Alibaba and ByteDance are preparing to place significant orders for Huawei’s newest AI chip, the Ascend 910C, as tightening U.S. export controls squeeze access to high-end Nvidia GPUs, according to sources familiar with the plans. The reported orders reflect a broader strategic shift among China’s largest technology firms, which now face pressure from both Washington and Beijing to rethink how they source the processors that power artificial intelligence. What makes this moment distinct is the collision of two opposing policy forces: American restrictions designed to slow China’s AI progress and Chinese government directives pushing domestic alternatives.
Why U.S. Export Rules Are Forcing the Shift
The legal architecture behind these procurement decisions traces back to the U.S. Commerce Department’s Bureau of Industry and Security, which administers the export regulations governing AI chip sales to China. These rules require export licenses for high-performance AI accelerators shipped to Chinese buyers, and the licensing process has grown increasingly restrictive since late 2022. For companies like Alibaba and ByteDance, which have built cloud and AI infrastructure around Nvidia hardware, the practical result is deepening uncertainty about whether they can continue sourcing the chips they need.
Nvidia’s own financial disclosures confirm how significant these restrictions have become. In its regulatory filings, Nvidia describes export controls as a material risk, details licensing uncertainty for China-bound shipments, and acknowledges ongoing product adjustments for restricted markets. The company also discloses substantial China-related revenue exposure, a figure that has drawn scrutiny from investors trying to gauge how much business Nvidia stands to lose. Those filings make clear that the export regime is not a theoretical constraint but an active drag on one of the world’s most valuable chipmakers.
Beijing’s Dual Signals on Chip Procurement
China’s policy response has been anything but simple. On one hand, Chinese regulators have reportedly banned foreign AI chips from state-funded data centers, according to Reuters reporting, a move that creates direct policy pressure favoring domestic suppliers like Huawei. On the other hand, competing reports paint a more complicated picture. Bloomberg has reported that Chinese authorities told Alibaba and other tech firms to prepare orders for Nvidia H200 chips, and a separate Reuters account indicated that Beijing gave a green light for ByteDance, Alibaba, and Tencent to purchase over 400,000 Nvidia H200 units.
These two signals appear contradictory, but they likely reflect different procurement channels. State-funded infrastructure may be reserved for domestic chips, while commercial operations at private tech giants retain some flexibility to buy foreign hardware when available. The tension between these directives reveals a government trying to accelerate self-sufficiency without crippling the AI capabilities its companies need to compete globally. For Alibaba and ByteDance, the practical takeaway is that relying solely on Nvidia is no longer a safe bet, even when Beijing occasionally approves specific purchases.
Alibaba’s Supply Chain Calculus
Alibaba’s public filings offer a window into why the company would turn to Huawei. In its annual report, Alibaba discloses AI and cloud capital expenditure priorities alongside explicit supply-chain risk factors tied to export controls. The filing acknowledges reliance on third-party accelerators, a category that includes Nvidia GPUs, while also referencing investments in in-house chips such as those developed by its T-Head semiconductor unit.
That dual-track approach, buying externally while building internally, has been Alibaba’s hedge against supply disruptions. But the hedge has limits. T-Head chips are not yet competitive with Nvidia’s top-tier products for large-scale AI training, and export restrictions make it harder to fill the gap with American hardware. Huawei’s Ascend line represents a third path: a domestic supplier whose chips are improving and whose supply chain is not subject to U.S. licensing requirements. For Alibaba’s cloud division, which sells AI computing services to thousands of Chinese businesses, securing a reliable chip supply is not just a technical question but a commercial imperative.
ByteDance’s Training Ambitions
ByteDance, the parent company of TikTok and Douyin, has its own reasons for turning to Huawei. Sources told Reuters that ByteDance plans to train a new AI model using Huawei Ascend chips, a significant step that would mark one of the first major large-language model efforts built primarily on Chinese hardware. The shift reflects both the tightening export environment and ByteDance’s strategic calculation that it cannot afford to depend on a supply chain that Washington can restrict at any time.
