ByteDance wants its own chips. The company behind TikTok and a string of powerful AI models no longer accepts dependence on foreign suppliers. Executives have set an aggressive timeline. They plan to complete the design of a next-generation central processing unit by early 2027. Mass production and internal deployment should follow in the second half of that year.
An early version already runs inside ByteDance data centers. It has operated there since late 2025. The move comes as the firm’s computing demands explode. Its Doubao chatbot and Seedance AI model generate enormous workloads. These systems require not just graphics processors but general-purpose CPUs to orchestrate tasks efficiently.
Beijing’s Export Controls Force a Faster Path
U.S. restrictions on advanced chips hit hard. Nvidia’s H100 and H20 models face tight controls. ByteDance responded the way other tech giants have. It follows the playbook written by Google with its Tensor Processing Units, Amazon with Trainium and Graviton processors, and Microsoft with its Maia chips. Build in-house. Reduce reliance. Control the stack.
Qualcomm lends a hand on the manufacturing side. The partnership helps ByteDance secure foundry capacity at a moment when capacity remains scarce. Tape-out dates for the new CPU, once announced, will offer markets a clearer signal of ByteDance’s semiconductor seriousness. Until then speculation fills the gap.
Yet ByteDance stands far from alone. A separate Chinese firm founded in 2019 has drawn fresh capital and public market attention in recent months. Biren Technology listed on the Hong Kong Stock Exchange in January 2026. Shares jumped 76 percent on debut, according to Bloomberg. The company raised roughly $717 million in that initial public offering.
Now Biren seeks even more. It aims to raise HK$7 billion, about $892.5 million, through a new share sale. Sixty percent of the proceeds target commercialization and mass production of its next-generation general-purpose GPUs. Another 20 percent goes to research and development. The South China Morning Post first detailed the fundraising plans just days ago.
Biren’s BR100 chip once claimed performance on par with Nvidia’s A100 in certain metrics. Company specifications listed 1 PFLOPS in BF16 and 2 POPS in INT8. Those numbers looked impressive on paper. Real-world adoption told a more complex story. U.S. sanctions placed Biren on export control lists, limiting access to advanced manufacturing tools.
Still the firm pushes forward. Its BR20X series targets inference workloads with support for FP8 and FP4 precision. Plans call for release this year. Revenue reached about $150 million in 2025, up 207 percent from the prior year. Research spending exceeded revenue. The company invested $210 million in R&D while generating far less in sales. That imbalance shows confidence. Or urgency. Probably both.
Investors appear to share the sentiment. Biren’s stock has climbed more than 150 percent since its Hong Kong debut. The rally reflects broader excitement around domestic AI hardware. MetaX and Moore Threads also went public recently. Their shares soared hundreds of percent on first trading days, CNBC reported in December 2025. Moore Threads, sometimes called “China’s Nvidia,” was founded by a former executive from Nvidia’s Chinese operations.
Enflame, another player started by ex-AMD engineers, focuses on training and inference chips for data centers. Huawei, Alibaba, Baidu and Cambricon round out the field of serious contenders. None yet match Nvidia’s full software ecosystem or raw performance at scale. But they don’t need to. Not in China.
Executives at Chinese firms plan to increase spending on domestic AI accelerators from 30 percent today to 46 percent over the next 12 months. That’s according to a Bloomberg survey cited in recent social media analysis. Beijing eyes roughly 2 trillion yuan in data center investment across the next five years. At least 80 percent of core technologies should come from local suppliers.
The shift carries strategic weight. Washington intended export controls to slow China’s AI progress by three to five years. Instead the policy accelerated domestic innovation. “Good enough” silicon produced inside China now competes for contracts once reserved for Nvidia hardware.
Biren’s latest funding round underscores the point. Customers, the filing states, include cloud providers, AI data centers and large enterprises. All of them expand computing capacity rapidly. Demand outstrips supply. Foreign chips remain difficult to obtain at scale. Domestic alternatives fill the void.
And the alternatives keep improving. Biren’s BR200 series offers better cost structures than its predecessor. The company developed a software platform called BIRENSUPA to ease deployment. Early commercialization efforts focus on high-performance computing clusters where multi-GPU scaling matters.
ByteDance’s CPU project points to an even deeper ambition. Training large models requires massive parallel compute. Inference and orchestration benefit from specialized CPUs that handle scheduling, data movement and lightweight tasks. A vertically integrated stack gives ByteDance control over latency, efficiency and data security.
Other Chinese internet giants pursue similar strategies. The pattern repeats across the industry. Sanctions didn’t cripple progress. They redirected it. Chinese firms now treat silicon independence as a national habit rather than a temporary fix.
Challenges remain. Advanced semiconductor manufacturing still depends on equipment and processes dominated by Western and allied suppliers. Yield rates on leading-edge nodes lag. Software compatibility with global AI frameworks takes time to mature. Power consumption and cooling demands grow with every performance gain.
Yet progress arrives faster than many outsiders predicted. Biren’s IPO success and rapid follow-on fundraising demonstrate investor belief. So does the debut performance of peer companies. The market has spoken. Domestic AI chips represent real business in China.
ByteDance’s timeline adds pressure. If the firm hits its early 2027 design target, competitors will feel the heat. Internal deployment in the second half of that year could validate the architecture at scale. Success would encourage other hyperscalers to accelerate their own chip programs.
The broader implications stretch beyond hardware. China’s AI compute capacity expands regardless of U.S. policy. Data centers under construction prioritize local silicon wherever possible. That reality reshapes global supply chains, investment flows and technology competition.
Nvidia still leads in absolute performance and software maturity. Its chips remain the gold standard for many workloads. But in the world’s second-largest economy, the conversation has changed. Buyers ask different questions now. Not whether a chip comes from Santa Clara. But whether it meets requirements, scales reliably and avoids future supply shocks.
Biren, ByteDance and their peers provide answers. Their chips may trail on some benchmarks. Their ecosystems lack the polish of CUDA. None of that stops adoption when the alternative means waiting on export licenses or paying premiums on gray-market hardware.
So the race continues. Billions pour into new designs, new fabs, new software layers. Timelines compress. Ambitions swell. The TikTok creator’s quiet move into silicon captures the moment perfectly. What began as a content platform now fuels one of the most determined technology catch-up efforts in modern history.
Watch the tape-out dates. Track the production ramps. Measure the actual performance inside live data centers. Those numbers will decide whether these Chinese challengers blunt Nvidia’s dominance or merely nibble at its edges. For now the momentum sits firmly with the domestic players. And they show no sign of slowing down.
WebProNews is an iEntry Publication