Logistics Investment Theme
The structural pressures reshaping global supply chains create an investment landscape that echoes past infrastructure transitions: a system operating at its physical limits, a new cohort of technologies capable of relieving those constraints, and a geopolitical alignment where countries are focusing more on themselves. But the logistics transformation is fundamentally distributed. It unfolds simultaneously in buildings, transport networks, and the digital orchestration layer that stitches everything together.
Sustainable logistics is therefore best understood as a multi-sector theme with heterogeneous business models and risk profiles. On one end sit the Logistics Real Estate Companies, providing the physical backbone where goods are stored, sorted, and processed. In the middle are the Transportation & Logistics Operators, the firms that actually move products across increasingly complex global and regional networks. Layered above both is the fast-growing class of Software & System Optimizers, the digital infrastructure orchestrating flows, improving asset utilization, and maximizing emissions efficiency across the entire “supply chain of supply chains.”
This week’s note will highlight the leading companies in each subsector for investors seeking exposure to the ongoing transformation of sustainable logistics.
I. Logistics Real Estate: Warehousing as Critical Infrastructure
Warehousing has become one of the most structurally constrained components of the global economy. Despite intermittent recession fears, U.S. industrial vacancy remains near 4%, well below historical averages. European markets are even tighter, with modern logistics space in the Netherlands and Germany at sub-3% vacancy. When space is available, much of the global stock is outdated – not designed for automation, electrification, or high-velocity returns. The result: lower storage density, inefficient material flows, and higher energy use per pallet.
As throughput requirements surge and tenants electrify fleets, grid capacity, rooftop solar, EV-charging yards, and digital backbones have become mission-critical. This is pushing logistics real estate owners into the role of modernization utilities – enabling tenants to meet operational, emissions, and resilience requirements that legacy warehouses simply cannot support.
Within the subsector, Prologis stands out as the most advanced on sustainability investment, supported by green-financing programs, integrated climate governance, and large-scale infrastructure upgrades across its portfolio. Segro plc is a close second followed by Goodman Group with credible long-term decarbonization strategies and capex commitments.
II. The Transportation & Logistics Operators: Fixing the Network Itself
If logistics real estate companies control the physical nodes, transportation and logistics operators manage the edges – the links that move goods between every point in the system. Those links remain structurally inefficient. U.S. trucking networks still operate with about a third of empty miles, and global road-freight emissions reached a record 2.2 Gt CO₂ in 2023. Inefficiency is both a cost burden and an emissions problem.
Operators such as C.H. Robinson, DSV, SG Holdings, and Brambles are increasingly closing these gaps with software-driven, data-coordinated transport networks. Freight brokerage has become algorithmic; consolidation and routing decisions are automated; and real-time emissions data is shaping modal-shift decisions. Asset-sharing systems – most notably Brambles’ CHEP pallet-pooling model – reduce waste, improve reuse rates, and drive circularity at global scale.
Within this subsector, Brambles stands out as the sustainability benchmark due to the inherent circularity of its CHEP model and its progress against science-based targets. DSV ranks as the strongest of the global Third-Party Logistics companies (3PLs), supported by robust absolute emissions-reduction commitments and leading disclosure. SG Holdings has quietly built a high-quality sustainability profile. C.H. Robinson is making credible improvements and launching customer-facing tools, but its transitions appear more incremental relative to the leaders.
III. Software & System Optimizers: The Intelligence Layer
The third pillar – and the highest-growth, highest-margin segment – is the software and data infrastructure that orchestrates modern logistics. This includes warehouse management systems (WMS), transportation management systems (TMS), real-time visibility platforms, RFID ecosystems, and automation-control engines. As regulatory scrutiny tightens and operational complexity scales, the digital layer offers the fastest, most capital-efficient path to emissions reductions and productivity gains.
Several technology vectors are now converging. RFID adoption has finally passed its inflection point: global shipments exceeded 30 billion tags in 2023, up 25% year-over-year. Persistent labour shortages are accelerating warehouse automation capex, which is expanding 10–12% annually. And visibility tools that were niche pre-COVID are now mainstream, with 70% of Fortune 500 shippers deploying real-time tracking, exception management, and digital documentation.
