China announces 50,000-unit GPU procurement from domestic producer Iluvatar CoreX, signaling major shift away from Nvidia.

This afternoon, A-shares delivered major news in the final trading session: ByteDance is negotiating the purchase of at least 50,000 AI inference chips from Tencent Chips.

Many treat this as a concept trade. But its significance extends far beyond the closing rallies—it signals a inflection point for China's domestic computing power sector.

ByteDance is currently China's largest AI application company. Its Doubao assistant has 368 million monthly active users and processes 1.4 million trillion tokens daily, representing over 1,000 times year-over-year growth. Globally, only one company can meet inference demands at this scale: Nvidia.

But ByteDance has now taken two critical steps. First, it split its supply chain into separate training and inference pipelines, using Huawei Ascend and Cambricon for large-scale model training while introducing Tencent Chips for massive concurrent inference. Second, it has proven through real capital allocation that domestic inference GPUs are now viable in large-scale production environments.

Previously, domestic chips carried the label of government cloud and AI compute centers—policy-backed scenarios. But ByteDance's move changed that calculation in a single transaction.

Three things happened in today's market: advanced packaging surged, wafer manufacturing strengthened, and inference chips moved sharply lower. This was not isolated.

SMIC hit a limit-up, TSMC raised prices across the board, HongHui Electronics reached all-time highs. The inference chip volatility had clearer logic—ByteDance's 50,000-unit purchase is just the tip of the iceberg.

When China's heaviest-weight AI application begins running inference on domestic chips, the entire domestic computing ecosystem's demand-side logic shifts fundamentally: from policy-driven to demand-driven.

The Shanghai Composite oscillated near 4,100 and closed red today. The Sci-Tech 50 surged over 3%. Yet across the market, more than 4,400 stocks declined.

This extreme structural divergence won't end in a day, but the trend is clear: semiconductor chips are becoming the market's most certain main thread.

Tomorrow, A-shares will likely continue the structural pattern. Inference chips could open higher, but note: the news circulated on June 17th; today's close represented an emotional release following compounding catalysts.

Tomorrow's focus is not chasing rallies but whether turnover can expand further—new capital participation confirms the main thread's sustainability. Watch overall market volume. Today's estimated combined turnover was around 3.3 trillion yuan. If tomorrow rebounds above 3.5 trillion, the structural bull market's foundation remains solid.

Guotai Securities emphasized: exploding token calls create massive demand for computing power. Huawei Securities calculated that domestic chip supply remains constrained by foundry and HBM segments, with significant shortfalls persisting through 2026.

Institutional analysis points to one conclusion: the competitive landscape for AI computing is shifting from Nvidia's single-pole dominance toward a binary structure of overseas training cards paired with domestic inference cards.