China's national chip champions timed a wave of product launches to land days before Xi Jinping's state visit to Washington, and the stack details matter more than the diplomatic optics: Huawei pulled forward its next AI chip by three quarters, Alibaba shipped a processor it calls the country's most powerful, and DeepSeek reportedly assembled the largest known cluster of Huawei chips built to date, according to CNBC's report on the summit run-up. For architects scoping GPU or accelerator procurement across US-China supply chains, the timing is a signal, not a spec sheet — the underlying compute gap has not closed.
The two flagship parts are named and dated. Huawei said at its annual Huawei Connect conference last week that its Ascend 960DT will ship in the first quarter of 2027, three quarters ahead of its original schedule, CNBC reported. At Alibaba's Apsara Conference in Hangzhou on Tuesday, the company's chip unit unveiled the Zhenwu V900, which the report says triples the performance of its predecessor and goes on sale in early 2027. Sigrid Wang, an analyst at China-focused Hutong Research, told CNBC "there does seem to be genuine acceleration in its production timeline" on the Huawei part.
The architecture behind these launches is a workaround, not a breakthrough. Chinese firms have compensated for weaker individual chips by linking thousands of them into large clusters over faster interconnects — the same clustering logic DeepSeek used at scale, per the CNBC report. But Huawei's latest SuperPoD computing architecture links far fewer processors than originally planned, and each chip still delivers roughly half the compute of an Nvidia chip, according to the report. That per-chip deficit is the number that should anchor any procurement model: doubling cluster size to match a single Nvidia part's throughput changes the power, cooling and interconnect budget for a deployment, even before yield and availability are factored in.
Supply is the harder constraint than performance. Huawei Chairman Eric Xu acknowledged, as reported by CNBC, that Huawei may not be able to make enough chips to meet domestic demand. Xiaomeng Lu, a director at Eurasia Group, told CNBC "Chinese chip companies haven't made any groundbreaking progress in the past few years" and that "the fundamental competitive landscape hasn't changed." Chris Miller, a Tufts University professor and author of "Chip War," told CNBC that Chinese firms remain substantially short of the chips needed to train and deploy advanced models and rely heavily on smuggling chips or accessing data centers outside China — "China is still highly dependent on the U.S. for advanced chips," he said.
The revenue gap tells the same story from the software side. Miller told CNBC that American firms like Anthropic and OpenAI make 50 to 100 times the revenue of comparable Chinese firms, and that US models remain ahead on technical benchmarks and profitability. George Chen, partner and chair of digital practice at The Asia Group, characterized the launch timing to CNBC as a "deliberate" confidence move ahead of the Trump-Xi summit, framing it as Xi having "many cards" to bring to the table rather than technological parity.
On the policy side, the trade that matters to compliance teams did not happen. Treasury Secretary Scott Bessent and US Trade Representative Jamieson Greer met Vice Premier He Lifeng in New York over the weekend and set up a formal AI channel for incident warnings, but Greer said explicitly that export controls on advanced chips and chipmaking equipment were not on the summit agenda, per CNBC. Lizzi Lee, a fellow at the Asia Society Policy Institute, told CNBC that a "recalibration of U.S.-China tech relations takes a lot of substantial work" and is "not a summit deliverable... at least not this time." That means the current control regime on Nvidia's most advanced chips holds for now, with no announced loosening or tightening to plan around.
The unresolved part for anyone building a multi-region GPU roadmap is that the gap and the workaround are both real at once: China is not closing the per-chip performance deficit, but it is proving a cluster-based path to usable AI compute exists without those chips, and that path scales with domestic chip volume China itself admits it doesn't have enough of yet, per Huawei's own chairman.
The takeaway for procurement planning: treat this week's launches as evidence that export controls slow China's roadmap rather than block it, and keep sourcing decisions anchored to the documented compute-per-chip gap and the unchanged control regime, not to press-cycle timing around a summit.