Companies such as Zebra Technologies, Descartes, and WiseTech Global increasingly run the critical logic of logistics. Their platforms determine pallet routing, slotting optimization, picking sequences, carrier selection, and customs compliance. The impact is measurable: shrinkage reductions of up to 30%, stockout improvements of 10–20%, and materially higher asset-utilization rates across fleets, containers, and warehouse equipment.
Within this subsector, Zebra Technologies and WiseTech Global emerge as the clearest sustainability leaders, integrating emissions-aware tools and efficiency metrics directly into their solutions. Descartes remains an essential enabler of efficient logistics flows with solid ESG reporting, though it communicates its sustainability positioning more quietly than peers.
Sustainability x Financial Analysis
Building on the investment theses for each subsector, we evaluate the selected companies through a dual lens of sustainability and financial performance. The names highlighted in yellow represent our top picks, offering the most compelling intersection of ESG impact and valuation.
- Logistics Real Estate: While Prologis is the sector benchmark, its current valuation is prohibitive. We prefer SEGRO plc, which offers a comparable sustainability profile at roughly half the multiple, albeit with a heavier UK/EU geographic exposure.
- Logistics Services: Both Brambles and DSV A/S stand out for offering a robust balance of sustainability leadership and strong financial fundamentals.
- Technology: Zebra Technologies is our sole selection in this category, driven by its attractive valuation and accelerating focus on sustainability initiatives.
Ultimately, this theme offers a distinct advantage for portfolio construction: it provides natural diversification across asset classes and business models. Logistics Real Estate Companies provide defensive ballast through real-asset duration, inflation-linked income, and structurally tight markets. Transportation & Logistics Operators add cyclical torque, outperforming when global trade accelerates and capacity tightens. Software & System Optimizers supply the long-duration growth engine, driven by automation, digitization, and the rising value of real-time data. Combined, these segments create a tunable mix of quality, growth, and cyclicality depending on an investor’s risk posture.
Taken together, these three pillars form an ecosystem undergoing simultaneous modernization. Buildings are being retrofitted, networks are being re-optimized, and software is finally coordinating the whole system with real-time data. It is rare to find a theme where physical infrastructure, operational networks, and digital layers are all experiencing multi-year investment cycles in parallel – and rarer still when the economics and decarbonization objectives align this cleanly.
The Inevitable Clash
China is the world’s manufacturer, and the United States is its consumer. In the 2000’s, the commercial trade deficit of the United States was larger as a percentage of GDP (6%), it was balanced across many countries and built on the hope that the rise of China would bring new consumers. Obama realized late in his second term that China would not become a consumer, and he timidly pivoted to Asia. In his first term, Trump tried to make a deal with China using sticks (tariffs). Biden maintained those tariffs but used carrots (industrial subsidies for United States production). Trump will now use bigger sticks to reverse China’s manufacturing domination. The chart below highlights the biggest global economic issue of our time between the two largest economies in the world; China produces 5% more than it consumes, and the U.S. consumes 3% more than they produce.
At the top of the following graph, we can see the countries on the target list of the Trump administration’s goal to rebalance production towards the United States. Unsurprisingly, Mexico, Canada and China have been targeted for tariffs for national security reasons even before the required analysis to implement tariffs for economic reasons is completed (the latter should be completed by April 1st). We can expect global tariffs following the economic analysis with special negotiation tariffs for specific countries when opportunities arise.
The chart below describes the items for which the United States has the largest deficits and the largest surplus. The commercial balance is negative for manufactured items and positive for energy and food commodities, as well as high-tech circuits and semi-conductor machinery. The United States will fight hard to keep exporting high-tech circuits and semi-conductors.
On the opposite side, the picture below highlights sectors where China dominates global production. Those sectors have not been kind for global companies because profit margins have essentially disappeared. Chinese companies are choosing volume over profits, supported by Chinese banks that continue to lend to non-profitable companies at the directive of the government.
The table below shows the sectors where China will concentrate its efforts in the future. These are high-value-add sectors and the commercial clash between China and the United States will shape the global economy for the next decade.
The First Chinese Curveball to the New Trump Administration
Just as Donald Trump announced the Stargate project dedicated to huge investments in AI/data centres in the United States, performance results for DeepSeek models were released: they were truly impressive, especially considering the high-performance chips embargo that China is under and the relatively small size of the DeepSeek team. Moreover, their models are available as open source and use specialization techniques that reduce chip performance demand.
