Microsoft (NASDAQ: MSFT) CEO Satya Nadella published a 1,200-word essay on X on June 14, titled “A frontier without an ecosystem is not stable,” which drew 28 million views.
The essay drew a direct line between AI concentration and globalization, arguing the AI industry risks the same structural damage that offshoring inflicted on industrial economies.
Nadella’s warning is not isolated, as similar concerns are growing louder across venture capital, crypto infrastructure, and AI research circles.
Microsoft, Google, Amazon, and Meta collectively plan to spend approaching $700 billion on capital expenditure in 2026, with most of that directed at AI infrastructure.
A single company supplies approximately 75% of the specialized hardware used to train AI at scale, while one foundry manufactures around 90% of the leading-edge chips the entire industry depends on.
“The chokepoint is physical. The risk was never five model companies. It’s a compute layer only five balance sheets can afford to build,” Fan told TheStreet.
Building competitive compute infrastructure requires land, power, cooling, and hardware procurement on timelines measured in years, making capital and physics the primary barriers rather than engineering skill.
Fan further cautioned that broad access to AI infrastructure means little without verifiability, adding, “Broad access means nothing if it’s permissioned by the people collecting the tolls. And the blocker is technical, not political.”
Kamiya told TheStreet, “We think concentration is a real concern, but the risk is dependency. If every startup depends on the same few companies for the core AI stack, the market becomes less open even if innovation is still happening.”
A startup can build a useful AI product on rented infrastructure, yet its pricing, data access, and customer relationships can still be controlled by the hyperscaler sitting beneath it.
Nadella’s essay argued that GDP numbers improved during offshoring, but the jobs and industrial ecosystems that left did not return, warning AI value concentration could produce the same pattern.
He also described a specific mechanism: AI models absorb company expertise and commoditize it, turning a firm’s proprietary capabilities into standardized services accessible to its competitors.
Wish Wu, co-founder and CEO of Pharos, said model and infrastructure advantages look durable today but may not hold as the market matures over time.
“Model capabilities will continue to improve, and infrastructure will remain critical, but those advantages may become less differentiated over time,” Wu said, pointing to application-layer integration as the more durable long-term position.
Wu noted that in past technology cycles, including social media, search, and mobile software, companies that owned user relationships captured more lasting value than those owning underlying infrastructure.
“I don’t think concentration itself is necessarily the problem. The risk is if access to intelligence becomes too restricted and limits innovation at the application layer,” Wu added.
AI agents capable of holding balances, executing payments, and coordinating autonomously will require financial and coordination infrastructure that does not yet exist and that no single entity currently controls.
Wu described that undeveloped layer as one of the most significant opportunities in AI, carrying large consequences depending on who builds it and how open they choose to keep it.
