EVAS Intelligence, a Chinese chip startup, raised $295 million on September 18 in a funding round that values the company at $2.21 billion, betting on the RISC-V instruction set as a way to build data center AI chips that US export controls cannot directly restrict. The round adds to roughly $309 million the company had raised previously and follows a $211 million Series B in April that funded its Epoch series chips, already in production for large-model training and inference. The new capital is earmarked for next-generation chip development, software ecosystem expansion and international growth.
The choice of RISC-V is the strategic point. RISC-V is an open instruction set governed by RISC-V International, a Swiss nonprofit that relocated from the United States to Switzerland in 2019, a move widely read as reducing exposure to American trade regulation. The architecture can be used royalty-free by anyone, and Washington cannot impose export controls on it the way it restricts Nvidia’s GPU shipments. US lawmakers proposed in November 2023 to require export licenses for US persons engaging with China on RISC-V, but that restriction has never been enacted. The instruction set remains outside the primary US enforcement mechanism, which is the whole reason a chip startup would organize itself around it.
EVAS pairs RISC-V vector extensions with a proprietary AI computing architecture called EVAMIND, a domain-specific design the company describes as TPU-like. Its Epoch series is China’s first RISC-V AI compute chip with native FP8 block-wise quantization, released in April and in mass production. The company, founded in 2022 and headquartered in Guangzhou with a new base in Beijing, counts China Mobile’s Chain Development Fund among its strategic investors and reports commercial orders exceeding 10 billion yuan with strategic customers. Its ELYNX automotive chip line, launched in December 2023, gave it revenue before the data center push.
The harder question is software
Whether the money buys competitiveness is a different matter from whether it buys chips. RISC-V’s share of new silicon designs globally reached an estimated 25 percent by January 2026, but nearly all of that was IoT devices, embedded systems and edge AI. Scaling an open instruction set into data center AI training clusters is a qualitatively harder problem, where ecosystem lock-in, compiler maturity and operator library completeness decide real-world throughput. Customers who currently default to Nvidia, or to Huawei’s Ascend for domestic Chinese buyers, do not switch on instruction set politics alone. CUDA’s two decades of accumulated software remains the moat, and Huawei has spent years trying to build its equivalent with limited traction outside government-backed projects.
The funding lands amid a familiar pattern in Chinese AI infrastructure. Alibaba supplied roughly 20,000 Nvidia H200 processors to Moonshot AI through a cloud deal despite the US export ban, a Bloomberg report confirmed earlier this year, showing that Western hardware still finds its way to Chinese labs through rental and cloud intermediaries. The White House Office of Science and Technology Policy accused Moonshot of acquiring Blackwell chips through a rental path in Thailand and of distilling an Anthropic model to train its Kimi K3. Against that backdrop, a domestic chip company that needs no workaround at all has obvious appeal to Chinese buyers who cannot rely on gray channels and cannot afford to be the next enforcement headline.
EVAS’s trajectory tracks the broader state-backed push. The company raised nearly 200 million yuan in angel funding in 2022, completed a 1.5 billion yuan Series B in April 2026, and moved its headquarters into Beijing’s National Information Technology Application Innovation Park in June. Its investor list is dense with Beijing municipal funds, including the Beijing AI Industry Investment Fund and the Beijing Economic-Technological Development Area Industrial Upgrade Fund. That profile, private capital wrapped around state money, is now the standard structure for Chinese chip ventures aiming at strategic sectors, and it signals that Beijing intends EVAS to survive the long software build-out regardless of near-term sales.
For the US export control regime, the development is uncomfortable but not yet decisive. Controls have never restricted RISC-V itself, only American chipmaking tools, advanced GPUs and US persons’ involvement in certain Chinese semiconductor projects. A Chinese firm designing competitive AI accelerators on an open, Swiss-governed instruction set, fabricated presumably on domestic or non-US process nodes, would sit outside every major control lever. The question Washington faces is whether that changes its calculus on restricting RISC-V collaboration, something it has considered and declined to do three years running. Each new round of funding for a RISC-V data center chip company makes that answer harder to defer.
The competitive frame matters too. EVAS is not the only Chinese RISC-V bet, but it is among the few with a produced training-class chip rather than a roadmap. Huawei’s Ascend line remains the domestic incumbent, protected by procurement preferences and a larger software investment. EVAS’s pitch to Chinese cloud providers is efficiency and total cost of ownership on modern AI workloads, a claim that will be tested in deployments rather than benchmarks. The company says its architecture natively supports large language model workloads through a streaming parallelism design, which, if the numbers hold in production clusters, would distinguish it from retrofit GPU designs.
The coming 12 to 24 months will show whether EVAS can win training workloads or remains confined to inference and edge deployments. The company has capital, state backing and a production chip. What it does not have, yet, is proof that anyone outside a captive domestic customer base will run large models on its silicon, and that proof is the only thing that turns a $2.21 billion valuation into a business.