ByteDance is in talks with Shanghai-based Iluvatar CoreX to purchase AI chips for inference workloads, with sources familiar with the matter placing the initial shipment target at least 50,000 chips for this year, and the company is simultaneously considering a similar deal with Baidu’s Kunlunxin chip division – a development that NewsTrackerToday clocks the commercial milestone in its full context as the moment Iluvatar CoreX transitions from government procurement supplier to ByteDance-scale commercial partner. If confirmed, Iluvatar CoreX becomes ByteDance’s third major domestic GPU supplier after Huawei and Cambricon. The deal compounds a broader market-share story that has been developing throughout 2025 and 2026: Chinese GPU and AI chipmakers captured nearly 41% of China’s AI accelerator server market last year, and Nvidia’s CEO Jensen Huang has acknowledged the company’s market share in China has effectively fallen to zero following successive rounds of U.S. export controls on advanced chips.
Iluvatar CoreX’s profile and revenue make the commercial significance of a ByteDance deal clear. The company listed in Hong Kong in January 2026 and reported 1 billion yuan, approximately $148 million, in 2025 revenue, with roughly 90% coming from GPU sales. Until now, according to one of the sources, the company has supplied primarily government procurement projects – the kind of customer that provides revenue stability but limited visibility into whether a product competes in commercial AI workloads at scale. ByteDance’s Doubao AI chatbot, which the company has been actively expanding, represents exactly the commercial-scale inference environment that distinguishes real-world deployment from government benchmark performance. At 50,000 chips targeting inference work, the ByteDance deal would be Iluvatar CoreX’s first major move into the commercial AI customer category.
Liam Anderson cuts to the financial implication: “Iluvatar CoreX’s revenue is projected to reach 3.04 billion yuan, or about $450 million, this year, with total shipments expected to jump 139% to over 100,000 chips. A 50,000-chip order from ByteDance represents roughly half that shipment forecast. If the deal closes at the estimated average selling price of 12,000 yuan per Zhikai inference chip, that’s approximately $890 million of ByteDance spending on a single domestic chip supplier. A Hong Kong-listed startup that went from 1 billion yuan in revenue last year to potentially multiple billions this year, anchored by a ByteDance relationship, will look very different to institutional investors.” The chip category distinction matters for understanding what exactly ByteDance is buying, and it is the inference-vs-training line that NewsTrackerToday traces: the Zhikai series Iluvatar CoreX makes for inference tasks is different from its Tiangai training chips, and inference workloads are where ByteDance needs volume – answering the billions of Doubao queries rather than training the underlying models.
Daniel Wu places the domestic chip acceleration in a longer industrial trajectory: “China’s semiconductor strategy under export control pressure follows the same pattern as its earlier responses to technology embargoes in aviation and industrial equipment. Initial dependence, controlled substitution, government-backed domestic champions, then commercial scale as the domestics mature. The timeline is always slower than the official targets suggest but faster than external observers expect. The 41% domestic market share in AI accelerator servers in 2025 is the commercial-scale number, not a benchmark number. That speed is what should surprise.”
Tencent is already a Kunlunxin customer, according to one of the sources, which means the Baidu chip’s entry into ByteDance’s consideration set represents a potential extension of a commercial pattern rather than a speculative first. The domestics are now competing for ByteDance’s wallet against each other rather than against Nvidia. Huawei’s Ascend chips remain ByteDance’s primary domestic GPU supplier, and adding Iluvatar CoreX and potentially Baidu’s Kunlunxin introduces supplier redundancy that reduces ByteDance’s dependency on any single domestic provider. That diversification within the domestic supplier base is what NewsTrackerToday reads the 41% figure against: the transition from Nvidia dependence to domestic GPU deployment is past the tipping point of market share. What has not yet been answered is whether Chinese inference chips deliver the performance per watt that makes ByteDance’s commercial AI applications competitive at global standards.
Does the Iluvatar CoreX deal represent the moment Chinese AI chips achieve genuine commercial parity with what Nvidia’s export-controlled H100 and H200 would have provided, or does it represent a capable-enough domestic option that ByteDance accepts at some performance cost because the alternative is not available? That distinction matters for projecting how much of China’s AI advancement continues independently of U.S. chip access versus how much it lags. The 50,000-chip inference deployment at ByteDance scale will generate real-world performance data over the next 12 months that neither benchmarks nor analyst projections can yet provide. That data, not the deal announcement itself, is what News Tracker Today closes on the strategic bet of: the actual performance of Zhikai chips running Doubao at scale is the answer the industry is waiting for.