The implication is that the huge moat that supposedly encircled large current Western language models no longer looks so large. If a small firm can build highly performing models at a fraction of the cost, one can expect a race to low profitability similar to other sectors where China is competing aggressively.
American Exceptionalism
The U.S.’ brightest spot relative to China was Silicon Valley’s tech dominance. However, since ChatGPT came out in 2023, the big question has been how large language models (LLM’s) will be monetized. Just like the prospects from the Internet in the late 1990’s or those of the Chinese consumer in the early 2000’s, investors are really excited by the opportunity of LLMs and assume that Apple, Amazon, Microsoft and Google will be able to extend their existing large moats to artificial intelligence. It’s an open question whether LLMs will follow the profitability path of the Internet that eventually delivered or to the Chinese consumers that have proven disappointing.
Investor enthusiasm has led Open AI, Anthropic, Google, Meta and Microsoft to spend enormous amounts to be AI leaders. So far, their investments seem based on the motto “If you build it, they will come.” Open AI and Anthropic are dedicated to LLMs whereas Google, Meta and Microsoft aim to control the artificial intelligence ecosystem. So far, the large language model builders have adopted a brute force approach that resulted in huge demand for the best hardware. This created scarcity for Nvidia’s H100 chips (made for about $3,000 with a lifespan of 2-3 years and sold for 10 to 50 times that price). As a result, Nvidia became the most valuable company in the world.
Winners and Losers
The table below highlights our view of future market behavior in artificial intelligence. Software developers should benefit from more cost-friendly development leading to greater adoption (Jevons Paradox). Profitability winners from lower cost could be users that have not participated in the LLM arms’ race, most small businesses and everyday users. The big losers are risking being pure LLM developers like Open AI, which happens to be in fund-raising mode, and Anthropic; they could quickly become obsolete due to their high operating cost if a race to the bottom starts. Google, Microsoft and Meta have made large investments in LLM development, they are also likely losers because they have overspent but they could still benefit from greater adoption.
In terms of the infrastructure build-out, all else equal, Nvidia is a likely loser because the scarcity value of its chips will diminish, and their margins risk falling from their current stratospheric levels. Meanwhile, we expect data center enablers like energy producers and energy efficiency providers, to fare somewhat better; they will benefit from greater use of artificial intelligence but be negatively impacted by lower energy needs. Data centers will likely charge less for computing but pay less for chips. The economics behind the build-out of independent data centers (not the ones built by Google, Microsoft and Meta) generally assume a 20-year payoff. These will some under scrutiny from uncertainty in compute pricing even if the lower cost of chips could compensate for lower revenues.
The Likely Reaction
No one can “unsee” DeepSeek’s cost-to-performance results. Nonetheless, we expect a witch hunt to target the origin of their chips and intellectual property theft, leading to an eventual ban of DeepSeek in the United States.
Significance for Sustainability Investors
One of our investment themes is Productivity with Purpose, whereby technological innovations enable new solutions to improve quality of life. DeepSeek’s lower cost and energy demands could make LLMs more readily available and produce economy-wide gains as described in our note, “History Lessons for an AI Future”, June 20, 2024.
A core SMS megatrend is the age of electricity. From a pure energy efficiency standpoint, DeepSeek is an improvement over previous LLMs and could reduce the strain on projected energy demand emanating from the AI sector. However, given that AI is only a one piece of the electrification trend, we believe that the punishing market reaction to energy efficiency and nuclear power companies since Deepseek entered the scene is overdone.
DeepSeek Market Implications
The mega-cap names based in the United States have been clear beneficiaries of globalization. Their domination has grown since 2023 from the expected moats of artificial intelligence and investors assumed that the largest companies would be the ultimate beneficiaries, but a small company like DeepSeek should create doubts about their continued domination. Just like with the Internet around Year 2000, the ultimate winners have not been the ones people expected. From this standpoint, we expect subpar performance from the large companies that have optimistic implied expectations and outperformance from smaller companies that have reasonable implied expectations that will benefit from low-cost AI solutions.
Immediate Investment Opportunities in Electrification
The table below shows five companies that have fallen between 10-20% since the DeepSeek moment. Considering the long-term electrification trend should continue (even if data centers construction slows down), the activities in which they are involved will see sustained demand.
