Nvidia’s chief executive Jensen Huang says the company’s next artificial-intelligence chip for China will not come from the Hopper architecture, underscoring the challenges US export curbs pose to America’s most valuable semiconductor designer.
“It’s not Hopper because it’s not possible to modify Hopper anymore,” Huang told reporters during a livestreamed appearance in Taipei on 17 May.
His comments followed a whirlwind April visit to Beijing, where he met Vice-Premier He Lifeng and Shanghai mayor Gong Zheng only days after Washington tightened rules on shipments of Nvidia’s H20 accelerator.
The H20 is currently the only Hopper-based product the company may sell in China without an export licence.
Earlier this month Reuters reported that Nvidia intends to launch a downgraded version of the chip within two months in an effort to defend market share against domestic rival Huawei.
Huang confirmed that China remains “critical” to Nvidia’s growth strategy despite an increasingly complex regulatory landscape.
The world’s second-largest economy generated $17 billion in revenue for the firm during the fiscal year that ended on 26 January, or roughly 13 percent of total sales.
Alibaba, Tencent and ByteDance alone have reportedly placed $18 billion in H20 orders since January, highlighting the scale of local AI demand.
Searching for a new foothold
Sources tell Reuters that Nvidia has begun scouting locations in Shanghai’s Minhang and Xuhui districts for a dedicated research-and-development centre.
City officials are said to be offering tax breaks and a sizable land parcel to lure the project, part of broader efforts to attract global tech investment.
Analysts see the potential facility as a hedge that would keep Nvidia close to key customers while insulating some engineering work from export restrictions.
Shanghai already hosts major operations for Tesla and scores of international chipmakers, providing a deep talent pool.
Questioning export-control logic
Huang reiterated his long-held view that earlier US rules aimed at hampering AI diffusion were misguided.
He argued that regulations should strive to maximise US technology globally instead of fragmenting supply chains.
The current framework, formalised at the end of the Biden administration, restricts advanced AI accelerators to most jurisdictions and caps performance thresholds for those that can be shipped.
Former president Donald Trump, now campaigning for a return to the White House, has vowed to scrap the measures if elected.
Lobbyists representing US chip firms contend the curbs could backfire by accelerating indigenous hardware development in China.
Huawei’s Ascend 910B, unveiled last year, already competes head-to-head with the H20 in many benchmark tests.
Downgrade strategy carries risks
Cutting memory bandwidth or disabling interconnect lanes may keep Nvidia on the right side of regulators, but customers could balk at lower performance.
“Chinese cloud providers will compare cost-per-training-token and may favour domestic chips if the gap widens,” said Kevin Xu, founder of the consultancy Interconnected.
Xu noted that Beijing’s subsidy programmes for local semiconductors further tilt the playing field.
Some analysts warn that segmenting product lines for different regions could inflate Nvidia’s engineering costs and complicate its long-term roadmap.
However, they add that walking away from such a large market is not an option for a company whose valuation increasingly depends on global AI dominance.
Political cross-currents in Washington
The Commerce Department maintains that the diffusion rules are essential to prevent potential military applications of cutting-edge AI systems.
Lawmakers on Capitol Hill remain divided, with hawks urging tougher action and moderates worried about collateral damage to US firms.
An attempt to pass a bipartisan bill that would mandate licensing for any AI chip above a certain flops-per-watt threshold stalled in committee last month.
Industry insiders expect export-control policy to feature prominently in the 2026 US presidential race, adding further uncertainty.
China’s AI ambitions undimmed
Despite restrictions, China’s AI sector is projected to hit $50 billion in annual revenue within three years, according to Huang.
He warned that shutting US companies out of that growth would amount to a “tremendous loss” for both sides.
Chinese regulators, for their part, have signalled openness to foreign participation so long as national-security standards are met.
Local media report that Shanghai is drafting guidelines specifically tailored to foreign AI chip developers, a move viewed as a conciliatory gesture toward Nvidia.
What comes after Hopper
Huang declined to provide details on the architecture succeeding Hopper but hinted it would incorporate lessons learned from operating under export controls.
Some observers speculate that Nvidia could design a modular platform with performance governors that can be toggled at the factory, simplifying compliance.
Others believe the company may prioritise software differentiation through CUDA-compatible optimisation layers that run equally on high- and mid-tier silicon.
Meanwhile, board partners are pressing for clarity on memory configurations to prepare server-rack designs ahead of China’s all-important Singles’ Day shopping season.
Supply-chain managers caution that lead times for high-bandwidth memory remain elevated, making early specification vital.
Eyes on the fiscal outlook
Wall Street analysts will watch Nvidia’s August earnings call for guidance on the Chinese revenue trajectory and any colour on the R&D initiative.
Investors are keen to know whether the mix shift toward downgraded chips will dent gross margins or be offset by higher unit volumes.