Export controls are a defensive crouch. Wendy R. Anderson, writing and op-ed in the Washington Post this week as a former senior official in the Defense and Commerce departments, argues that the United States needs to stop treating the global artificial intelligence infrastructure race as a containment problem and start treating it as a market competition problem. She is right about the problem. Whether the institutions she is asking to execute can move at the speed the market requires is a harder question.
Anderson reads the competitive threat correctly. The question her op-ed does not fully answer is whether the institutions she is asking to execute can actually move at the speed the market requires.
The Part Anderson Gets Right
Earlier this year I wrote about the frog in the well problem in AI strategy: the assumption, common in Western boardrooms and policy circles, that the AI infrastructure race is being run on Nvidia hardware and decided in Washington. It is not. Huawei's Ascend 950PR chip is shipping to ByteDance and Alibaba at scale. DeepSeek's V4 model is being trained to run on that hardware. A parallel CUDA-compatible software layer, Huawei's Compute Architecture for Neural Networks framework, is reducing the switching cost for developers who built on Nvidia tooling.
Anderson adds the layer my post did not cover: the commercial deployment strategy beyond China's borders. Huawei is not just selling chips to Chinese companies. It is selling complete AI infrastructure packages to governments and enterprises in markets where U.S. companies have historically been slow or absent. Chips, cloud infrastructure, implementation support, and financing, bundled together into a single procurement decision. That is a different competitive threat than raw hardware benchmarks. A country choosing Huawei for its national AI infrastructure is not just choosing a chip. It is choosing a vendor relationship, a software ecosystem, and a financing structure that may span a decade.
This is the strategic reality that export controls cannot address. Restricting Nvidia chip sales to China does not stop Huawei from building an installed base in Malaysia, Indonesia, Saudi Arabia, or Nigeria. If anything, it accelerates Huawei's pitch: we are here, we are reliable, and we are not subject to the policy volatility of Washington.
Where the Playbook Gets Hard
Anderson's proposed response has three components: a trusted-partner framework for chip and cloud exports, reliable software access for allies in contested markets, and directing the U.S. International Development Finance Corporation toward artificial intelligence infrastructure investment. The corporation has a $60 billion statutory investment cap. That is real capital. Whether the agency can deploy it into AI infrastructure on a timeline that competes with Huawei's existing deal pipeline is a different question.
Huawei's bundled model works because it removes procurement complexity. A government ministry in Southeast Asia does not need to separately contract a hardware vendor, a cloud provider, a systems integrator, and a financing institution. Huawei handles the bundle. The U.S. alternative, even with DFC capital deployed, requires assembling a coalition of private sector companies, coordinating export licensing across multiple agencies, and managing a procurement process that a counterpart government has limited bandwidth to run. The competitive disadvantage is structural, not financial.
The corporation's track record in technology infrastructure is thin. Established in 2019 under the Better Utilization of Investments Leading to Development Act, the agency was built with two rationales: advancing U.S. development finance and countering Chinese investment in developing markets, particularly the Belt and Road Initiative. It has deployed capital in telecommunications, notably supporting alternatives to Huawei in 5G markets, but AI infrastructure is a different scale of commitment. A national AI data center is not a cell tower contract. The integration requirements and the ongoing vendor dependency are substantially larger. Corporation financing gets a country into the room. It does not hand them a working product on day one, which is what Huawei's bundled offer does.
The Policy Creates the Opening. Someone Still Has to Walk Through It.
The better parallel is the hyperscaler expansion of the 2010s. When Amazon Web Services, Google Cloud, and Microsoft Azure entered Southeast Asian markets, they won because they offered a credible, documented, scalable product that a procurement team could evaluate and a finance team could model. Policy helped at the margins. The product did the work.
U.S. export frameworks and U.S. International Development Finance Corporation financing create the conditions to compete. They do not substitute for a bundled offer that a government ministry can actually procure. The missing piece in Anderson's argument is the product layer: which U.S. companies are positioned to bundle hardware, cloud, implementation, and financing into a single offer for a national artificial intelligence infrastructure buyer, and what policy changes are required to let them do it at the speed the market is moving?
Huawei already has that answer.
Anderson is writing about policy levers, not product strategy. For a chief information officer whose organization operates in the markets being contested, the policy debate is background noise. The operational question is which vendors can show up with a credible bundled offer, and on what timeline. Right now, in Southeast Asia and the Gulf, Huawei can answer that question more concretely than the U.S. technology sector can.
Five Eyes Covers the Wrong Geography
Anderson calls for a Five Eyes export corridor as a trusted-partner framework for AI infrastructure. The Five Eyes alliance, comprising the United States, United Kingdom, Canada, Australia, and New Zealand, already coordinates on sensitive infrastructure decisions. Extending that to AI hardware export coordination is a natural institutional step, and it is probably the right one for those five countries.
The problem is that the markets where Huawei is winning are not Five Eyes countries. They are Indonesia, Malaysia, Thailand, the United Arab Emirates, Saudi Arabia, Kenya, and Nigeria. An export framework optimized for trusted allies does not directly address the procurement decisions being made in Kuala Lumpur or Riyadh. Anderson acknowledges this and argues for a tiered system that extends to a broader set of strategic partners. That tiered system does not currently exist in operable form, and building it requires exactly the kind of interagency coordination that U.S. policy machinery executes slowly.
If your organization operates in Southeast Asia, the Gulf, or Sub-Saharan Africa, the infrastructure your regional competitors are evaluating may not be built on Western silicon or Western cloud. The AI offense debate in Washington matters, but it runs on a policy timeline. The procurement decisions are happening now. Which of your current AI vendors can show you a credible deployment path in those markets that does not depend on U.S. export policy resolving in your favor?
Anderson, Wendy R. "The U.S. Needs to Go on AI Offense." The Washington Post, 8 Apr. 2026.
Bellamkonda, Shashi. "The Frog in the Well Cannot See the Chip War." Shashi.co, Apr. 2026, shashi.co/2026/04/the-frog-in-well-cannot-see-chip-war.html.
Bellamkonda, Shashi. "Huawei's 950PR: The Chip That Learned to Speak CUDA." Shashi.co, Mar. 2026, shashi.co/2026/03/huaweis-950pr-chip-that-learned-to.html.
Council on Foreign Relations. "China's AI Chip Deficit." Council on Foreign Relations, Dec. 2025, cfr.org.
Reuters. "DeepSeek V4 to Run on Huawei Chips." Reuters, Apr. 2026, reuters.com.
Reuters. "Huawei's New AI Chip Finds Favour With ByteDance, Alibaba." Reuters, 27 Mar. 2026, reuters.com.
U.S. International Development Finance Corporation. "About DFC." DFC.gov, 2026, dfc.gov.