Emergence of AI Applications
As discussed above, the emergence of DeepSeek marks a shift in AI’s evolution; suddenly the technology is significantly cheaper and more efficient - if the data holds true. Moreover, as an open-source AI model, broad accessibility will facilitate further development. Based on these factors, we believe investors should shift their focus on software rather than hardware. Lower costs and improved accessibility create tailwinds for companies leveraging AI in their products.
From a sustainability perspective, we focus on AI applications that drive meaningful productivity gains (SMS theme Productivity with Purpose). Two areas with the most impactful opportunities are industrial and healthcare applications.
In industrial settings, AI enhances resource efficiency, reducing waste through optimized processes or improved design in the case of engineering tools. We have explored this topic in past research: “Finding Real Returns in Artificial Intelligence”, April 27, 2023. These companies stand to benefit not only from effective use of AI, but also from the ongoing deglobalization trend leading to new factory construction. Autodesk, PTC Inc. and Parker Hannifin are well positioned in this space and experienced strong stock performance in the past year. Dassault Systemes and Rockwell have lagged their respective peers.
In healthcare, AI is being used to improve diagnostics and accelerate drug discovery. In contrast to some of the industrial plays, healthcare names have underperformed the market in recent months offering a more attractive entry point. Both GE Healthcare and Quest Diagnostics are leveraging AI in their diagnostic tools and processes; Quest Diagnostics is also notable for its acquisition of PathAI. On the drug discovery side, Eli Lilly had partnered with OpenAI to find treatments for drug-resistant bacteria and GSK is touting the benefits of its AI use. Investors should also watch life science companies like Danaher, Agilent and Thermo Fisher, which provide tools that support drug development and research. These too can benefit from cheaper, more readily available AI.
Beyond industrial and healthcare applications, greater AI accessibility will be disruptive across other sectors, though the societal benefit varies. AI enhancements to tools like Workday, Shopify and Duolingo are beneficial, but less transformational. A step further, companies like Verisk (insurance), Wisetech Global (logistics) and Wolters Kluwer (finance, ESG) can leverage AI for specialized applications in more critical industries. However, the most significant beneficiaries of expanding AI adoption are enablers like Accenture, CGI Inc and Capgemini which help businesses across a broad range of sectors integrate AI to enhance operational efficiency and better decision-making. As this technology becomes more accessible and cost-effective, demand for their services will surely rise.
Since its launch in November 2022, ChatGPT has become hugely popular and it’s no wonder: with an IQ of above 120 (placing it among the top 10% of humans), the generative AI tool can do things that many humans can’t.
But ChatGPT and the like are just a small sliver of what artificial intelligence can and will be able to do.
General purpose technology (GPT), which artificial intelligence is presumed to be, is defined as ‘a single generic technology, which initially has much scope for improvement and eventually comes to be widely used and to have many spillover effects. Examples of GPTs include electricity and the personal computer. In other words, a technology that is pervasive and significantly impacts aggregate productivity growth.
However, the pervasiveness and productivity growth associated with GPTs often take very long to develop. A popular example is electricity: the big impact of electricity came in the 1920s, about 40 years after Thomas Edison first distributed electrical power in New York in 1882.
The long lag time between invention and widespread application occurs because GPTs tend to entail major investments, learning costs, and complementary innovations. Often, the net benefits of the new technology itself are low early for the earliest adopters.
Prior to artificial intelligence, the most recent example of a GPT is the personal computer. The first desktop computers available for the mass market were introduced in the 1970s, but it took two decades before a significant share of households had adopted them. In 1993, still only 23% of American households owned a computer. The share of workers using computers in their jobs also grew relatively slowly before becoming ubiquitous in the 2000s.
This widespread adoption of computers was followed by an increase in U.S. productivity growth. Between 1991 and 2004, total factor productivity (TFP) grew 1% per year on average, almost twice as fast as its average growth rate in the 1980s (about 0.6%).
Like other GPTs, it is highly likely that artificial intelligence will spur a strong, sustained increase in productivity growth across whole economies that could last decades. Like past GPT related productivity booms, however, these advances could surprise by how long they take to be truly transformative.
An early indicator of enhanced penetration of AI into the overall economy should be the observation of a wave of increased capex, IT and training expense. This has clearly started, but remains timid in most areas.
