China's AI accelerator supplier Biren posts 2,000% year-over-year revenue growth — US export controls benefit homegrown chips as Nvidia and AMD exit market

Skyler Monroe (0:10)

Hey, everyone, welcome back to the AI Hardware Show. I'm your host Skyler Monroe, and if you're into chips, silicon, and the hardware that's literally powering the future, you're in exactly the right place. Huge thank you to our sponsors today. Ada, helping businesses integrate AI into their workflows seamlessly. Ago Consulting, that's Ago, your go-to for silicon development consulting from AI to SAW, and Zen Semiconductor building the processors and AI fabric behind modern data centers, with ventures like Sierra RISC-V CPUs and Loom AI Fabric, plus Ada for deploying AI at scale in any business. Alright, we have got a packed episode today. We're talking about a Chinese chip company posting some absolutely jaw-dropping revenue numbers, Nvidia wanting to put a mini data center on your living room wall, and TSMC scrambling to double a critical packaging capacity. Let's get into it.

Okay, first story, and honestly, this one hit me like a freight train when I read it. Biren Technology, a Chinese AI accelerator company, just posted 2000% year-over-year revenue growth in the first half of 2026 2000%, let that sink in for a second. That is not a typo. That is not a rounding error. That is an absolutely staggering number by any measure in any industry, let alone semiconductors. So who is Biren?

For those who haven't been tracking the Chinese chip space closely, Biren Technology is one of China's most ambitious homegrown AI accelerator companies.

They've been developing their own GPU-class chips designed to compete with the kind of hardware Nvidia and AMD were selling into China before the export controls kicked in. In that last part, export controls is really the key phrase here. Here's the context. The US government has been progressively tightening restrictions on what kind of advanced AI chips can be sold to China.

Nvidia's H100, the A100 gone from the Chinese market. They tried to make export control-compliant versions like the H800 and A800, but those got restricted too.

Eventually, even lower-performance chips like the H20 got caught in the net. AMD has faced similar restrictions. So essentially, the two dominant AI hardware companies on the planet have been pushed out of one of the world's largest and fastest growing technology markets. And nature, as they say, abhors a vacuum.

When Nvidia and AMD can't sell there, at least not officially, someone has to fill that gap. And Biren is raising its hand very enthusiastically. Think of it like this. Imagine the two biggest coffee chains in the world are suddenly banned from operating in a massive city. Every independent coffee shop in that city is about to have the best year of their life. That's essentially what's happening here. Now 2000% growth sounds incredible, and it is, but we need to add some nuance. Biren was starting from a relatively small revenue base, so percentage growth from a low baseline can look dramatic, even when the absolute numbers are more modest. That said, even accounting for the base effect, the trajectory here is clearly real and clearly significant. Shipments are reportedly up massively in the first half of the year. What's also interesting is how this fits into China's broader strategic push for semiconductor self-sufficiency. Biren isn't alone, you've got Cambricon, Hygon, MoreThreads and others, all competing in this space. But Biren seems to be emerging as one of the more capable players, particularly for large-scale AI training and inference workloads. Their BR100 chip has been getting attention for its architecture and memory bandwidths, which are critical for AI workloads. The implications here are massive for the industry. In the short term, this is a windfall for Chinese domestic chip companies. They're essentially getting a government protected market handed to them. But longer term, this is accelerating China's ability to develop genuinely competitive AI silicon. The more shipments they do, the more engineering feedback they get, the better their chips get. It's a flywheel. Export controls may be slowing China's access to the very best chips, but they're simultaneously turbocharged the development of China's own ecosystem. That's the paradox that policymakers are wrestling with. And for investors and hardware watchers, Biren is absolutely a company to keep your eye on. Alright, let's switch gears entirely, because story number two is one of those that made me do a genuine double take. The headline? Nvidia wants to put 16 Blackwell GPS and four server CPUs into a liquid-cooled, wall-mounted box in your home, and they want to offer it at $0 upfront cost. I read that three times.

Wall-mounted home, zero upfront. Let's unpack this. So Nvidia has been making moves into what's being called personal AI or AI PC territory for a while now. Project Digits was the first major signal, a compact, powerful AI computer designed for individuals and researchers. But this new concept takes things to a completely different level. We're talking about something more akin to a home server rack, just designed to mount on your wall like an oversized electrical panel or a modern TV mount. 16 Blackwell GPUs. Let's just appreciate that for a second. A single Blackwell GPU we're talking about Nvidia's current generation flagship architecture is an absolute beast. The GB200 NVL72 rack systems that hyperscalers are buying contain 72 of them and cost tens of millions of dollars. Now obviously a home targeted Blackwell would be a different lower tier variant but 16 of anything in that family is still enormously powerful. We're talking about Compute That just a few years ago would have been the exclusive domain of well-funded research labs and enterprise data centers. The liquid cooling angle is critical here and totally makes sense. 16 GPS generates serious heat. You cannot air cool that in a residential setting without turning your house into a turbine engine. Liquid cooling keeps it quiet, efficient and manageable. It's the same reason high-performance gaming PCs increasingly go liquid-cooled, just scaled up dramatically. Now, the $0 up-front cost model, that's where this gets really interesting from a business model perspective. This isn't Nvidia giving stuff away. The obvious play here is a subscription or compute as a service model. You get the hardware installed and you pay monthly, either for your own use of the compute, or potentially Nvidia uses your idle compute capacity as part of a distributed cloud network, offsetting your costs. Think of it like solar panels with a net metering agreement, but for AI, compute instead of electricity. This is actually a fascinating strategic move if you think about it. Nvidia has been so dominant in the data center that the next frontier is literally everywhere else– the edge, the home, personal devices.