As 2026 kicks off, China’s top internet giants are taking their rivalry to a new frontier: AI chips. Baidu’s AI chip arm, Kunlun Chip, recently filed confidentially for an IPO in Hong Kong, sending Baidu’s U.S.-listed shares soaring 15% in a single session.
The move has investors reassessing the tech muscle of the internet veteran, long thought to be trailing behind in China’s fast-moving AI scene.
Baidu isn’t alone. The second half of 2025 saw a wave of eye-catching domestic chip IPOs, including Moore Threads, hailed as the “first domestic GPU stock.” Its debut on the STAR Market was spectacular: shares surged 425% on day one, creating instant paper millionaires among lucky retail investors.
Analysts say the contrasting stories of Kunlun Chip and Moore Threads illustrate the current state of China’s AI chip sector—one grounded in the certainty of established ecosystems, the other fueled by big-picture ambition and imagination.
Kunlun Chip has built its case gradually. After an independent financing round in 2021, the unit was valued at 13 billion yuan, rising to 21 billion yuan following a Series D round in 2025. Growth has been measured, but firmly anchored in results—and in Baidu’s backing.
Baidu’s chip ambitions trace back to 2011, when soaring costs for search operations pushed the company to build its own chips, years before Google unveiled the TPU. Kunlun Chip mostly supported Baidu’s internal operations until its spin-off in 2021 under Chief Chip Architect Ouyang Jian, when commercialization became a focus.
The company’s P800 third-generation chip hit mass production in 2024, helping Kunlun Chip move beyond internal deployment. Its integration into Baidu’s AI ecosystem—including search, ERNIE Bot, and autonomous driving—provided a built-in testing ground and initial customer base.
Revenue is already climbing. In 2025, Kunlun Chip won a billion-yuan centralized procurement project from China Mobile and serves clients including China Merchants Bank, China Southern Power Grid, Geely Auto, and vivo. Expected 2025 revenue tops 3.5 billion yuan, with JPMorgan forecasting 2026 revenue could hit 8.3 billion yuan. Haitong International recently estimated the unit’s market value could approach $50 billion after the IPO.
Moore Threads, by contrast, has been more of a market sensation than a profit story. Its STAR Market debut on Dec. 5, 2025, drew 4.8 million subscriptions for an IPO with a winning rate of just 0.036%, creating a frenzy among retail investors. Its first-day market cap hit 280 billion yuan, generating nearly 8.2 billion yuan in paper gains.
But Moore Threads is still operating at a loss. In 2024, revenue totaled 438 million yuan with a net loss of 1.618 billion yuan. In the first half of 2025, losses continued at 271 million yuan. The high valuation reflects investor faith in the company’s full-featured GPU strategy and its pedigree: founder Zhang Jianzhong and key team members come from NVIDIA and AMD. Moore Threads is betting it can replicate NVIDIA’s ecosystem success in China using its MUSA architecture, even before it turns a profit.
Technical Paths: Specialist vs. Generalist
The two companies have taken very different technical approaches. Kunlun Chip is an AI specialist, laser-focused on deep learning workloads. Its self-developed XPU architecture powers large model training and inference efficiently, outperforming NVIDIA’s China-only H20 in key AI benchmarks. A single machine with eight P800 cards can train massive models with hundreds of billions of parameters.
Cluster performance is another edge. Kunlun Chip deployed a 30,000-card cluster in early 2025, supporting multiple concurrent model trainings and serving over a thousand enterprise clients. Such scale is rare and demonstrates operational reliability—a critical factor for enterprise adoption.
Moore Threads takes a different approach: a full-featured GPU strategy. Its MUSA unified architecture handles AI training and inference, graphics rendering, video encoding/decoding, and scientific computing. Chips like Huashan and Lushan form a collaborative multi-product ecosystem. The advantage: versatility across industries such as gaming, professional design, and digital twins. The drawback: high R&D costs and a long road to profitability.
In today’s chip market, performance alone doesn’t win. Ecosystem matters. Kunlun Chip benefits from Baidu’s closed-loop system. Its chips integrate with the PaddlePaddle deep learning framework and the Wenxin large model, supporting over 1.1 million models. Internal business units act as testing grounds and initial customers, enabling rapid monetization.
Moore Threads is betting on open-source co-creation. Facing NVIDIA’s entrenched CUDA ecosystem, it launched Moore Academy and provides open-source libraries to attract developers. By the end of 2025, over 760,000 enterprises were part of the MUSA ecosystem. While this approach promises long-term resilience, adoption is gradual and capital-intensive.
“Kunlun Chip converts technology into revenue fast; Moore Threads is a long-term bet on building a Chinese version of NVIDIA,” said a Beijing-based semiconductor analyst. “One offers certainty, the other imagination.”
Short-Term Gains, Long-Term Plays
In the near term, Kunlun Chip is well-positioned. Its integration into Baidu’s AI ecosystem, proven performance, and early revenue wins give it a first-mover advantage. Moore Threads, while technologically promising, must scale commercialization and ecosystem adoption to justify its valuation.
Looking further ahead, Kunlun Chip’s position may strengthen. Baidu has invested over 180 billion yuan in R&D over the past decade, supporting continuous chip iteration. Roadmaps include the M100 inference chip in 2026, the M300 training chip in 2027, and clusters at million-card scale by 2030.
Moore Threads still has opportunities in areas like graphics rendering, professional computing, and consumer GPUs. Emerging sectors such as embodied intelligence and digital twins could favor its full-stack capabilities. The company must accelerate ecosystem growth and achieve profitability to maintain investor confidence.
What This Means for China’s AI Chip Race
The competition between Kunlun Chip and Moore Threads reflects a broader national push for semiconductor independence. Investors and analysts see the two companies as complementary case studies: one demonstrating the power of ecosystem synergy and operational certainty, the other showcasing ambition and ecosystem-building potential.
While capital markets often favor quick results, industry insiders emphasize patience. “The real test isn’t first-day trading gains,” said a Shanghai-based chip investor. “It’s who can consistently deliver technology, build an ecosystem, and navigate cycles over the next five to ten years.”
As IPOs continue to reshape China’s AI chip sector, both Kunlun Chip and Moore Threads will need to prove that their strategies—whether leveraging a giant ecosystem or betting on open innovation—can translate into long-term market dominance.
For investors, the message is clear: the AI chip race is no longer just about specs on paper; it’s a marathon of technology, ecosystem, and execution. The winners will be those who endure the cycles and turn early promise into sustainable growth.