Capex is currently significantly greater for Nasdaq listed (i.e. tech-focused) companies compared to indexes that reflect the broader economy, as shown in the chart below. This suggests that so far, much of the acceleration of IT spending has been fairly concentrated.
A sign of AI pervasiveness could be when the capex efforts of financials, health care or other sectors match levels in the IT space.
In the meantime, we examine what the overall effects (and roadblocks) of artificial intelligence might be beyond sheer productivity gains.
Artificial Intelligence Will Create More Jobs Than It Destroys
AI, like all new technologies, has been accompanied by a raft of concerns about how its usage will affect certain aspects of society. The first is that job automation will lead to significantly higher unemployment rates. In our view, this fear is widely exaggerated.
The periods of greatest unemployment tend to occur during economic recessions, typically brought on by financial crises and spikes in oil prices.
In fact, studies have shown that technology tends to create more jobs than it destroys. Deloitte has recently published a convincing set of charts across a number of industries where jobs were created/and destroyed over the past 140 years of innovation. No doubt the debate suffers from a common cognitive bias, loss aversion. Analysts tend to focus on the job-destroying effects of technological change (because they are more easily observed) rather than its creative aspects.
Moreover, there is already evidence that the impact of artificial intelligence on the job market is playing out exactly as described above: census bureau data of U.S. firms show that adoption of AI (and other technologies like robotics) so far varies substantially across industries, but is concentrated in larger firms. Adopters have higher labor productivity and greater demand for skilled labor, but the effects on overall employment levels at these firms were limited or ambiguous. This is also borne out by survey data, as highlighted in the chart below.
Importantly, while general purpose technologies prior to the personal computer most severely impacted “blue collar jobs”, it is most likely that service sector employment, such as IT and finance, will see their operations being the most impacted by AI in the coming years.
The bottom line is that there is no precedent that suggests the introduction of artificial intelligence will be different than past GPTs: job losses will occur - but will be offset by job growth in new sectors, This topline approach, however, does mask important weaknesses in the labor market, most notably income inequality.
Artificial intelligence Will Exacerbate Income Inequality
A more serious potential concern regarding AI adoption is a rise in socio-economic inequality.
There is a great deal of historical evidence to suggest that general purpose technologies have increased economic inequalities and there is potential – though not a foregone conclusion – that artificial intelligence will accentuate this trend.
Previous technological advances have been skill-biased, meaning that high-skill workers experienced wage gains due to productivity improvements while people in lower-skilled jobs, (i.e. more repetitive tasks), experienced both job losses and decreased real wages due to automation. It’s estimated that between 50% and 70% of changes in the U.S. wage structure (among the most flexible job markets in the world) over the past four decades are due to the relative wage declines of workers specializing in routine tasks in industries that experienced rapid automation.
This phenomenon, unsurprisingly, bears heavily on the financial outcomes of those with low levels of educational attainment. In the U.S., real earnings of individuals without a high-school degree are now 15% lower than they were in 1980, whereas real wages of workers with a post-graduate degree rose significantly.
The next phase of technological innovation is no different. In the words of technologist and investor Kai-Fu Lee, “corporate profits will explode, showering wealth on the elite executives and engineers lucky enough to get in on the action….”.
The remedy to income inequality stemming from general purpose technologies is already known and well documented. It requires a redistribution of wealth via taxation and assurance of a social safety net that includes access to quality health care and education. Differences in Gini coefficients among the most technologically advanced countries underscores that the tendency toward technologically-induced income inequality can be counteracted through policy. In this sense, although it is a foregone conclusion that increasing automation means greater dispersion in the price of labor according to skill, it is not inevitable that artificial intelligence will lead to greater social inequalities.
Contrary to income inequality within nations, income inequality between countries has been falling in recent decades as faster-growing emerging economies like China and India have narrowed the income gap with advanced economies. This trend is in danger of reversing, given that AI is developing in a monopolistic fashion that risks concentrating economic power within a handful of actors.
Left Unchecked, Artificial Intelligence Will Concentrate Economic Power Even More
A common tendency among general purpose technologies is that they tend to create monopolies. Anti-trust laws exist and are enforced to prevent customers from punitively high prices and to ensure that economic power does not translate into egregious political power (witness the U.S. oil industry). In a review of general purpose technologies, the U.S. Department of Justice concludes that in all cases, “dominant firms have accumulated far more monopoly power than is necessary to motivate and sustain the most rapid and beneficial rate of technological progress”.
