Who Controls the Chips Controls the AI

Who Controls the Chips Controls the AI

Most organizations thinking about AI strategy are focused on models, tools, and workflows. Very few are focused on the hardware underneath all of it. That is the wrong place to stop. DeepSeek, the Chinese AI startup that stunned Silicon Valley with its low-cost reasoning models, is now developing its own AI chip, quietly hiring semiconductor engineers through private channels and holding discussions with chip design firms, foundries, and memory suppliers (Reuters, July 2026). It is the latest move in a semiconductor race that is reshaping the entire AI infrastructure landscape, and the enterprise buyers building production AI workflows on top of that landscape are the ones least prepared for what comes next. The question is not whether the chip shortage, the export controls, and the domestic silicon race are relevant to the organizations deploying AI. They already are. The question is whether those organizations understand how.

The Shift the Market Is Underweighting

Most enterprise AI conversations treat hardware as someone else's problem. The cloud provider handles the compute. The AI vendor handles the model. The enterprise buyer handles the workflow. That division of responsibility made sense when the hardware supply chain was stable, predictable, and largely invisible. It does not make sense anymore.

Nvidia, which once commanded over 90% of the Chinese AI chip market, has seen that share decline to approximately 50% as of early 2026, as US export controls pushed Chinese enterprises toward domestic alternatives (Oplexa, March 2026). DRAM prices climbed over 50% in some categories in 2025, with server contract prices up as much as 50% quarterly. SK Hynix told analysts that all memory scheduled for 2026 production is already sold out, with shortages potentially persisting until late 2027. Major cloud providers including Google, Amazon, Microsoft, and Meta issued open-ended orders to Micron stating they will take as much inventory as the company can provide (AI News, January 2026).

Those numbers are not background noise for enterprise AI buyers. They are the supply chain conditions under which every AI infrastructure decision made in the next 18 months will play out. The organizations that understand that context will make better vendor decisions, better contract decisions, and better architecture decisions than the ones that do not.

What the Semiconductor Race Actually Looks Like

The chip story has two parallel tracks running at the same time, and both of them matter for enterprise buyers.

The US-China divide is producing two distinct AI infrastructure ecosystems. In April 2025, the US Commerce Department declared that the H20 chips that had powered DeepSeek's breakthrough AI model were noncompliant with export controls, cutting off one of the last remaining legal pathways for Chinese companies to access competitive Nvidia hardware (Brookings, June 2026). China's response has been systematic: Huawei filling the gap with domestic chips, Alibaba and Baidu building their own silicon, and now DeepSeek joining the race. The result is not just a geopolitical competition. It is the active formation of two separate AI infrastructure stacks, and the enterprise buyers whose vendors operate across both of them are carrying supply chain dependencies they may not have fully mapped.

The inference chip is where the next wave of competition is focused. DeepSeek's chip is designed specifically for inference, the stage where a trained AI model generates responses for users, rather than for training new models. OpenAI recently unveiled its first custom inference chip called Jalapeno developed with Broadcom, and Anthropic has also been exploring building its own AI processors (Reuters, July 2026). The shift toward inference-specific silicon is not just a technical detail. It reflects where AI computing demand is actually going. As AI applications spread from research and development into production workflows, more of the industry's computing work moves from training models to running them at scale. The organizations that understand which part of that pipeline their AI vendors are optimizing for are the ones that can make informed decisions about cost, performance, and supply chain risk.

Export control enforcement is becoming more aggressive, not less. US Congress approved the Chip Security Act on March 26, 2026, which would directly embed tracking technology into chips. Two federal cases in March 2026 resulted in arrests for diverting approximately 170 million dollars worth of Nvidia chip-equipped servers to China via Taiwan and Southeast Asian intermediaries using falsified documentation (BISI, April 2026). The compliance environment around AI hardware procurement is tightening at exactly the moment that demand is accelerating. Enterprise buyers in regulated industries who have not reviewed their AI vendor's hardware sourcing and supply chain practices are carrying compliance exposure they may not have priced.

Why This Creates Disproportionate Risk for Unprepared Enterprise Buyers

The hardware layer of the AI stack is not a background condition. It is a risk multiplier that compounds across performance, cost, and compliance in ways that most enterprise buyers are not currently measuring.

Performance dependencies are invisible until they produce an incident. The AI vendor that delivers consistent performance today is delivering it on a hardware supply chain that is actively being disrupted. Memory shortages, export control shifts, and domestic chip development timelines are not stable variables. An enterprise buyer that has not mapped its AI vendor's infrastructure dependencies against the current supply chain environment does not know how its AI performance will hold when the next disruption arrives.

