China's 'Little Nvidias' Hit Profit. The Executive Bench to Scale Them Doesn't Exist

Moore Threads, the Shanghai GPU maker long dismissed as a state-subsidized long shot, reported first-half 2026 revenue of 1.74 billion yuan — about $257 million — a 147 percent leap from a year earlier. In August it filed for a Hong Kong listing, months after debuting on Shanghai's STAR Market. It is no longer an outlier. Across China's domestic AI-chip firms the story reads the same: years of subsidy have turned, with striking suddenness, into profit. The speed of the turn has caught even the firms themselves off guard, and investors have noticed — the Hong Kong filing values Moore Threads at a multiple that would have seemed fanciful two years ago. The problem nobody planned for now dominates boardroom talk. These companies have silicon, cash and customers. What they lack, by their own admission, is the executive bench to run a real business. A GPU does not run a company. Somebody has to.

The numbers form a coherent picture. Moore Threads' first-half revenue roughly tripled year over year, according to filings reported by BriefAsia and Nation Press. Rival Cambricon, the older AI-chip designer, grew revenue 108 percent in the same stretch, net profit climbing from 1.01 billion yuan in the first quarter to 1.30 billion in the second, per Nation Press. Around them sit MetaX, Iluvatar CoreX and Biren Technology — the cohort the trade press has taken to calling China's "little Nvidias." The trigger is well documented. Washington's export controls, tightened in successive rounds since 2022, cut Chinese buyers off from Nvidia's most advanced accelerators, and a captive market handed these firms their first real revenue. The controls arrived in waves — first in late 2022, then a sharper round in October 2023 that severed access to the top accelerators, then further limits on the downgraded chips Nvidia had designed for China. What none of those rounds could touch was the software problem: Nvidia's CUDA stack carries a decade of developer habit, and winning a customer means winning their developers first. The talent arithmetic is less cheerful. Deloitte projects a global shortfall of more than one million semiconductor workers by 2030. Industry estimates put China's own gap at 200,000 to 300,000 professionals, sharpest in chip design, electronic design automation and advanced packaging. The shortage that gets the loudest attention, though, is not the one that will decide the winner.

Look closer at Moore Threads. Its founder James Zhang spent years at Nvidia, and its Sudi and Chunxiao GPU lines were engineered to do three jobs at once — 3D graphics, AI training and video processing, with support for DirectX and OpenGL. That is the firm's quiet problem. A company built by engineers, for engineers, hits a wall the moment the work shifts from "can the chip be made" to "can it be sold." The shift is already visible. Cambricon, founded in 2016 and once a supplier to Huawei's Kirin phones, is the veteran of the group. Its second-quarter profit growth slowed to 90 percent year over year from 185 percent in the first, per China Fund News figures cited by TrendForce — not a stumble, but a sign the easy gains from pent-up domestic demand are flattening. Each member of the cohort faces the same cliff in a different disguise: MetaX in inference and rendering, Iluvatar in general-purpose GPUs, Biren in AI training. All of them spent a decade hiring for the cleanroom. None of them built the commercial layer above it. The software gap is the sharpest edge of the problem. CUDA did not win by being the only option; it won because a generation of developers built their work on it, and that community is as much a product as any chip. Building a rival community is an executive job, not an engineering one.

