Biren Technology, one of a handful of chipmakers trying to build a GPU business outside the Nvidia-AMD duopoly, posted first-half 2026 revenue of $183.9 million, up roughly 1,998% year over year from about $8.665 million in the same period of 2025, according to Tom’s Hardware. The Shanghai-based company also booked a $56.2 million net loss and a 42.7% gross margin, a split that captures exactly what it costs right now to chase a market Nvidia and AMD still dominate. The numbers landed just as Huawei, Intel, Qualcomm, MediaTek, and Imagination Technologies all pushed out their own GPU and accelerator news within the same few weeks, turning “who else makes GPUs” from a trivia question into a live earnings story.
None of these companies are close to displacing Nvidia or AMD in raw volume. A broader look at the non-Nvidia, non-AMD GPU landscape published on this site earlier in 2026 found the market share gap still sits above 90%. But the pattern across Biren’s earnings, Huawei’s Atlas 960 SuperPoD launch, Intel’s Crescent Island roadmap, and fresh silicon from Qualcomm and MediaTek shows a second tier of GPU and accelerator makers scaling faster than at any point in the last several years, largely because export restrictions and AI compute shortages are forcing customers, especially in China, to buy from whoever can actually ship.
Biren Technology’s Revenue Jumped Nearly 20x. The Loss Grew With It
Biren’s first-half 2026 results, reported by Tom’s Hardware, are the clearest data point yet on how fast China’s homegrown AI chip sector is scaling without Nvidia or AMD parts. Revenue hit $183.9 million against roughly $8.665 million a year earlier, a jump of about 1,998%. Gross profit came in at $78.552 million, giving the company a 42.7% gross margin, which is respectable for a chipmaker still building out its product line. The catch is the bottom line: a $56.2 million net loss, driven by continued spending on new accelerators, software stacks, and optical-interconnect rack-scale systems.
That combination, fast revenue growth paired with a widening loss, is the standard shape of a chip company trying to buy market share with R&D spend before the business model catches up. Biren’s AI-accelerator share of the Chinese market remains below 3%, according to TrendForce data cited by Tom’s Hardware, which puts the growth in context: the company is scaling from a very small base, not challenging incumbents at scale.
What Biren Actually Sells
Biren’s current shipping lineup includes the BR106, BR110, and BR166 GPUs, aimed at AI training and inference workloads. The company is also developing a next wave of accelerators under the BR20X, BR30X, and BR31X names, according to Tom’s Hardware’s reporting. That roadmap mirrors the pattern set by Nvidia and AMD: a current-generation product line shipping today, and a named future generation already in development to keep customers committed to the platform rather than shopping elsewhere.
Biren went public on the Hong Kong Stock Exchange in January 2026. At the listing ceremony, founder, chairman and CEO Zhang Wen said, “Starting today, Biren Technology will become a public company and will also shoulder greater social responsibility,” according to a report carried by RTHK. He added that the company would “invest even more resolutely in technological innovation and will do a better job in managing the business, creating long-term value for shareholders, employees and society,” a line that, nine months later, lines up with where the first-half results landed: heavy investment, growing revenue, and a loss to match.
Zhang also framed the listing as a vote of confidence in Hong Kong as a base for hard-tech startups, saying, “As one of the first key high-tech firms brought into Hong Kong, we deeply feel the city’s emphasis on and support for hard-tech startups,” and that “Hong Kong’s excellent business environment and its efficient link to global capital will become an important platform for Biren Technology and other companies to develop in Hong Kong,” per the same RTHK report. He described the IPO itself as a transition point rather than a finish line, calling it “not the end but a brand-new beginning.” Read together with the first-half numbers, those comments track: a company treating a loss-making stretch as the price of building out a platform, not a warning sign to slow down.
Why Export Controls Keep Creating New GPU Suppliers
The reason Biren, Huawei, and a cluster of smaller Chinese accelerator makers exist as a GPU category at all traces back to US export restrictions that have limited the sale of Nvidia’s and AMD’s most advanced AI chips into China. That policy created a gap that domestic suppliers are racing to fill, not because their technology has caught up to Nvidia’s, but because Chinese AI labs and cloud operators need accelerators and can’t always get the ones they’d otherwise buy.