Cost assumptions built on 2024 hardware economics are already outdated. The Trump administration's December 2025 decision to allow conditional sales of Nvidia's H200 chips to China came too late to prevent widespread supply disruption. Memory prices had already surged and cloud providers had locked up available inventory under open-ended supply agreements (AI News, January 2026). Enterprise buyers whose AI cost models were built before the memory price spike and the H20 export control shift are operating on assumptions that no longer reflect market reality. The organizations that have not updated those models are the ones that will face the largest budget surprises in the second half of 2026.

Compliance exposure is growing alongside enforcement activity. The Trump administration has been using its authority in novel ways that combine targeted policy flexibility with heightened compliance and diligence expectations on US and multinational companies. Expanded commercial access may not equate to reduced enforcement and compliance risk (Morrison Foerster, February 2026). An enterprise buyer in a regulated industry whose AI vendors source hardware through complex international supply chains is not just carrying performance and cost risk. It is carrying potential compliance exposure that becomes visible only when enforcement does.

How Enterprise Buyers Should Assess Their Actual Hardware Exposure

Five questions separate the organizations that understand the hardware layer underneath their AI deployments from those that are about to find out they did not.

1. Can the organization produce a current map of the hardware infrastructure underlying every AI system in production, including the chip architecture, geographic sourcing, and cloud provider dependencies associated with each vendor?

2. Has the organization reviewed its AI vendor contracts in the last six months to identify pricing terms that were set before the 2025 memory price surge and the H20 export control shift, and to assess which renewal windows fall within the current supply disruption period?

3. Does the organization have visibility into whether its AI vendors are building or sourcing inference chips through supply chains subject to US export control requirements, and has it assessed what a supply disruption would mean for each production workflow?

4. Has the organization assessed the compliance implications of its AI vendors' hardware sourcing practices, specifically whether any part of the supply chain involves hardware procured through markets or intermediaries subject to export control enforcement?

5. If the hardware infrastructure underlying a critical AI deployment became unavailable or significantly more expensive in the next 90 days, does the organization have a documented contingency plan and an alternative vendor path?

An organization that cannot answer most of these is not managing AI infrastructure risk. It is hoping the supply chain remains stable, which is a different thing entirely.

Bottom Line for Enterprise Buyers

The AI model race gets the headlines. The chip race determines who wins it. Export controls may have limited China's ability to deploy AI at scale. After the release of its R1 model, DeepSeek had to restrict access to its API, presumably because it could not provide enough inference compute to meet user demand (AI Frontiers, July 2025). That constraint is the clearest illustration available of what hardware dependency actually costs at the moment it becomes visible. The enterprise buyers that map their hardware exposure now, review their vendor supply chain dependencies, and build contingency into their AI infrastructure decisions are the ones that will scale confidently when the next disruption arrives. The ones that treat hardware as someone else's problem will find out it is not at the worst possible moment. Cost is what organizations pay to deploy AI. Value is what understanding the hardware layer underneath it protects across every performance commitment, every budget cycle, and every compliance conversation that follows. For enterprise buyers, the ratio is not close.

Works Cited

"China's DeepSeek Developing Its Own AI Chip, Sources Say." Reuters via Taipei Times, 8 Jul. 2026, www.taipeitimes.com/News/biz/archives/2026/07/08/2003860388.

"Ball Game's Over: The US Is Out of the AI Chip Market in China." Brookings Institution, June 2026, www.brookings.edu/articles/ball-games-over-the-us-is-out-of-the-ai-chip-market-in-china.

"US China Chip War 2026: Export Controls, Semiconductor Impact." Oplexa, 24 Mar. 2026, oplexa.com/us-china-chip-war-2026-semiconductor.

"AI Chip Smuggling: The Limits of US Export Controls." Bloomsbury Intelligence and Security Institute, 6 Apr. 2026, bisi.org.uk/reports/ai-chip-smuggling-the-limits-of-us-export-controls.

"2025's AI Chip Wars: What Enterprise Leaders Learned About Supply Chain Reality." AI News, 6 Jan. 2026, www.artificialintelligence-news.com/news/ai-chip-shortage-enterprise-ctos-2025.

"Managing Export Control Risks in the AI Chip Ecosystem." Morrison Foerster, 9 Feb. 2026, www.mofo.com/resources/insights/260209-managing-export-control-risks-in-the-ai-chip-ecosystem.

"How US Export Controls Have and Have Not Curbed Chinese AI." AI Frontiers, 8 Jul. 2025, ai-frontiers.org/articles/us-chip-export-controls-china-ai.