In May 2026, members of the House Foreign Affairs Committee travelled to Silicon Valley to discuss AI and export controls with leading technology companies.
That sentence would have been incoherent a decade ago. Foreign affairs committees visit capitals, not campuses.
It is now routine — and it marks the clearest signal yet that the boundary between American technology policy and American foreign policy has dissolved entirely.
Key Takeaways
- The consensus has fractured. What was bipartisan agreement on restricting China’s chip access has become a triangular standoff between Congress, the White House and the chip industry.
- The money dwarfs the diplomacy. US hyperscalers are expected to invest $650 billion into AI in 2026 alone.
- Enforcement is widening. DOJ and BIS investigations now reach intermediaries, financial institutions and data centre operators.
- Chinese models are inside US companies. Cost and speed advantages are driving adoption regardless of policy.
- Platforms perform state functions. Private firms took operational roles in Ukraine historically held by allied governments.
From Containment to Contest
The original logic was straightforward. Progressively tighten export controls on high-end chips and the equipment used to fabricate them, while investing in domestic capacity through the CHIPS and Science Act.
That framework is no longer stable. The United States is intensifying its technology confrontation with China, but the situation has become harder to read, with the former bipartisan consensus replaced by a three-way contest reshaping strategy in real time.
The three positions:
Congress wants harder limits. At the January 2026 House Foreign Affairs hearing on competing with China on AI, Chairman Brian Mast noted that private companies in China purchase chips for applications that ultimately assist adversary militaries. Former Deputy National Security Advisor Matt Pottinger argued Congress needs to step in and put durable guardrails in place.
The industry wants market access. Restrictions cede the largest addressable market for AI infrastructure to domestic Chinese alternatives.
The White House wants leverage. Export licences are negotiating instruments, and instruments lose value when Congress removes discretion over them.
The Legislative Vehicles
Two bills define the terrain:
- The GAIN AI Act, which would impose further restrictions on exports of US-made chips.
- The AI Overwatch Act, introduced December 2025, which would require congressional review of export licences for advanced AI chips to China — giving both chambers 30 days to review and potentially block a licence, modelled on existing oversight of arms sales.
That modelling is the tell. Advanced semiconductors are being legislatively treated as weapons systems.
The Export Strategy Has a Second Half
Restriction is only one half of American technology statecraft. The other is export promotion.
This is the “tech stack diplomacy” model: partner countries adopt American chips, American cloud infrastructure and American model standards as an integrated package, creating dependencies more durable than any treaty.
Where It Runs Into Trouble
Financing is inadequate to the task. Against $650 billion in private hyperscaler AI investment, DFC and EXIM’s combined funding must also cover transportation, energy and other projects. EXIM leadership has signalled plans to deploy up to 100% of its remaining $100 billion in energy and critical minerals projects, leaving limited government incentive to offer already well-resourced American companies.
Partner countries resist lock-in. Prospective partners explicitly seek to avoid technological lock-in, with sovereign AI initiatives reflecting mounting anxiety about foreign control. The UAE case is distinctive given Abu Dhabi’s technological capacity and capital.
The stack leaks at the model layer. Airbnb’s CEO noted the company’s AI customer service agent relies heavily on Alibaba’s Qwen model because it is fast and cheap, with reporting suggesting many Silicon Valley start-ups are similarly inclined.
That last point is the strategy’s central vulnerability. You can control chip exports. You cannot control the adoption of open-weight models — and American start-ups are choosing Chinese ones on economics.
Compliance Is the New Geopolitics
For companies, this has stopped being an abstract policy debate.
Open questions shaping 2026 compliance include whether a full replacement for the AI Diffusion Framework arrives, and the extent to which governments enforce security, monitoring and reporting requirements for permissible exports.
| Actor | Exposure |
|---|---|
| Chip designers | Licence dependency, revenue concentration |
| Cloud providers | End-use verification, deployment monitoring |
| Data centre operators | Customer screening obligations |
| Financial institutions | Transaction diligence on restricted end-users |
| Resellers/intermediaries | Highest enforcement risk |
Any company touching advanced compute now carries foreign-policy exposure it did not underwrite.
Platforms as Foreign Policy Actors
The Ukraine precedent reshaped expectations permanently. During the Russia-Ukraine war, private technology companies performed operational roles historically held by allied governments — Microsoft supporting Ukrainian network cybersecurity, SpaceX’s Starlink providing connectivity for military and civilian communication.
The framing has since broadened. In 2026, diplomacy extends far beyond embassies and summit halls, marking a transition from a Westphalian order grounded in territorial sovereignty toward an era shaped by silicon statecraft, where influence circulates through platforms, algorithms, data centres and corporations.
This model prioritises resilient supply chains, trusted partnerships and coordinated export controls on advanced chips and AI accelerators — operating as a coalition of capabilities rather than a traditional military bloc.
What Investors Should Take From This
Three practical implications:
- Regulatory risk is now a primary valuation input for semiconductor and cloud names. A licensing change can alter addressable market overnight in ways earnings models do not capture.
- Geographic revenue disclosure matters more than it used to. Concentration in restricted or restriction-eligible markets is a discrete risk factor.
- The compliance cost curve is steep and rising. Firms without mature export-control frameworks face enforcement exposure that scales with their growth.
What This Means for the Global Market in 2027
Congressional review would change the pricing of chip revenue. If the AI Overwatch Act model passes, export revenue to sensitive markets becomes contingent rather than contracted — a discount, not a multiple.
Sovereign AI is the counter-trend to watch. Partner-country resistance to lock-in is producing domestic model and infrastructure programmes. Every one of these erodes the stack diplomacy thesis.
Open-weight models are the unregulated channel. Chip controls address hardware. The competitive pressure from cheap, capable open models runs entirely outside that framework — and US start-up adoption suggests the market has already voted.
Power and minerals become foreign policy. With EXIM prioritising energy and critical minerals, the diplomatic contest is shifting upstream from chips to the electricity and materials that produce them.
Corporate foreign policy capacity becomes a competitive advantage. Firms that build genuine government-relations and export-compliance capability will access markets that competitors cannot.
Frequently Asked Questions
Why is Silicon Valley involved in foreign policy?
Control of semiconductors, AI models and digital infrastructure has become a primary instrument of national power. House Foreign Affairs Committee members travelled to Silicon Valley in May 2026 specifically to discuss AI exports and controls.
What are AI chip export controls?
Restrictions on exporting advanced semiconductors and fabrication equipment to designated countries. Proposed legislation including the GAIN AI Act and AI Overwatch Act would tighten them further and add congressional review.
How much are tech companies investing in AI?
US hyperscalers are expected to invest approximately $650 billion into AI during 2026 alone — a figure that dwarfs available US government export financing.
Do US companies use Chinese AI models?
Yes. Airbnb’s CEO has confirmed the company’s AI customer service agent relies heavily on Alibaba’s Qwen model, and reporting suggests many Silicon Valley start-ups use Chinese models for cost and speed reasons.
