Beijing-based Moore Threads spent 50 percent of its revenue on research and development in the first quarter of 2026. Its Shanghai rival MetaX wasn't far behind at 45 percent. By comparison, American semiconductor giants AMD and Intel typically dedicate just 20 to 30 percent of their sales to R&D. Even Nvidia — the undisputed king of AI chips — saw its R&D ratio plummet to a mere 8.6 percent last year as surging demand inflated its top line beyond $215 billion.
The numbers, revealed in fresh exchange filings and reported by the South China Morning Post on May 6, paint a stark picture of two radically different innovation strategies. While U.S. chipmakers are riding a wave of AI-fueled profitability, China's domestic GPU champions are burning cash at a rate that would make Silicon Valley venture capitalists wince — and they're doing it because they have no other choice.
The 50 Percent Club: Moore Threads and MetaX
Moore Threads and MetaX are not household names outside of China. But inside the world's second-largest economy, they represent the vanguard of a national mission that Beijing has made unmistakably clear: achieve technological self-reliance in semiconductors, or risk permanent dependence on foreign suppliers.
Moore Threads, headquartered in Beijing and founded in 2020 by former Nvidia vice president James Zhang Jianzhong, has emerged as China's most credible challenger to Nvidia's GPU dominance. Zhang spent 14 years at Nvidia, rising to become general manager of the company's China operations, before leaving to build what he now claims is the only Chinese firm currently mass-producing fully functional GPUs.
The company's financial trajectory tells a story of explosive growth fueled by relentless investment. Moore Threads' revenue surged 243 percent year-over-year to reach 1.5 billion yuan (approximately $208 million) in 2025. Its gross margin — a critical indicator of manufacturing maturity — swung from a devastating negative 70 percent in 2022 to a healthy 70.71 percent by 2024, according to data compiled by industry analyst TrendForce. That margin recovery is the kind of improvement that typically takes legacy semiconductor firms a decade to achieve.
But the headline figure is the R&D burn. Over the past three years and three quarters, Moore Threads has poured 4.67 billion yuan into research and development — roughly three times its cumulative revenue during the same period. In Q1 2026 alone, half of every dollar coming through the door went straight back into engineering labs, chip design software, and fabrication partnerships.
MetaX, based in Shanghai and founded by former AMD senior director Chen Weiliang, is running a similarly aggressive playbook. The company's revenue exploded from a modest 53 million yuan in 2023 to 743 million yuan in 2024 — a 1,301 percent year-over-year leap — before climbing another 121 percent to 1.644 billion yuan in 2025. MetaX has accumulated 2.943 billion yuan in R&D spending over the past three years and three quarters, roughly 1.5 times its revenue.
MetaX's product focus is narrower but strategically important: training-and-inference GPU boards and servers. By 2024, GPU boards accounted for nearly 69 percent of its revenue, with GPU servers contributing another 28 percent. In the first quarter of 2025, GPU board revenue alone surged to 97.55 percent of total sales — a sign that the company has found product-market fit in the AI infrastructure space.
The American Contrast: Profitability Over Investment
While China's GPU upstarts are hemorrhaging cash to close the technology gap, their American counterparts are enjoying a golden age of profitability that has actually reduced their R&D intensity.
Nvidia, the $3.4 trillion titan whose GPUs power everything from ChatGPT to autonomous vehicles, spent $18.5 billion on R&D in the year ended January 25, 2026 — an enormous absolute figure that nonetheless represented only 8.6 percent of its revenue. That's down from 27.2 percent in 2022, before the generative AI boom turned the company's financial profile upside down. Revenue surged to $215.9 billion, making Nvidia's R&D ratio look almost modest by comparison.
AMD spent approximately $8 billion on R&D for the year ended December 27, 2025, while Intel invested $13.8 billion. Both firms typically allocate 20 to 30 percent of revenue to research — healthy by most industrial standards, but modest compared to the Chinese startups' all-in approach.
The absolute spending gap remains enormous. Nvidia's annual R&D budget alone exceeds the combined cumulative R&D spending of Moore Threads and MetaX over their entire corporate lifetimes. Chinese chip designers, despite their ferocious intensity, still lag far behind on the total dollars available for hiring top talent, licensing design tools, and securing early access to advanced manufacturing processes.
Why the Desperation Makes Strategic Sense
The paradox of China's chip R&D frenzy is that it is both economically irrational and geopolitically inevitable. No venture-backed startup in Silicon Valley could sustain spending 50 percent of revenue on R&D for multiple years without triggering a boardroom revolt. But Moore Threads and MetaX are not operating in a normal market — they are arms suppliers in a technological cold war.