This is also why the growth numbers look the way they do. A 1,998% year-over-year jump sounds like a company eating into Nvidia’s business. In Biren’s case it’s closer to a company going from nearly nothing to a small, real business because the alternative, buying restricted Nvidia or AMD silicon, isn’t available to Chinese buyers at the volume they need. The same dynamic is shaping Huawei’s accelerator push and the broader interest in domestic GPU IP across Chinese chip design houses, and it echoes the supply pressure already visible in Micron’s decision to wind down 2GB GDDR7 production, which is squeezing memory supply for RTX 50-series cards built around Nvidia and AMD silicon.
Huawei’s Atlas 960 SuperPoD Targets the Same Gap
Huawei used its Connect 2026 conference in Shanghai on September 17 to unveil the Atlas 960 SuperPoD, a large-scale AI computing system positioned as a domestic alternative for customers who can’t access the newest Nvidia or AMD hardware. Huawei also laid out a roadmap for two new Ascend processors: the Ascend 960DT training chip, targeted for the first quarter of 2027, and the Ascend 960PR inference chip, scheduled for the third quarter of 2027.
Separately, reporting has pointed to DeepSeek deploying at least 160,000 Ascend 950DT accelerators in Inner Mongolia, though that figure should be treated as a reported deployment number rather than an independently verified shipment total. Huawei’s Ascend line and Biren’s BR-series chips are effectively competing for the same customer base: Chinese AI companies that need training and inference capacity and can’t rely on an uninterrupted supply of Nvidia or AMD GPUs.
Intel Is Still Building a Data Center GPU Business Too
Intel’s path back into the GPU conversation runs through Crescent Island, a 160GB accelerator built for inference workloads in air-cooled enterprise servers. The chip is built on Intel’s Xe3P architecture and uses LPDDR5X memory, and Intel has said it plans to sample the part with customers in the second half of 2026, according to CRN.
Intel CEO Lip-Bu Tan addressed the company’s accelerator ambitions directly during a recent earnings discussion, saying “the accelerator remains central to frontier AI, and we will continue to participate, innovate and partner in that category,” as reported by CRN. The comment reads as a direct response to skepticism that Intel can compete in standalone AI accelerators after years of GPU roadmap delays and canceled projects, including the kind of scrutiny Intel’s 14A manufacturing node is facing from chip giants still evaluating it. Crescent Island is explicitly scoped for inference rather than frontier model training, which puts it in a different lane than Nvidia’s flagship training GPUs and closer to where Biren and Huawei are also positioning their newest chips.
Mobile GPU IP: Qualcomm, MediaTek, and Imagination Move Too
The GPU conversation outside Nvidia and AMD isn’t limited to AI data center silicon. Qualcomm used its 2026 Snapdragon Summit, held September 22 to 24, to introduce the Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6 mobile platforms, both of which include new integrated GPU designs for flagship phones launching later this year and into 2027.
MediaTek unveiled the Dimensity 9600 Pro on September 15, which the company describes as a major architectural revision of its flagship mobile chip, including GPU and AI engine updates aimed at closing the gap with Qualcomm’s top-tier silicon, per MediaTek’s own product materials.
Imagination Technologies, the UK-based GPU IP licensor whose designs power chips from multiple Android vendors, published initial performance data for its E-Series GPU architecture and introduced a feature called Neural Super Resolution, which combines graphics rendering, general compute, and AI upscaling in one programmable design, according to Semiconductor Engineering. Imagination doesn’t sell finished chips the way Biren or Huawei do; it licenses GPU designs to phone and automotive chipmakers, which makes it a different kind of “GPU maker” but a real one, since its IP ends up inside a meaningful share of non-Qualcomm, non-Apple mobile silicon.
How the Non-Nvidia, Non-AMD GPU Field Stacks Up
Putting these companies side by side shows how differently each one is approaching the GPU market. Some are chasing AI data center share, others are focused entirely on mobile silicon, and their business models range from selling finished chips to licensing IP blocks to other manufacturers.
