Major procurement order announced for 50,000 units of domestic Chinese GPU chips.
This afternoon, A-shares trading session broke with major news: ByteDance is negotiating with Tianshu Zhixin to purchase at least 50,000 AI inference chips.
Many treat this as a speculative concept play. But the significance of this news extends far beyond a few trading halts in late-session trading—it signals an inflection point for China's domestic AI computing chip sector on A-shares.
ByteDance is currently China's largest AI application company. Doubao, its AI assistant, boasts 368 million monthly active users, with daily token invocations surpassing 14 trillion—a year-over-year increase exceeding 1,000x. Globally, only one company can meet inference demand at this scale: NVIDIA.
But ByteDance has now taken two decisive steps. First, it has split its supply chain into two lines—training and inference. Large-scale model training uses Huawei Ascend and Cambricon, while high-volume concurrent inference draws from Tianshu Zhixin.
Previously, domestically made chips were confined to policy-backed scenarios: government cloud and AI computing centers. But now ByteDance has proven with real capital deployment that domestic inference GPUs are already viable in large-scale production environments.
Three things converged in today's market: advanced packaging surged, wafer manufacturing strengthened, and computing chips showed late-session volatility—none of these moves were isolated.
ChipTech Technology hit a trading halt. TSMC raised prices across the board. Huahong Semiconductor reached all-time highs. The late-session computing chip movement carries direct logic: ByteDance's 50,000-chip purchase is just the tip of the iceberg.
When China's most significant AI application begins running inference on domestic chips, the demand-side logic for the entire domestic computing power chain undergoes fundamental change—from policy-driven to demand-driven.
Today the Shanghai Composite oscillated around 4,100 points and turned positive. The Star 50 Index surged over 3%. Yet over 4,400 stocks across the market declined.
This extreme structural divergence won't end in a single day, but the trend is unmistakable: semiconductor chips are becoming capital's most certain main line.
A-shares will likely continue this structural pattern tomorrow. Computing chips have potential for a gap-up opening, but note this: the news already circulated on June 17th. Today's late-session movement represents emotional release following additional catalysts.
Tomorrow's key focus is not chasing trading halts, but whether trading volume can continue expanding. New capital taking over is what will confirm the main line's sustainability.
Also monitor overall market volume. Today's estimated combined trading volume across both markets is around 3.3 trillion yuan. If tomorrow rebounds to 3.5 trillion yuan or higher, the foundation for a structural bull market remains solid.
Guojin Securities points out that explosive growth in token invocations creates massive demand for computing power. Huatai Securities calculates that domestic chip supply is constrained by foundry services and HBM segments, with significant shortages still persisting through 2026.
Institutional views converge on a single conclusion: the competitive landscape for AI computing power is shifting from NVIDIA's unipolar dominance to a binary structure—overseas training cards paired with domestic inference cards.