Biren Technology’s own filing did the bragging for it. The Shanghai GPU maker told the Hong Kong exchange that revenue for the first six months of 2026 would land somewhere around 1.15 billion yuan, which works out to nearly 2,000% growth year over year. No spin required. Just a number big enough that everyone in the industry stopped scrolling.
My first reaction, as someone who spends most days poking holes in AI product claims: 2,000% of what, exactly?
Percentages are the most dishonest honest number in tech
Nothing in that figure is fabricated. Biren really did grow that fast. But a 2,000% jump tells you almost nothing about a company’s capability and almost everything about where it started. If you sold 20 units last year and 400 this year, congratulations, you posted the same growth rate as a company going from 20,000 to 400,000. One of those is a supply chain. The other is a pilot program that went well.
Biren is small. That’s the part the headline number quietly depends on. It listed in Hong Kong in January 2026, and the H1 revenue estimate sits in the low billions of yuan, not the tens of billions of dollars. That’s a real business now, not a demo, but it’s not yet a company that keeps Jensen Huang up at night on the merits of its silicon.
The analyst math deserves a second look
Here’s where I’d push back on the coverage. Financial analysts are projecting operating revenue of $4.09 billion in 2026, $13.39 billion in 2027, and $27.38 billion in 2028, with growth rates of 165%, 228%, and 104%. Stack those against the company’s own H1 2026 estimate of roughly 1.15 billion yuan and the numbers don’t sit comfortably together. Either the second half of 2026 is expected to be wildly, historically larger than the first, or those projections are measuring something different from what the filing describes.
I’m not calling anyone a liar. I’m saying that when a forecast implies a 22-fold surge, the burden of proof sits with the forecast. Projections that far out are marketing documents with footnotes. Treat them accordingly.
What’s actually driving this
The demand is real, and it isn’t primarily about Biren winning on engineering. It’s about Nvidia and AMD walking out of the room. US export controls pushed the top-end Western parts out of reach, and Chinese buyers who still need accelerators are buying whatever they can legally get delivered. DIGITIMES estimates Chinese vendors will ship 2.123 million high-end cloud AI accelerators in 2026, a 136% year-on-year increase. That’s a market being rebuilt from the supply side up.
Policy created a captive customer base. Domestic vendors are filling it. That’s the whole mechanism, and it’s a little funny in a bleak way: the controls meant to slow China’s AI buildout handed its chip startups the one thing startups almost never get, which is a protected market with urgent, well-funded buyers and no incumbent to undercut them.
And Biren isn’t even the scary one
If you’re looking for the company that should worry Western vendors, look at Huawei. In March 2026 it announced the Atlas 350, a single-card part it claims delivers 1.56 quadrillion calculations per second, roughly three times the performance of Nvidia’s best China-legal chip. Huawei also has system integration muscle that Biren does not. Biren’s 2,000% is the flashier number. Huawei is the more serious threat.
What I still can’t tell you
This is the part that matters for anyone actually choosing hardware, and it’s the part nobody’s answering. I have no independent benchmarks for Biren’s accelerators. I have no data on real-world utilization, thermal behavior under sustained training loads, or failure rates at scale. Most importantly, I have nothing solid on the software stack, and software is where accelerator companies go to die. CUDA’s moat was never really the transistors.
Revenue growth measures how many chips shipped. It does not measure whether developers can get anything useful out of them without a support engineer on speed dial. Those two things diverge constantly in this business, and the gap is where buyers get burned.
How I’d read this if I were you
- The growth is real but the base is small. Don’t read 2,000% as a capability claim.
- The demand is policy-made. Export controls did more for Biren’s order book than any product launch could have.
- The 2027 and 2028 forecasts are unproven. They imply a step change that the current filing doesn’t support on its own.
- Watch software, not silicon. Whoever builds a usable toolchain wins the Chinese accelerator market, regardless of who ships the most cards in 2026.
Biren had an excellent year. It earned the headline. But a growth rate is a story about the past, and this particular past was written mostly in Washington.
🕒 Published: