Two superpowers walked into Shenzhen this week with completely different playbooks for artificial intelligence, and honestly, the gap between them has never looked wider.
At the APEC summit, US President Donald Trump came out swinging against what he called "globalist regulation" of AI, warning against rules that would control or stifle superintelligence. Days earlier, Washington tightened export controls on advanced AI chips routed through offshore data centers. Beijing took the opposite tack—pushing AI cooperation proposals, pitching applied industrial scale, and holding up the Greater Bay Area as an open hub for hardware and models.
Short headlines love framing this as a simple race over who builds the smartest model first. That misses the real battle. It's a tug-of-war over compute bottlenecks, third-country loopholes, and who sets the rules for everyone else.
What the APEC Summit Actually Means for AI and Tech Competition
If you read quick dispatch summaries on AI and tech competition between the US and China around APEC, you might walk away thinking leaders are about to sign a tidy truce on algorithms. They won't.
Washington's posture right now is a paradox. On one hand, Trump told APEC leaders in Shenzhen that the US opposes globalist rules that slow down AI development. On the other hand, the US Commerce Department rolled out new rules on November 16 targeting remote access to advanced chips via overseas data centers. You don't need a degree in trade law to see the tension there. American policy rejects top-down international AI governance while doubling down on unilateral export licensing and hardware chokepoints.
Beijing is playing a very different hand. Ahead of the summit, China floated an AI cooperation push aimed at cross-border dialogue, safety research, and predictability. Chinese officials also used the Shenzhen stage—right in the middle of the Guangdong–Hong Kong–Macao Greater Bay Area—to pitch practical industrial AI, mature manufacturing nodes, and open-weight ecosystems.
Watching tech policy up close, I've seen this pattern play out repeatedly in boardrooms and policy briefings. Whenever one side locks down upstream hardware, the other side competes on deployment speed, cost, and supply-chain integration. Neither side is standing still.
Why Export Controls Keep Shifting Toward Offshore Compute
Early export controls focused on a straightforward question: can a Chinese buyer import a restricted AI chip directly? By 2026, that question feels almost quaint.
Modern frontier training and inference don't require owning the physical server rack down the hall. A company can rent cloud capacity or route workloads through facilities in third countries. That's why the November 16 Commerce Department rule zeroed in on overseas data centers and flagged five markets—Singapore, Malaysia, Thailand, Indonesia, and Brazil—for heightened licensing scrutiny, giving operators a 90-day compliance window.
Chip restrictions work best on paper when hardware is static; they get messy fast when compute can be rented across borders.
Let's be real about how hard this is to enforce. Data center operators have to identify end users supplying significant computing power to China-headquartered firms, while US-operated facilities in those five countries and certain consumer platforms face different treatment. Even then, enforcement depends on cloud telemetry, contract audits, and diplomatic cooperation.
Industry voices have also been blunt about collateral damage. Analysts and executives have pointed out that strict hardware curbs pressure domestic Chinese alternatives and reshape global market share, even as Washington tries to keep its highest-end processors—like Blackwell-class systems—out of rival hands. Whether export controls permanently slow China's frontier capability or mostly force architectural workarounds is genuinely debated among defense and tech analysts. Anyone claiming certainty on that score is guessing.
The Middle East and Southeast Asia Are Becoming Swing Arenas
APEC isn't just about Washington and Beijing talking past each other in a conference room. It's about who wins over everyone else in the room.
Look at the Gulf and Southeast Asia. American cloud and chip giants have poured billions into regional AI data centers and partnerships, such as deals tied to Saudi Arabia's Humain and broader Middle East infrastructure pledges. At the same time, Washington has launched supply-chain alignments like Pax Silica—where the US and nine partners signed the San Francisco Declaration—to build trusted hardware and mineral networks.
Meanwhile, Chinese firms and officials have pushed cloud infrastructure, open models, and regional digital partnerships across Asia and the Gulf. In real-world deployments, countries often mix and match—pairing American compute hardware with Chinese models or regional software stacks. For emerging economies, "choosing a side" often looks less like ideological loyalty and more like practical hedging.
AI Safety Talks vs. Geopolitical Distrust
One detail from the APEC window deserves closer attention than it usually gets: Washington and Beijing haven't completely stopped talking about AI risk.
Following a September arrangement to open a direct AI incident communication channel, US Treasury Secretary Scott Bessent indicated that bilateral discussions around the Shenzhen summit would touch on AI risks, pre-release testing practices, and keeping humans in control. At the same time, Beijing's Ministry of Commerce has moved on draft export rules targeting AI tools capable of autonomously generating or modifying malicious code.
Here's the catch. Having a safety channel doesn't mean the two sides define "safety" the same way.
Historically, Chinese pre-release reviews have leaned heavily toward content compliance and political control, whereas Western safety evaluations have focused more on capability risks like cyber offense, bio-risks, or loss of oversight. You can exchange technical notes on testing without agreeing on philosophy. Plus, broader geopolitical friction—from tariff truces approaching expiration on January 10 to regional security flashpoints—can freeze working-level tech talks overnight.
Common Mistakes Businesses Make Reading APEC Tech News
When geopolitical summits hit the front page, companies tend to overreact in one of two ways. Both are expensive.
First, some teams assume a bilateral summit means export rules will suddenly vanish. In reality, executive meetings rarely erase years of layered export controls, statutory pressure in Washington, or industrial policy goals in Beijing. Second, other teams treat the US and China as hermetically sealed silos, ignoring how third-country cloud hosting, open-weight models, and cross-border supply chains still touch their stack.
In my experience advising product and infrastructure leaders, the smartest move isn't chasing every headline. It's building portability and auditability into your compute and model layer so you aren't trapped if a single vendor, cloud region, or licensing rule changes.
Practical Next Steps If You Build or Buy AI
If you run a tech company, manage cloud infrastructure, or invest in AI hardware and software, treat APEC as a signal to tighten your operational homework—not a reason to panic.
- Map your compute and hosting footprint: Check where your training and inference workloads actually run, especially if you rely on third-party clouds or data centers in Southeast Asia or Latin America.
- Audit model and dependency provenance: Document where your foundation models, weights, tooling, and hardware components originate so compliance reviews don't turn into fire drills.
- Avoid single-vendor lock-in: Keep evaluation pipelines flexible across model families and cloud providers so you can adapt without rewriting core workflows.
- Track regulatory deadlines, not just speeches: Pay close attention to concrete compliance windows—such as the 90-day licensing timeline on the offshore data center rule and upcoming trade truce dates—rather than broad summit rhetoric.
The US and China aren't going to stop competing over chips, talent, and influence anytime soon. Build systems that stay resilient either way.