The US depends on private financing. Chipmakers and other tech leaders are leveraging their booming profits to invest in, guarantee, and otherwise finance customers. The Economist has even labeled NVIDIA the “central bank of AI.”
China uses a different model: state-led funds, state-run bank financing, local government investment vehicles, subsidized compute, and state coordination.
Although the US private financing approach works well at the start of a capital-intensive build-out, it may come under stress during a downturn. If AI demand disappoints, chip prices fall faster than expected, or AI applications take longer to produce paying customers, the pain could spread fast and wide, hitting equipment sales, data center revenues, customer creditworthiness, and the value of the supplier’s equity stake.
China suffers its own debt and overcapacity risks. State-directed funds can create duplication and poor allocation; local governments already carry high debt burdens; and state banks may roll over weak projects rather than force rapid restructuring.
But during a long downturn, China can maintain politically prioritized investment even if near-term returns are low. State capital can refinance weak projects, direct procurement toward domestic suppliers, and preserve strategic firms through a period when Western investors may demand immediate retrenchment.
Consider a close look at two investments. NVIDIA agreed to provide up to $105 billion in lease and power guarantees for an OpenAI datacenter in Ohio — in return for OpenAI buying NVIDIA chips. Marvell sold Google a warrant allowing it to acquire up to $12.2 billion of its shares, with vesting tied largely to purchasing milestones under their custom-silicon agreement.
These deals, while not identical, share a central feature: a supplier’s returns are tied not only to selling technology, but to financing, rewarding, or underwriting the customer’s future demand. They shift part of the project’s commercial risk from the data center developer and its lenders onto the supplier of the chips it will buy. Or they shift the risk from an AI startup to a rich search giant.
The danger is not an accounting trick or an automatic replay of the dot-com crash. AI data centers can generate real value, and vendor guarantees can help otherwise viable infrastructure reach financial close. The policy issue is less about whether NVIDIA and Google are “creating” demand than whether financial markets can distinguish independent end-user demand.
China’s incentives in directing state funding to AI extend beyond commercial return; it is technological self-reliance and the ability to keep building despite US export controls. Beijing is also pursuing a more open-weight model strategy. A US-China Economic and Security Review Commission report argues that Chinese labs’ publication of models and weights can help spread Chinese standards, tools, cloud services, chips, and applications across foreign markets.
A Western AI correction would not automatically make China the winner in the AI race. China remains constrained by restrictions on its access to the most advanced chips and chipmaking tools, and Chinese AI firms also face the challenge of turning huge infrastructure spending into sustainable revenue.
The US may innovate faster during a boom, but it can also cut back faster after a correction. China may waste more capital, but it can keep its industrial learning curve moving while competitors pause.
Before a potential downturn, the US needs to establish transparency and stress-testing for AI infrastructure financing. American and European regulators should require clear disclosure of vendor guarantees, equity stakes, prepayments, minimum-revenue commitments, residual-value support, and data center capacity purchase agreements. Banks and institutional investors should stress-test projects against lower GPU rental prices, slowing utilization, fast hardware depreciation, higher power costs, and customer defaults.
Chip and data center investments may create productive infrastructure. But what happens if the bubble bursts? The West should act before it does.
Christopher Cytera CEng MIET is a senior fellow with the Tech Policy Program at the Center for European Policy Analysis and a technology business executive with over 30 years of experience in semiconductors, electronics, communications, video, and imaging.
Bandwidth is CEPA’s online journal dedicated to advancing transatlantic cooperation on tech policy. All opinions expressed on Bandwidth are those of the author alone and may not represent those of the institutions they represent or the Center for European Policy Analysis. CEPA maintains a strict intellectual independence policy across all its projects and publications.
Steel and Silicon: Allied Innovation for the Next Century
2026 CEPA Tech & Security Conference