Thus, fears about monopolistic power concentrated in the hands of artificial intelligence “first movers” is warranted, based on the historical precedent of previous GPTs.
Aspects of the AI industry are already acting as a monopoly. The table below shows the major links in the artificial intelligence supply chain (more commonly known as the “AI stack”) with the degree of market concentration at each step.
Currently, it is the early inputs that are most monopolistic. For example, TSMC, the semiconductor manufacturer, is known to prioritize its contracts with Apple over other chip consumers during periods of shortages. Note that the companies in the “AI supply chain” with the most monopolistic power (NVDA, TSMC, ASML, AMZN, MSFT, GOOG) are also the stock prices that have been among the biggest gainers in dollar terms since the launch of ChatGPT.
The stakes of monopolistic behaviour are arguably far higher with artificial intelligence relative to past GPTs. Concentration in the number of foundation models, i.e. the number of artificial intelligence models on which data is trained and uses self-supervision, means that model providers can have an outsized influence on information and can shape the public discourse in profound ways. The current concerns about the outsized influence of media platforms owned by Meta would pale in comparison to the control possible by AI models in the hands of a select few.
The bottom line is that the artificial intelligence industry currently displays monopolistic tendencies. Similar to income inequality, there are policy prescriptions to counteract this, but left unchecked, historical precedents suggest that artificial intelligence will facilitate a higher concentration of economic power.
Artificial Intelligence’s Biggest Roadblock: Less Energy Inefficient Than the Status Quo
By status quo, we mean the human brain.
Artificial intelligence is different from previous general purpose technologies in that it is - in its current form – highly energy inefficient compared to a human. Note that in all other general purpose technologies, the energy needs of the technology to complete a task were vastly inferior to the preceding method of task completion. Consider the steam engine: the very first steam engine trains are estimated to have the power of approximately 300 horses and while steam engine powered locomotives are not energy efficient by today’s standards, they are vastly superior to the horse.
Artificial intelligence is already more powerful (i.e. smarter) than the human brain to solve problems with defined scope. However, to complete tasks, artificial intelligence today is less energy efficient than the human brain. The chart below shows the energy requirements of a mouse and human brain compared to computers. To pass from artificial intelligence to general intelligence, i.e. to move from solving defined tasks to solving problems in various domains (i.e. like a human being can) would take almost unimaginable quantities of energy.
Indeed, according to the Frontiers in Artificial Intelligence, researchers estimate that biological computing is about 9 × 108 times more energy efficient than artificial computing architecture. The authors uncover that to compete with the human brain across an economy, artificial intelligence’s energy use would need to be orders of magnitude higher than that of a large, highly industrialized nation. In other words, the authors conclude that “the principles of thermodynamics of computation pose a significant, if not insurmountable barrier, to the emergence of any artificial general intelligence system”. Of course, AI energy efficiency will no doubt see gains in the coming years, but it seems that humans have an important head start.
The chemical industry is low on the list of what comes to mind for sustainable investors. It is generally considered to be among the largest sources of industrial greenhouse gas emissions, with steel and cement being the other major contributors. The chemical industry is complex, covering a wide variety of processes and applications that are present in nearly 95% of manufactured products. As an industry with a tight regulatory framework and large emissions, it is devoting significant efforts to reduce its carbon footprint.
The industry is divided into four major categories: commodity chemicals, fertilizers, industrial gases, and specialty chemicals. In this note, we start at the base and focus on commodity chemicals which encompass petrochemicals and inorganic commodities.
Chemicals Outlook
The overall outlook for the chemical industry is one that is intimately tied to the state of global manufacturing. The chart below shows that global manufacturing is starting to recover after more than a year of slow manufacturing activity.
To identify investment ideas, we focused more on the operational performance of companies than their products. The list of end-products produced from chemicals is vast, making it difficult to narrow down a company to a single product. The purpose of this note is to guide sustainability-minded investors through this sector to make more informed investment decisions mainly based on operational performance.
Commodity Chemicals
Commodity chemicals remain at the bottom of the chemical product value chain. Compared to more complex sectors of the chemical industry like industrial gases and specialty gases, technical barriers to entry are low and differentiating factors are smaller.
There are two types of commodity chemicals: petrochemicals and inorganic chemicals. Polyvinyl chloride (PVC) is a combination of the two. In our search for companies, we included PVC in the petrochemical segment.
