The defining investment question of the AI era may not be which company builds the most powerful model, but which institutions end up controlling the infrastructure, standards and relationships through which AI actually gets deployed. That distinction is becoming more consequential as two competing risks pull the industry in opposite directions: excessive concentration, where a handful of foundation-model and infrastructure providers capture a disproportionate share of economic value while enterprises hand over data, institutional knowledge and bargaining power; and fragmentation, where governments respond to that dependency by trying to build sovereign AI stacks that lack the capital, compute, talent and scale to stay globally competitive.

For investors, this tension points to something bigger than technology policy — a structural shift in where durable value is likely to accrue across the AI economy. The most plausible endpoint is neither a fully centralized system dominated by a few platforms nor a patchwork of self-sufficient national ecosystems, but a network of trusted technology markets, where countries, companies and institutions stay operationally independent while becoming increasingly integrated through shared standards, supply chains and governance frameworks. If that thesis holds, the next phase of AI investment will be shaped as much by the economics of trusted interdependence as by raw technological capability.

Microsoft CEO Satya Nadella has warned against an AI economy in which companies across every industry cede value to a small number of models. Jacob Helberg, the architect of the Pax Silica initiative, has flagged the mirror-image danger: governments pursuing technological sovereignty through isolated national AI stacks that end up weakening their own competitiveness. The two men are arguing from different vantage points, but they are describing the same structural problem. AI systems carry extraordinary economies of scale — training frontier models demands enormous capital, compute, energy, data and technical expertise, all of which naturally favor concentration — even as the organizations and governments deploying that technology have every incentive to keep control of their proprietary data, critical infrastructure and strategic decisions. Scale wants concentration; sovereignty wants control. Neither side is winning outright, and that unresolved tension is where the opportunity sits.

For capital markets, transitions like this one tend to follow a pattern: value concentrates first at the infrastructure layer, then migrates toward applications, distribution and specialized ecosystems. The internet made fortunes in telecoms and computing infrastructure before the largest pools of value moved elsewhere. Cloud computing followed the same arc — infrastructure stayed highly profitable, but the bigger prize turned out to be the entirely new software markets it enabled. AI looks to be approaching a comparable inflection point. The first investment cycle has been a story of scarcity, with capital chasing compute capacity, advanced semiconductors, data centres, electricity and frontier-model capability because supply couldn't keep up with demand. The next cycle is likely to be defined by something less glamorous but ultimately more valuable: deployment.

Enterprises and governments now have to work out which models can touch sensitive data, where workloads should be processed, how systems get audited, which supply chains can be trusted, and how dependency on any single vendor can be managed without becoming a liability. That set of requirements creates a potentially significant economic layer sitting between frontier AI infrastructure and end users — because in this context, trust isn't an abstract political concept, it's an operational requirement. A bank rolling out AI across regulated financial systems, a government embedding models into defence infrastructure, or a manufacturer weaving AI into industrial processes cannot judge the technology on performance alone; provenance, security, jurisdiction, interoperability, governance and continuity of supply all become material to the purchasing decision as AI penetrates critical systems.

That has a direct investment implication: the economic value of AI is likely to broaden out from the companies that build intelligence to the companies that make intelligence deployable. Cybersecurity providers, identity platforms, data-governance companies, semiconductor traceability systems, compliance software, sovereign cloud infrastructure, observability platforms and AI-auditing technology could all become increasingly important layers of the stack. The market has spent a great deal of time pricing the producers of AI capability. It has spent much less time pricing the companies that determine whether that capability can be trusted at all — and that gap is where some of the least crowded trades may be found.

Geopolitics reinforces the thesis. The Pax Silica initiative and the Joint Statement on AI Opportunity both reflect an attempt to organize AI supply chains and technology markets around trusted international partnerships, and the fact that European countries are joining the United States in that effort suggests competition is increasingly being framed around economic blocs rather than individual national strategies. That framing makes sense because almost no country has every resource needed to build a competitive AI ecosystem alone: advanced semiconductor manufacturing is geographically concentrated, critical minerals are unevenly distributed, frontier-model development eats enormous pools of capital and talent, data centres depend on energy infrastructure, and applications need access to customers, proprietary datasets and industry expertise. Since no single democratic economy controls every layer, the logical response isn't technological autarky — it's deeper integration between markets that already trust one another.

