Two closely timed headlines on Monday captured the direction of the global AI race: China is building a more self-contained technology stack, while Washington is tightening the rules around Chinese AI inside government systems. For investors and analysts, that combination matters. It points to a market where China’s large-model developers, chipmakers, and software infrastructure are no longer just competing on model quality. They are also aligning hardware, software, and procurement in ways that could reshape costs, supply chains, and commercial adoption across Asia and emerging markets.
According to The Information, as reported in ZeroHedge, DeepSeek has moved beyond being a Huawei customer and is now effectively a co-designer of Huawei’s next-generation silicon. Huawei engineers have reportedly been stationed at DeepSeek’s offices for an extended period, folding the AI lab’s requests and suggestions directly into chip design decisions. That is a notable milestone because it suggests a tighter loop between model builders and chip designers, similar to how leading US players have optimized their stacks over time.
The report also says Huawei plans to give DeepSeek most of the supply of its Ascend 960DT training chip when it launches next year. On the software side, both companies have been using TileLang as an alternative to Nvidia’s CUDA, with DeepSeek’s models now largely handled by TileLang running on Huawei chips. In practical terms, that is the blueprint for a homegrown AI ecosystem: local models, local chips, and local tooling.
The same report adds that DeepSeek said in September it aimed to deploy over 160,000 Huawei AI chips. It also says that on Sept 30 DeepSeek open-sourced an Ascend port of TileLang, along with Ascend versions of its DeepGEMM and DeepEP libraries, and explicitly credited Huawei while touting a jointly optimized 128-chip Ascend 950 supernode. Huawei, meanwhile, pulled the launch of the Ascend 960DT forward to the first quarter of 2027, three quarters ahead of schedule.
That matters because the real story is not a single product or one isolated benchmark. It is the pace of industrial coordination. China’s AI sector appears to be moving toward a more integrated stack where the lab, the chipmaker, and the software layer are all working toward the same optimization target. For global investors, that raises the odds of faster domestic deployment, stronger cost discipline, and a broader export opportunity for Chinese AI infrastructure over time.
The same day, the Department of Homeland Security published an order directing federal civilian agencies to remove DeepSeek products, and any third-party services built on the Chinese model, from their IT systems. The order is Exclusion and Removal Order 2026-01, issued under the 2018 Federal Acquisition Supply Chain Security Act. It was signed on July 24 and only became public Monday through a Federal Register filing, according to the report.
The order covers civilian executive branch agencies such as Treasury, State, and Justice. The Pentagon and the intelligence community are carved out, largely because the Pentagon already banned DeepSeek and High-Flyer AI via the FY2026 NDAA. DHS says DeepSeek’s models pose multiple threat vectors and may transfer sensitive data to the PRC, citing China-based servers, PII collection at sign-up, and possible user fingerprinting. The order stops short of an Entity List designation, so commercial use is said to remain untouched for now.
The most important detail may be the reach of the order. It goes beyond DeepSeek’s own app and extends to third-party products developed using DeepSeek implementations. In plain English, that means vendors selling tools built on fine-tunes or distillations of DeepSeek’s open weights could face procurement problems with the US government. That is a bigger issue than a simple app ban. It turns model provenance into a compliance question for contractors and subcontractors across the federal software supply chain.
That is why this is not just a security story. It is a market-structure story. When government procurement starts treating model origin as a screening factor, Chinese open models lose some room in the highest-trust segments of the US enterprise market. But the same policy pressure also pushes Chinese developers further toward domestic hardware and software ecosystems. In effect, Washington’s restrictions and Beijing’s industrial policy are reinforcing each other, even if neither side intends that outcome.
The sequence now looks familiar. NASA, the Navy, and roughly 17 states blocked DeepSeek on government devices in early 2025. The FY2026 NDAA pushed it out of the Pentagon and its contractors. Now civilian agencies are following. The Senate Armed Services Committee’s version of the FY2027 defense bill would extend the Pentagon ban beyond DeepSeek to AI from ten other Chinese companies and their derivative models. Taken together, those steps show how fast policy can spread once a model becomes a security concern.
At the same time, the broader Chinese AI field keeps advancing. In July, after Moonshot’s Kimi K3 launch, Axios reported that the US administration was reviving plans for de facto bans on Chinese open models, with Kimi and Alibaba’s Qwen in the crosshairs through procurement rules, Entity List threats, and security advisories. Scott Bessent added a sanctions threat over distillation a day later. That tells investors the policy risk is not isolated to one company. It is widening across the entire Chinese open-model stack.
There is another layer here: Beijing may be moving to define the boundaries of its own AI ecosystem. The report says China’s commerce ministry floated export controls that would, for the first time, cover AI model weights and source code above certain performance thresholds, while promising a fast-track license for open-source releases. Chinese AI commentators took that to mean the next Kimi, GLM, and DeepSeek flagships may never be open-sourced at all.
If that interpretation holds, the era of easy access to frontier-grade Chinese open weights could close from both directions at once. Washington would restrict usage in government and procurement. Beijing would restrict export of the very weights and code that made Chinese open models so widely usable. That would leave the market with less cross-border sharing, but potentially stronger domestic clustering on both sides of the Pacific.
For investors, the key insight is that China’s AI strategy is becoming more vertically integrated, not less. DeepSeek’s reported work with Huawei shows how rapidly Chinese firms can align around a shared stack when policy and market incentives point in the same direction. Huawei’s willingness to accelerate the Ascend 960DT schedule, and DeepSeek’s use of TileLang and Ascend libraries, suggest a serious effort to reduce dependence on US technology.
That is bullish for China’s domestic AI supply chain and for companies tied to local compute, software tooling, and enterprise deployment. It is also a reminder that Chinese AI should no longer be viewed only through the lens of model quality. The more relevant frame is ecosystem resilience. If China can keep improving the loop between chips, frameworks, and applications, it gains leverage not just at home, but across markets that want low-cost AI infrastructure without US policy exposure.
Monday’s twin headlines show a world splitting into more distinct AI stacks. On one side are CUDA, Nvidia, and closed US frontier models. On the other are TileLang, Ascend, and increasingly coordinated Chinese open-model infrastructure. DeepSeek’s reported co-design role with Huawei is an important milestone in that transition, and Washington’s latest DHS order shows how seriously US agencies are treating the Chinese AI layer.
For analysts, the next phase is clear: watch how fast Huawei’s Ascend roadmap advances, how far DeepSeek’s software ecosystem spreads, and whether Beijing’s export-control thinking hardens around model weights and source code. The future of AI competition is no longer just about who has the best model. It is about which country can build the most complete stack, and move it fastest.