This is where Henderson Executive Search sees the gap most sharply. The firm's consultants, who have followed the semiconductor market through several cycles, note that China's AI-chip industry spent ten years stocking up on architects, designers and process engineers — the people who make silicon. It did not, over the same period, build a bench of the executives who turn silicon into a business: heads of go-to-market, developer-relations leaders who can nurture a software community the way CUDA did for Nvidia, international sales directors for export markets across Southeast Asia and the Middle East, and chief financial officers who can steer a dual listing. Add to that a supply chain leadership role that barely existed five years ago, because the entire stack — advanced packaging, high-bandwidth memory, leading-edge foundry capacity — now sits inside the blast radius of export controls. "The resumés in this space run deep on the technical side and thin everywhere else," a senior consultant at Henderson Executive Search observed. For a sector sprinting toward public markets, that imbalance is not a staffing wrinkle; it is the binding constraint on scale. Henderson Executive Search has fielded a wave of briefs from GPU firms for exactly these roles. Turns out, the qualified pools are, by the firm's measure, a fraction of what a comparable U.S. chip company would draw. One GPU maker's hunt for a head of go-to-market ran past nine months, according to Henderson Executive Search's consultants — longer than it took the same firm to tape out a new chip. The competition is cross-border by definition: the few executives who have actually scaled a semiconductor business are almost all already running something, frequently outside China. A single credible candidate can command several competing offers within a week, and compensation is no longer the obstacle — at the senior end, offers have climbed past what a comparable role pays in Silicon Valley once equity is counted.

To be fair, the profitability is not entirely a triumph of execution. A slice of it is a captive-market gift. With Nvidia largely shut out, Chinese cloud and enterprise buyers had nowhere else to spend, and prices held. That cushion is already thinning, and the moderation in Cambricon's growth is the first visible crease. There is also the retention problem: a firm that spends a year landing a scarce executive can lose them in a quarter to a rival waving a bigger equity package. What all of this points to is the next phase. The coming two to three years will be decided not in the cleanroom but in the meeting room — by whether these firms can hire and hold executives who have genuinely scaled a semiconductor business across borders. Henderson Executive Search's consultants put a finer point on it: the talent exists, but it is locked up in incumbent roles, and prying it loose takes credible equity, a real route to the public markets, and a search partner with reach on both sides of the Pacific. The firms that manage all three will separate from the pack. The ones that keep promoting engineers into commercial seats will watch the lead evaporate.

The pattern has played out before, and it ends the same way each time. The companies that treat the leadership bench as seriously as the product roadmap are the ones that survive the turn from subsidized champion to commercial operator. For China's little Nvidias, the silicon is no longer the hard part. Henderson Executive Search's consultants put the situation plainly: the scarcest component in the entire stack is no longer the chip. It is the person in the corner office, and the next twelve months will tell which firms understood that in time.

FAQ

Q: What roles do profitable AI-chip firms actually need next?

A: The commercial layer above the silicon. Heads of go-to-market, developer-relations leaders who can nurture a software community the way the incumbent's stack did, international sales directors for markets across Southeast Asia and the Middle East, dual-listing chief financial officers, and supply chain leadership now that advanced packaging, high-bandwidth memory and leading-edge foundry capacity all sit inside the blast radius of export controls. The industry spent a decade hiring for the cleanroom and never built this bench.

Q: Why is developer relations the hardest seat to fill?

A: Because the incumbent accelerator vendor did not win on hardware alone. A decade of developer habit built around its software stack makes that community as much a product as any chip, and building a rival community is an executive job rather than an engineering one. Firms that spent ten years staffing cleanrooms have no candidate pool for it, and the qualified pool is a fraction of what a comparable U.S. chip company draws.

Q: Why do these searches run past nine months, and who are they competing with?

A: The few executives who have genuinely scaled a semiconductor business are almost all already running something, frequently outside China, so the competition is cross-border by definition. One search for a head of go-to-market ran past nine months, longer than the same firm took to tape out a new chip. A single credible candidate can draw several competing offers within a week, and at the senior end, packages including equity have climbed past comparable pay in the U.S. market. Retention is the mirror problem: a firm can spend a year landing an executive and lose them in a quarter.

Q: How durable is the profitability that created this hiring wave?

A: Partly a captive-market gift. With the leading accelerator vendor largely shut out by export controls, Chinese cloud and enterprise buyers had nowhere else to spend and prices held. That cushion is already thinning, and the moderation in profit growth at the cohort's most established designer is the first visible crease. The next two to three years will be decided less in the cleanroom than in the meeting room, by whether these firms can hire and hold executives who have scaled a semiconductor business across borders.