U.S. export controls, tightened progressively since 2022, have choked China's access to Nvidia's most advanced AI chips. The restrictions have been so effective that Nvidia's market share in China collapsed from 95 percent to roughly 50 percent over four years, according to Reuters. Nvidia CEO Jensen Huang acknowledged in May 2025 that the controls were "a failure" from a policy perspective — not because they didn't hurt China, but because they created a domestic replacement industry that Washington may not be able to stop.
Huang also offered a sobering assessment of China's progress, noting that the country is merely "nanoseconds behind" the United States in chipmaking capability. That gap, while real, is narrowing faster than many American policymakers expected.
The domestic opportunity is enormous. China's AI market — spanning everything from large language models to autonomous driving to smart city infrastructure — requires GPUs by the millions. With Nvidia increasingly restricted, the addressable market for homegrown alternatives has expanded dramatically. Moore Threads' MTTS5000, its flagship training-and-inference GPU, has already entered mass production and is reportedly being deployed in clusters capable of supporting trillion-parameter model training at efficiency levels comparable to advanced GPU clusters of similar scale.
The IPO Race: Capitalizing on National Ambition
Both Moore Threads and MetaX are racing toward public markets to fund their R&D addiction. Moore Threads received approval for its IPO on Shanghai's STAR Market in just 88 days from filing — one of the fastest approvals on the Nasdaq-style board — and aims to raise 8 billion yuan (approximately $1.12 billion). MetaX is pursuing its own listing, though timing remains undisclosed.
The speed of Moore Threads' regulatory approval is itself a signal. China's securities regulators appear to be treating strategic technology firms as national priorities, fast-tracking listings that align with Beijing's "tech self-reliance" agenda. The STAR Market, created in 2019 specifically to channel capital toward hard-tech innovation, is increasingly becoming the financing engine for China's semiconductor counter-offensive.
Meanwhile, Cambricon — the third member of China's emerging GPU trio — has already achieved what Moore Threads and MetaX are still chasing: profitability. The Beijing-based AI chip designer reported revenue of 6.497 billion yuan in 2025, up 453 percent year-over-year, with net profit hitting 2.059 billion yuan — a dramatic turnaround from a 450 million yuan loss the previous year. Cambricon raised another 3.985 billion yuan through a private placement last year, with proceeds earmarked for next-generation large-model chip platforms.
However, Cambricon faces supply chain headwinds. Its next-generation Siyuan 690 chip, originally expected to enter mass production by late 2025, is now rumored to be delayed until the second half of 2026 due to manufacturing constraints — a reminder that China's chip ambitions remain bottlenecked by access to advanced fabrication.
The Talent Pipeline: NVIDIA and AMD Veterans Building China's Future
One of the most fascinating dimensions of China's GPU push is the human capital story. Moore Threads founder James Zhang spent 14 years inside Nvidia before decamping to build its Chinese competitor. MetaX founder Chen Weiliang joined AMD Shanghai as a senior director in 2007 and spent 13 years there before launching his rival venture in 2020. Both of MetaX's CTOs, Peng Li and Yang Jian, are also AMD alumni.
This pattern — Western-trained semiconductor veterans returning to China to build domestic champions — mirrors the talent flows that previously powered China's smartphone, solar panel, and electric vehicle industries. In each case, years of working inside global leaders provided the technical know-how, industry relationships, and strategic vision needed to launch credible domestic alternatives.
The risk for American competitiveness is not that China will match Nvidia's performance tomorrow — it won't. The risk is that China is systematically building the institutional knowledge, supply chain relationships, and engineering talent base that will make it increasingly difficult to maintain a permanent technological edge. Every year that Moore Threads and MetaX spend 45 to 50 percent of revenue on R&D, they are compounding capabilities that become harder to suppress with export controls alone.
What This Means for the Global Chip War
The R&D intensity gap between Chinese and American chipmakers is not a temporary aberration — it is a structural feature of a bifurcating global semiconductor market. U.S. firms are optimizing for profitability and shareholder returns in a mature market they dominate. Chinese firms are optimizing for survival and strategic independence in a market where foreign supply is being deliberately choked.
For investors and technology observers, the dynamics create a complex picture. Chinese GPU stocks are likely to remain volatile and loss-making for years, sustained by national policy support and the sheer scale of domestic demand. American chip giants, meanwhile, face the long-term risk of a shrinking China market being replaced by homegrown competitors that are improving faster than Western analysts expected.
The 50 percent R&D ratio is not sustainable in a conventional business sense. But conventional business logic does not apply when a sector has been elevated to national security priority status. Moore Threads and MetaX are not trying to maximize quarterly earnings — they are trying to ensure that China never again faces a semiconductor chokehold.
Whether they succeed will determine not just the future of AI computing, but the balance of technological power between the world's two largest economies for the next decade.