For investors, that could produce a new kind of geopolitical premium. Companies embedded in strategically important allied supply chains may benefit from regulatory preference, government procurement and infrastructure spending, along with a lower perceived political-risk profile — while businesses tied to opaque supply chains, contested jurisdictions or strategically vulnerable infrastructure could face rising compliance costs and valuation discounts. The semiconductor industry already shows how quickly geopolitical alignment can move capital allocation; AI is likely to extend that dynamic across a much wider range of industries. "Democratic coupling" is the term that captures the economic logic here — countries and companies pooling complementary resources, from compute and minerals to capital, research, energy and talent, without each trying to replicate the entire technology stack at home. Done well, that architecture could add up to something bigger than a geopolitical alliance: an integrated technology market in its own right.

The investment opportunity lies in the connective tissue that makes such a market function. Standards, credentialing systems and verification infrastructure could lower transaction costs between governments and enterprises operating across trusted markets. Supply-chain provenance systems should grow more valuable as semiconductor and critical-mineral networks become more strategically sensitive. AI governance platforms can help institutions deploy models across multiple jurisdictions while keeping compliance and operational control intact. None of these businesses will draw the attention that frontier-model developers do, but their economics can be genuinely attractive: infrastructure built around verification and interoperability tends to generate recurring revenue, high switching costs and real network effects, and once standards get embedded into procurement systems and enterprise workflows, displacing them becomes difficult.

There is a risk worth taking seriously here, though: trust itself could become concentrated. If a small number of technology companies get to define the standards that govern AI deployment, public technological sovereignty risks simply being replaced by private technological sovereignty — which is exactly why open, transparent and independently assessable standards matter as much economically as they do politically. Investors should draw a sharp line between ecosystems that genuinely expand interoperability and those that borrow the language of trust to reinforce a closed platform. The former can enlarge the whole market; the latter may post strong short-term economics before running into regulatory intervention and customer resistance. That distinction — between opening a market and merely dressing up a moat — will increasingly separate durable competitive advantage from a temporary one.

Companies that let enterprises keep control of their data, models and infrastructure are likely better positioned than those that require customers to surrender strategic assets in exchange for convenience. Nadella's idea of a "distributed frontier ecosystem" captures this shift well: institutional intelligence stays inside the organization while AI capability is deployed across shared infrastructure. If enterprise AI develops along these lines, the eventual winners may not be the platforms that capture the most data — they may be the companies that let customers extract value from their own proprietary data without ever losing control of it. That is a materially different framework from the AI narrative currently dominating markets, which still rewards scale, model performance and infrastructure ownership above almost everything else. Those advantages are real and will remain important. But as AI moves out of the experimentation phase and into critical economic systems, other sources of competitive advantage are likely to matter more than they do today: trust, interoperability, provenance, regulatory credibility, and the ability to operate across strategically aligned markets.

Investors should therefore be evaluating AI exposure on three dimensions rather than one. The first is dependency: how reliant is a company on a small number of model providers, semiconductor suppliers, cloud platforms or jurisdictions? The second is control: does the business retain ownership of proprietary data, customer relationships and institutional knowledge as AI adoption deepens? The third, and the one markets have priced least efficiently, is position within the emerging trust infrastructure — whether a company merely consumes AI, or actually provides the verification, security, governance and interoperability that allow others to deploy it safely. That third category may hold some of the least appreciated opportunities in the entire AI trade.

The AI investment cycle began as a race to build intelligence. It is now moving into the harder, less visible challenge of integrating that intelligence into governments, companies and critical infrastructure — and that shift changes where investors should be looking for value. Compute will stay essential, frontier models will keep improving, and the large platforms will retain real advantages. But the long-term economics of AI are unlikely to be determined by technological capability alone. They will increasingly turn on who controls the systems that let this technology move safely across companies, industries and borders. The next major AI infrastructure opportunity may therefore be far less visible than the first — not another model, another chip, another data centre, but the architecture of trust that connects them all.