Petrochemicals are obtained through the process of refining oil, they are fossil fuel-based products. Their most common use is as plastic in packaging and containers. Though plastic packaging can draw the ire from environmentalists, a lack of good alternatives means they do have a role to play in preserving food, among other uses (see notes A Reality Check about Plastic Fantasies, April 21, 2023 and Good Things Come in Minimal Packaging, May 16, 2024).
Inorganic commodities are co-products from the chlor-alkali process. This process is very energy-intensive, accounting for about 40 to 50% of manufacturing costs. There are two main products: chlorine and sodium hydroxide. Chlorine is a disinfectant, a solvent and is used in solar panels and PVC. Sodium hydroxide, also known as caustic soda, is a cleaning agent commonly found in soaps and detergents, used as a water treatment product and in aluminum production to break down bauxite. These products have very useful applications where alternatives are limited.
Commodity Chemicals Outlook
The short-term outlook for ethylene depends on the resolution path of the current oversupply; improved demand and/or shutdown of capacity. Improved demand prospects for the Chinese economy and recent capacity reduction have resulted in a recent increase in prices.
The short-term outlook for PVC is slightly better than ethylene because oversupply is not as pronounced. Just like ethylene, the recent price increases reflect improved prospects from China.
The caustic soda market is in better balance between supply and demand. This market is not as economically sensitive as ethylene which explains the less dramatic price increases observed recently.
Just like the United States, which used to be the biggest importer of energy and is now a net exporter, China was the world's largest importer of chemicals but has now become a net exporter. The significant growth in supply from China across the chemical industry is transforming the landscape.
The price and proximity of energy determines where commodity chemicals are produced. The shale revolution and the war in Ukraine have put significant pressure on European plants. The shale revolution improved access to petroleum in the United States with very cheap prices relative to other jurisdiction for natural gas. The war in Ukraine deteriorated access to cheap Russian energy for European chemicals company. As a result, many chemical producers are curtailing their operations in Europe. Higher cost of production and lower prospects for increasing demand from Europe are the driving force behind those shutdowns.
Petrochemical Companies
To find the most attractive companies within ethylene and PVC from a long and short perspective, we constrained our analysis to a minimum market capitalization of 1B$ and at least 30% of revenues from plastic resins and materials manufacturing industry. We use the Sustainability Accounting Standards Board (SASB) materiality framework to identify key issues such as GHG emissions, hazardous waste management, and safety & environmental stewardship of chemicals. To rank the companies within the industry we referred to their CSRhub scores for energy and climate change, and resources management and evaluated the number and importance of recent safety controversies to assess safety aspects.
The following table provides a ranking of petrochemical producers based on their percentile score in energy and climate change, resource management, and safety controversies.
Despite its top ranking, we decided to exclude Daicel Corp from long opportunities as it is also the leading Japanese producer of materials used in cigarette filters, as we try to avoid exposure to the tobacco industry.
We also excluded Alfa SAB de CV from short opportunities despite its bottom ranking because of conflicting signals: Alfa operates in the petrochemical and synthetic fibers industry (polyester, plastics, and chemicals), its petrochemical segment – Alpek - was spun off and is now a public company. Alpek has high CSRHub rankings and still represents 40% of the market capitalization of Alfa.
The two long opportunities derived from our analysis are Shin-Etsu Chemical and Covestro given their strong operational scores in high materiality elements of the business. Westlake and EMS-Chemie are short opportunities, especially given their poor operational scores, but also because of the concerns raised about Westlake's safety practices following the plant explosion at their Louisiana facility.
Inorganic Companies
To select companies in the inorganic chemical industry, we used the same methodology as for petrochemicals.
Based on our analysis, we identified 2 long opportunities, Tata Chemicals and Tokai Carbon, and 2 short opportunities, Genesis Energy and Olin Corp.
Finally, we looked at financial ratios to determine the most attractive long and short for each of petrochemicals and inorganics. From the graph below, we see that Shin-Etsu has good margins, strong growth, decent valuation and no debt whereas EMS-Chemie has similar margins with no debt but much lower growth and much higher valuation. For inorganics, Tokai Carbon has high margins, strong growth, decent valuation and no debt whereas Genesis has low margins, low growth, high valuation and a high debt load.