Training a frontier AI model requires tens of thousands of high-performance chips running in coordinated clusters for weeks or months. If ByteDance can demonstrate that Huawei’s Ascend processors handle this workload effectively, it would validate Huawei’s position as a viable alternative to Nvidia for the most demanding AI applications. That outcome would carry implications well beyond ByteDance, potentially encouraging other Chinese firms to follow suit and giving Huawei a major reference customer for its AI chip business.
Huawei’s Ascend Chips and the Competitive Gap
A U.S. research brief details how sanctions have spurred Huawei’s AI chip development as the company works to challenge Nvidia’s dominance. The report notes that Huawei has poured resources into its Ascend architecture, aiming to close the performance gap with leading U.S. accelerators while optimizing for Chinese software ecosystems. The Ascend 910C is part of that push, positioned as a training-class processor intended to anchor large clusters for generative AI workloads.
Even so, Huawei faces technical and ecosystem hurdles. Nvidia’s advantage has been built not only on raw chip performance but also on its CUDA software stack, developer tools, and a vast base of existing models optimized for its GPUs. Huawei has responded by investing in its own software platform and by courting Chinese AI labs with engineering support and incentives to port models. The prospective orders from Alibaba and ByteDance suggest that, at least for some workloads, Ascend-based clusters are now viewed as good enough to justify the migration costs.
Performance is not the only metric that matters. For Chinese firms, predictability of supply has become just as crucial. A slightly less capable chip that can be bought in volume and deployed without fear of sudden export bans may be more valuable than a top-tier GPU whose availability is uncertain from one quarter to the next. Huawei’s domestic manufacturing partnerships, and its alignment with Beijing’s industrial policy priorities, give customers some confidence that their hardware roadmaps will not be derailed by foreign governments.
Strategic Implications for the AI Ecosystem
The emerging pivot toward Huawei has several broader consequences. First, it accelerates the fragmentation of the global AI hardware ecosystem. As Chinese firms build and tune models on Ascend chips, their software stacks may diverge further from those used in the United States and Europe, complicating cross-border collaboration and model portability. Over time, this could lead to parallel AI ecosystems with different standards, optimization targets, and performance characteristics.
Second, the shift could blunt some of the leverage Washington hoped to gain from export controls. While U.S. rules have clearly constrained access to Nvidia’s latest products, they have also strengthened the business case for domestic alternatives. If Huawei succeeds in scaling production and winning marquee customers, it will emerge as a more formidable competitor in global AI hardware markets, not just a substitute for sanctioned Chinese buyers.
Third, Beijing’s mixed signals on Nvidia purchases underscore the balancing act it faces. Completely cutting off foreign chips could slow China’s AI progress in the near term, but allowing unlimited imports would undermine efforts to build a self-reliant semiconductor base. The current approach, steering state-backed projects toward domestic hardware while selectively approving foreign orders for private firms, suggests a phased transition in which companies like Alibaba and ByteDance are expected to help prove out Chinese alternatives before the policy screws tighten further.
For now, the reported orders for Huawei’s Ascend 910C chips are best understood as part of a broader hedging strategy. Alibaba and ByteDance are not abandoning Nvidia so much as they are diversifying away from a single point of geopolitical failure. By building parallel compute stacks around both Nvidia and Huawei hardware, they aim to preserve access to cutting-edge AI capabilities regardless of how export rules or domestic procurement mandates evolve.
The success of this strategy will depend on how quickly Huawei can improve its chips, how effectively Chinese software developers adapt to the new hardware, and whether U.S. policymakers continue to tighten controls. What is clear is that the era in which Nvidia could count on near-unquestioned dominance of China’s AI data centers is ending. In its place is a more contested, politically charged market in which chip choices are as much about regulatory risk and national strategy as they are about benchmarks and price-performance ratios.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
