Nvidia finds itself designing AI chips for China once again, aiming to thread the needle of ever-tightening U.S. export regulations. The chipmaker has reportedly briefed key Chinese tech firms, including Alibaba, Tencent, and ByteDance, on plans for new processors built specifically to comply with the latest American rules, according to reporting today by The Information.
This effort underscores Nvidia’s determination to hold onto its business in China – a market that accounted for $17 billion, or 13% of its total sales last fiscal year – even after Washington effectively barred its previous generation of China-specific H20 chips just weeks ago. The H20 was an AI accelerator GPU with deliberately reduced performance density (around 2.9 TFLOPS/mm² vs the H100’s 19.4) designed to meet earlier export thresholds.
The latest regulatory hurdle arrived on April 15, when the U.S. Commerce Department mandated an “indefinite license” for selling Nvidia’s H20 and AMD’s MI308 chips into China, Hong Kong, and Macau.
Officials cited national security risks and potential diversion to supercomputing projects. Commerce Department spokesman Benno Kass confirmed the action, stating the department is committed “to acting on the president’s directive to safeguard our national and economic security.”
This policy tightening under the Trump administration forced Nvidia to announce a $5.5 billion charge against its quarterly revenue, attributed to H20 inventory and purchase commitments it could no longer fulfill.
This marked the failure of Nvidia’s second attempt at creating a compliant chip for China, following earlier bans restricting its more powerful A100, H100, A800, and H800 GPUs in 2022 and 2023. Political pressure had also mounted before the ban, with Senator Elizabeth Warren stating in an April 14th letter, “The Commerce Department cannot further delay undertaking necessary and urgent action on the H20 to protect U.S. national security.”
Huawei Steps Up as Nvidia Adapts
As Nvidia recalibrates, domestic competitor Huawei Technologies is capitalizing on the situation. Sources indicate Huawei is beginning mass shipments of its new Ascend 910C AI processor to Chinese customers this month. State-owned telecom operators and AI developers like ByteDance are reportedly already boosting orders for the 910C.
This chip, constructed using advanced chiplet techniques (combining smaller, specialized silicon dies) to integrate two prior-generation 910B processors, is said to offer performance competitive with Nvidia’s H100. Paul Triolo, a partner at consulting firm Albright Stonebridge Group, observed that the US curbs “will mean that Huawei’s Ascend 910C GPU will now become the hardware of choice for (Chinese) AI model developers and for deploying inference capacity.”
Huawei isn’t stopping there; it announced the Ascend 920 chip shortly after the H20 ban, targeting mass production later this year with aims of over 900 TFLOPS (using the BF16 format, optimized for AI) and 4000 GB/s memory bandwidth via HBM3 (High Bandwidth Memory, stacked memory providing fast data access), potentially on SMIC’s 6nm process.
Another chip, the Ascend 910D, is reportedly being prepared with the goal of outperforming Nvidia’s H100, with samples expected soon. This competitive pressure comes as Nvidia’s China data center business could “fall to nearly zero” from its recent 13% share due to the restrictions.
Huawei’s strategy also includes large-scale systems like the CloudMatrix 384 cluster, which uses 384 Ascend 910C chips and boasts high raw performance metrics exceeding Nvidia’s GB200 NVL72 system but reportedly consumes considerably more power – a trade-off potentially palatable given China’s relatively lower industrial electricity costs.
Nvidia’s Public Defense and Private Maneuvers
The rapid competitive maneuvering by Huawei underscores the urgency for Nvidia to find a new path forward in the Chinese market, a challenge the company addressed both publicly and privately. While working on new compliant designs, Nvidia publicly pushed back against the premise of stricter controls just before this news emerged.
In a disagreement with AI developer Anthropic over the U.S. “AI Diffusion Rule,” Nvidia argued against tighter regulations, suggesting they stifle American innovation. Anthropic, in contrast, argued in a blog post that “maintaining America’s compute advantage through export controls is essential for national security and economic prosperity.”
Nvidia countered with a spokesperson stating, “American firms should focus on innovation and rise to the challenge, rather than tell tall tales that large, heavy, and sensitive electronics are somehow smuggled in ‘baby bumps’ or ‘alongside live lobsters.’”
This public defense aligns with CEO Jensen Huang’s actions following the H20 ban. He traveled to Beijing on April 17, meeting with China’s trade promotion head Ren Hongbin and Vice Premier He Lifeng. Huang reportedly told Ren that “China is a very important market for Nvidia” and expressed hope for “continued cooperation,” while He Lifeng stated that China welcomes companies like Nvidia.
Huang also met with DeepSeek founder Liang Wenfeng, reportedly to discuss “new chip designs for Chinese customers,” according to the Financial Times.
This outreach occurred even as Nvidia hadn’t given advance warning about the H20 license requirement to some major Chinese customers, despite being informed by the U.S. days earlier. The market Nvidia aims to re-engage is substantial; Chinese firms reportedly placed at least $16 billion in orders for H20 chips in the first quarter of 2025 alone, partly fueled by demand from AI models like DeepSeek’s.
Sanctions, Supply Chains, and Self-Sufficiency
This cycle of restriction and adaptation plays out against the backdrop of the U.S.-China technology rivalry and Beijing’s determined push for semiconductor self-sufficiency, heavily funded by initiatives like its $47.5 billion “Big Fund.”
Enforcing the U.S. controls remains challenging. Reports persist, though denied by Huawei and TSMC, that Huawei may be sourcing restricted components like TSMC-fabricated chiplets (via Sophgo) and Samsung HBM through intermediaries, prompting U.S. scrutiny of TSMC. Adding another layer, the U.S. House Select Committee on the Chinese Communist Party recently launched a probe into Nvidia’s chip sales across Asia, examining potential export rule violations.
Simultaneously with navigating the China restrictions, Nvidia announced a $500 billion plan for U.S.-based AI supercomputer production, primarily in Texas. Given that Nvidia knew about the H20 ban before this announcement, it paints a picture of a company making complex strategic calculations across a fractured global technology map.
