Gabriel Odeyemi
← Strategic Digest

Strategic Digest

The Model Was Never the Moat

Chinese labs are giving away frontier-grade AI, and the durable value is migrating to chips, devices, and geography while energy costs climb.

Gabriel Odeyemi · · 6 min read

For three years the assumption underneath the American AI trade was that the best model would command the market. That premise is now under strain. Chinese laboratories are releasing open-weight systems that observers say rival the best from OpenAI and Anthropic at a fraction of the cost, and they are doing it for free. When capability becomes cheap and abundant, the question stops being who builds the smartest model and becomes who controls everything around it. The evidence gathered this week points in one direction: the defensible value in artificial intelligence is migrating away from the model itself and toward the chips, the devices, the distribution, and the geography that surround it.

The Erosion of the Software Moat

Moonshot AI's Kimi K3 and new models from Alibaba have unsettled parts of the AI commentariat, and not only because of their benchmarks. Kimi K3 was developed by a lab cofounded by a former Carnegie Mellon computer science graduate student, and its reception has been strong enough that Ryan Fedasiuk of the American Enterprise Institute described the American lead in frontier software as less durable than many had hoped. His expectation, as reported by Fast Company, is that Chinese labs will keep distilling and freely releasing capable systems.

The strategic problem for US labs is not that Kimi K3 must beat GPT. It only has to be good enough and free. That combination attacks the premium pricing that underpins the economics of OpenAI and Anthropic. A model that is nearly as capable and costs nothing sets a reference price that paid frontier systems must justify against. The Verge framed the pair of Chinese releases as a one-two punch landing precisely as the technology commoditizes. The immediate exposure sits with any buyer whose vendor commitments assume US frontier models are uniquely capable, an assumption now worth revisiting before contracts renew.

Where the Durable Value Is Moving

If the model layer is softening, two developments this week show where value is hardening. Nvidia's Jensen Huang left Tokyo with deals spanning Japan's entire technology ecosystem, according to TechCrunch. Read against the Chinese releases, the significance is clear. Huang is locking in sovereign and industrial demand for AI infrastructure outside the US-China axis, which deepens Nvidia's grip on the physical layer regardless of which lab wins any given benchmark. Chips are consumed no matter whose model runs on them.

The second signal comes from a courtroom rather than a chip. TechCrunch reports that an Apple lawsuit hangs over OpenAI's much-discussed ambitions to enter hardware and to go public. The detail worth holding is that the contested territory is the device layer, not the model layer. If the durable battleground is the hardware that reaches users and the ecosystems that distribute AI, then a legal cloud over OpenAI's hardware path and its potential public offering is more consequential than any single release. The company that arguably leads on models is being challenged on the ground where value is accumulating.

Together these two stories describe a market reorganizing around the pipes rather than the water. This is analysis rather than reported fact, but the pattern is consistent: commoditizing models, hardening infrastructure, and a distribution fight moving to devices.

The Macro Squeeze on the Buildout

The timing is unforgiving. A month ago American drivers had some relief at the pump after a deal between the United States and Iran intended to reopen the Strait of Hormuz. That detente has broken. The New York Times reports that the average US gas price has returned to four dollars a gallon as the Iran crisis escalates, and that global oil moved above ninety dollars before pulling back on tentative hopes of reduced tension. Shipping through the Gulf is dwindling.

Energy prices are an input cost for the compute-hungry AI buildout and an inflation and consumer-spending risk at the same time. The squeeze arrives from both sides. The cost of constructing AI infrastructure rises just as the revenue model behind it faces price pressure from free Chinese systems. Executives should stress-test near-term budgets against a sustained environment of ninety-dollar oil and four-dollar gas rather than waiting for the next spike to force the exercise.

A quieter reading of the same macro picture appears in luxury. Fast Company, citing Bain, reports that the luxury market has lost roughly seventy million customers since 2022, shrinking from about 400 million to around 330 million by the end of 2025, and that sales continue to decline. LVMH has agreed to sell Marc Jacobs as it trims its portfolio. Whether the cause is desirability or macro pressure, the aspirational consumer is an early gauge of discretionary strength, and the reading is soft precisely as energy costs climb.

The New Liability Layer in Hiring

A parallel risk is surfacing at the point where AI meets employment. MIT Technology Review reports new research suggesting that large language models do not merely inherit human biases from training data; they can develop their own. For firms deploying AI resume screeners, this shifts discrimination liability onto ground that existing compliance frameworks were not built to catch. The prudent response is immediate: audit any AI-driven hiring or screening tools now and align legal and human resources functions before a headline or a regulator forces the question. As models commoditize and spread into operational workflows, the governance exposure spreads with them.

The Strategic Read

The center of gravity in artificial intelligence is shifting from who builds the best model to who controls the surrounding system of cost, hardware, distribution, and geography. Chinese open releases are draining pricing power from the software layer, Huang's Japan deals and Apple's lawsuit reveal that infrastructure and devices are where durable value now sits, and the broken Iran detente raises the cost of building AI at the moment its revenue model comes under pressure. The judgment worth acting on is that the frontier is becoming plumbing. The advantage will accrue to whoever owns the pipes rather than whoever ships the cleverest model, and the near-term risks that demand attention are concrete: renegotiate vendor terms premised on US model supremacy, budget for a sustained energy shock, and close the governance gap in AI-driven hiring before it becomes a lawsuit.

Sources

← Back to all analysis