MiniMax Co-Founder Reveals 280% Revenue Growth on Slower R&D Spending: What It Means for the AI Infrastructure Trade
A Chinese AI lab nobody on Wall Street watches just reported revenue growing at more than twice the pace of its research spending, and the implications land squarely on the three American chipmakers holding up a multi-trillion dollar buildout bet.
This post may contain links from our sponsors and affiliates, and Flywheel Publishing may receive compensation for actions taken through them.
A private Chinese AI lab that most American investors have never heard of stood up at a Goldman Sachs conference on Monday and told a story that should worry anyone holding the AI infrastructure trade. Yeyi Yun, co-founder of MiniMax, said first-half revenue grew more than 280% and had already reached 1.5 times the prior full year, with July token consumption running 20 times the January level.
She said the company is targeting $1 billion in full-year revenue while R&D spending grows at only 130%. MiniMax is privately held and unavailable to US investors, but the claim still matters as evidence, because the entire American AI trade rests on the assumption that frontier capability requires ever-larger training clusters.
If a lab can grow revenue at more than twice the pace of its research spending while giving away frontier weights, that assumption deserves scrutiny. The names underwriting the assumption are NVIDIA (NASDAQ:NVDA | NVDA Price Prediction), Advanced Micro Devices (NASDAQ:AMD), and Broadcom (NASDAQ:AVGO).
What MiniMax Actually Claimed
Yun framed the strategy in plain terms. “The key of our company is any other name minimax, minimizing the cost and maximizing the intelligence,” she said, adding that cost efficiency and intelligence have “more synergy” rather than a tradeoff.
She also said, “Revenue growth is much faster than our R&D spending,” and pointed to the H suite open-source model, which she said had been downloaded more than 24 million times on Hugging Face with more than 300 derivatives.
These are self-reported figures from a private company at a bank conference. Download counts measure curiosity rather than paid usage, and 280% growth from a small base is different from 280% growth on billions. Take the numbers as directional rather than audited.
But the underlying pattern, cheaper models trained with less compute achieving comparable output, has been visible since DeepSeek and is now showing up in monetizable revenue. That is the part American holders of AI capex-linked equities cannot dismiss.
Capex Bets Underneath the American Chipmakers
NVIDIA is priced for the buildout continuing. On the Q2 FY2027 call, management said cloud industry backlog exceeded $2 trillion and that the top five hyperscalers were expected to spend nearly $800 billion on capex in 2026 and $1.3 trillion in 2027. Numbers like that draw in a much wider circle of suppliers than the chipmakers do, which is why we pulled together seven of the power, cooling, and networking names riding the same buildout in a free report you can grab here.
The company reported Q2 revenue of $96 billion and Data Center revenue of $89 billion. It also disclosed $279 billion in supply obligations, largely tied to Vera Rubin memory. That is a forward commitment that only pays off if hyperscaler orders keep landing.
Broadcom is running the same trade through custom silicon. Q2 AI semiconductor revenue was $10.8 billion, up 143% year-over-year, and Hock Tan guided fiscal 2027 AI revenue to in excess of $100 billion. AMD reported Q2 revenue of $11.54 billion with Data Center up 107% to $6.718 billion, and Lisa Su has committed the company to gigawatt-scale Helios deployments with Anthropic, Meta, and Microsoft.
All three theses depend on compute demand outrunning efficiency gains. If MiniMax and peers make each GPU-hour more productive faster than workloads scale, the volume math breaks.
Why Jensen’s Rebuttal Actually Holds Some Water
Huang addressed this directly on the call. He argued that as inference becomes cheaper, usage will explode, and agentic workloads will run continuously. His framing: “If we had more compute, we could generate more profitable tokens, which results in more profit for all of the services.”
He also said frontier labs are “only limited by the amount of compute” and that NVIDIA’s platform supports both open and closed models. MiniMax’s H suite, notably, almost certainly trains and serves on NVIDIA silicon.
The Jevons argument is real. Cheaper unit costs have historically expanded total consumption in computing. But it requires that the new workloads materialize on schedule, and that hyperscaler capex, currently financed by operating cash flow and increasingly by third-party structured capital, holds up if returns disappoint.
NVIDIA itself is now backstopping some of that demand. Management said it has invested nearly $50 billion in frontier labs and organized partners to mobilize over $500 billion of third-party capital. That backstopping suggests management sees the need to underwrite demand it once expected to arrive on its own.
Is NVDA Stock a Buy?
NVDA trades at $217.44, up 25% over one year, with a market cap of $5.25 trillion. AMD is up 182.61% over one year and trades at a much richer multiple on still-scaling AI margins. AVGO carries steadier hyperscaler contracts and a fatter free cash flow yield.
Of the three, NVIDIA is the most exposed to the efficiency question because its revenue-per-gigawatt thesis assumes that each new architecture justifies a full replacement cycle. It is also the best hedged, because CUDA runs whatever wins, including MiniMax-style efficient models.
The risk/reward looks balanced here. The business is exceptional, and the buildout is real, but $279 billion in supply commitments against a capex cycle that a Chinese lab just credibly challenged is not the setup for adding at all-time highs.
Contact [email protected] for any questions or corrections.
Omor Ibne Ehsan is a writer at 24/7 Wall St. He is a self-taught investor with a focus on growth and cyclical stocks that have strong fundamentals, value, and long-term potential. He also has an interest in high-risk, high-reward investments such as cryptocurrencies and penny stocks.