| Company | Primary GPU/Accelerator Focus | Latest Product | Key Recent Data Point | Source |
|---|---|---|---|---|
| Biren Technology | AI training and inference (China) | BR106, BR110, BR166 | $183.9M H1 2026 revenue, up ~1,998% YoY; $56.2M net loss | Tom’s Hardware |
| Huawei | AI training and inference (China) | Atlas 960 SuperPoD | Ascend 960DT training chip due Q1 2027 | Huawei Connect 2026 |
| Intel | AI inference (data center) | Crescent Island (160GB, Xe3P) | Customer sampling planned for H2 2026 | CRN |
| Qualcomm | Mobile GPU | Snapdragon 8 Elite Extreme Gen 6 | Unveiled at Snapdragon Summit, Sept 22-24 | Qualcomm |
| MediaTek | Mobile GPU | Dimensity 9600 Pro | Unveiled September 15 | MediaTek |
| Imagination Technologies | GPU IP licensing (mobile/auto) | E-Series with Neural Super Resolution | Initial performance data published Sept 25 | Semiconductor Engineering |
The table makes one thing obvious: “besides Nvidia and AMD” isn’t really one market. It’s at least three separate competitions happening at once, AI data center accelerators, flagship mobile GPUs, and GPU IP licensing, each with different leaders and different customers.
The Historical Pattern: GPU Challengers Rarely Die, They Just Shrink
The GPU market has quietly supported more than two players for most of its history. 3dfx, Matrox, S3, and PowerVR all shipped competitive discrete GPUs in the late 1990s and early 2000s before consolidating or exiting the standalone graphics card business entirely. Imagination’s PowerVR lineage is a direct survivor of that era, having pivoted from discrete graphics cards to GPU IP licensing for mobile chips, which is exactly the business it’s still in today with the E-Series.
Intel’s current GPU push is itself a second attempt. The company’s prior standalone GPU efforts, including the i740 in the late 1990s and the Larrabee project that was cancelled before launch, show how long Intel has circled this market without full success. Crescent Island represents a narrower, more achievable bet: inference-only silicon rather than a general-purpose GPU meant to match Nvidia across every workload.
Market Impact: Why This Matters Beyond China
For global AI infrastructure buyers, the rise of Biren, Huawei, and other non-Nvidia, non-AMD accelerator makers mostly matters as a signal of how tight AI chip supply still is. When companies are willing to buy GPUs with a sub-3% market share and a 42.7% gross margin that still produces a loss, it tells you the alternative, waiting for more Nvidia or AMD allocation, is worse. That dynamic keeps pressure on global GPU and AI accelerator pricing even outside China, and it’s part of the same supply story behind Broadcom’s AI chip revenue climbing 221% to $16.7 billion, since custom accelerator demand keeps growing alongside, not instead of, the GPU shortage.
On the mobile side, Qualcomm’s and MediaTek’s new GPU architectures affect a much bigger and more immediate market: every flagship phone shipping in late 2026 and 2027. Imagination’s GPU IP licensing business, meanwhile, is a quieter bet on the idea that not every phone or car needs a GPU built from scratch; some just need a solid licensed design, which is a business model closer to Arm’s, whose own push into AI-tuned CPU cores is already squaring off against AMD’s data-center growth, than to Nvidia’s.
Competitive Comparison: Chip Makers vs. IP Licensors
It’s worth separating these companies by business model rather than just by product, because “GPU maker” means different things depending on who’s doing the making.
| Business Model | Companies | Customer | Revenue Model |
|---|---|---|---|
| Finished AI accelerator chips | Biren Technology, Huawei | Cloud providers, AI labs (primarily China) | Direct chip sales |
| Finished data center inference chips | Intel | Enterprise data centers | Direct chip sales |
| Finished mobile SoCs with integrated GPU | Qualcomm, MediaTek | Smartphone OEMs | Direct chip sales, licensing |
| GPU IP licensing only | Imagination Technologies | Chip designers (mobile, automotive) | Per-chip IP royalties |
This split explains why direct revenue comparisons between, say, Biren and Imagination don’t tell you much on their own. Biren books revenue every time it sells a finished BR166 accelerator. Imagination books a smaller royalty every time a licensee ships a chip using its GPU design, but that design might end up in tens of millions of devices a year across multiple manufacturers.
Why Gross Margin Alone Doesn’t Tell the Full Story
Biren’s 42.7% gross margin looks solid next to its $56.2 million net loss, and that gap is the detail worth watching. Gross margin measures what’s left after the direct cost of making a chip; the net loss includes everything else, R&D for the BR20X-series follow-ups, software engineering, and the optical-interconnect systems that tie racks of accelerators together. A chipmaker can have a healthy gross margin and still lose money for years if it keeps reinvesting at this pace, which is effectively the same playbook Nvidia ran through the early 2010s before its data center business took off.
What’s Still Unverified
Several of the claims circulating around this story are roadmap commitments or reported figures rather than confirmed, audited numbers. Huawei’s Ascend 960DT and 960PR launch windows are company-stated targets for Q1 2027 and Q3 2027, not confirmed shipping dates. The reported 160,000-unit Ascend 950DT deployment tied to DeepSeek in Inner Mongolia is sourced to reporting on the deployment rather than an independently verified shipment count, and should be read with that caveat. Intel’s Crescent Island sampling timeline for the second half of 2026 is also a stated plan, not a confirmed shipping product. None of this means the reports are wrong; it means the usual gap between an announced roadmap and a shipped product still applies here just as it does for Nvidia and AMD.
Predictions: Where the Non-Nvidia, Non-AMD GPU Market Goes Next
Biren’s losses will likely narrow before they close. A 42.7% gross margin on $183.9 million in revenue is a workable foundation; the $56.2 million loss is mostly a function of R&D and infrastructure spend that doesn’t scale linearly with revenue, so the next few reporting periods should show the gap shrinking if revenue keeps climbing.
Huawei’s Ascend roadmap will face the same scrutiny Nvidia’s roadmaps get. As Ascend 960DT and 960PR approach their 2027 target windows, expect more pressure for independent benchmarking rather than company-stated performance claims, especially from customers outside China.
Intel’s inference-only strategy with Crescent Island is the more realistic near-term bet than trying to match Nvidia on training hardware. Expect Intel to keep narrowing its GPU ambitions to specific workloads rather than relaunching a broad competitor to Nvidia’s full training stack.
Mobile GPU competition between Qualcomm and MediaTek will intensify through 2027 as both companies’ new architectures, the Snapdragon 8 Elite Extreme Gen 6 and Dimensity 9600 Pro, ship in flagship phones and get directly benchmarked against each other by reviewers.
Export-restriction-driven demand will keep propping up Chinese GPU makers’ growth rates regardless of whether their chips are actually competitive with Nvidia’s on a per-chip basis, since the growth is substitution-driven rather than purely merit-driven.
Frequently Asked Questions
Who makes GPUs besides Nvidia and AMD in 2026?
Companies with active GPU or GPU-adjacent accelerator businesses include Intel (data center and integrated GPUs), Qualcomm and MediaTek (mobile GPUs), Imagination Technologies (GPU IP licensing), and Chinese chipmakers including Biren Technology and Huawei, which build AI accelerators primarily for the domestic Chinese market.
How much revenue did Biren Technology report for 2026?
Biren reported $183.9 million in revenue for the first half of 2026, up from roughly $8.665 million in the first half of 2025, an increase of about 1,998% year over year, according to Tom’s Hardware.
Is Biren Technology profitable?
No. Despite a 42.7% gross margin, Biren posted a $56.2 million net loss for the first half of 2026 due to continued investment in new accelerators, software, and optical-interconnect systems.
What is Huawei’s Atlas 960 SuperPoD?
It’s a large-scale AI computing system Huawei unveiled at its Connect 2026 conference in Shanghai on September 17, positioned as a domestic alternative to Nvidia- and AMD-based AI infrastructure for Chinese customers.
What is Intel’s Crescent Island chip?
Crescent Island is Intel’s 160GB data center accelerator built on the Xe3P architecture with LPDDR5X memory, designed for AI inference workloads in air-cooled enterprise servers. Intel has said it plans to sample the chip with customers in the second half of 2026.
Does Imagination Technologies make its own GPU chips?
No. Imagination licenses GPU designs, including its E-Series architecture with Neural Super Resolution, to other chipmakers, who then build those designs into their own mobile and automotive chips.
Why are Chinese companies like Biren and Huawei building their own GPUs?
US export restrictions have limited the sale of Nvidia’s and AMD’s most advanced AI chips into China, pushing Chinese cloud providers and AI labs to buy from domestic suppliers even when those chips hold a small share of the overall AI accelerator market.
What’s the newest mobile GPU competing with Qualcomm’s chips?
MediaTek’s Dimensity 9600 Pro, unveiled September 15, 2026, is the company’s latest flagship chip and competes directly with Qualcomm’s Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6, both introduced at the Snapdragon Summit held September 22 to 24.