Backed by Beijing's Policies, Chinese AI Chip Makers Enter a Breakout Earnings Season
Miles Bennett
Beijing is ordering tech firms to swap Nvidia for domestic chips, and the numbers are landing — Cambricon, Metax, and Moore Threads all report triple-digit revenue growth. Planned domestic chip spending jumps from 30% to 46% of AI accelerator budgets. China's AI supply chain is shifting from backup to mainline.
How big is this revenue surge?
Cambricon expects a sharp revenue jump in H1; Metax projects sales to triple; Moore Threads disclosed H1 2026 revenue growth of up to 149%.
This means → three companies reporting triple-digit growth simultaneously is not one firm's luck — it signals the entire domestic AI chip pipeline scaling up at once.
The trigger: Beijing told local tech firms to find alternatives to Nvidia's flagships. The U.S. approved H200 sales to China, but officials said actual shipments were "negligible" — Nvidia and AMD are effectively locked out.
Who is grabbing Nvidia's lost share?
Huawei is ramping production of its flagship AI chip, the Ascend 910C, targeting roughly 600,000 units in 2026.
Alibaba's chip unit T-Head is also expected to expand its share.
A Bloomberg Intelligence survey in June found Chinese firms plan to spend 46% of their AI accelerator budgets on domestic chips over the next 12 months, up from 30% today. This means → nearly half of procurement dollars are migrating from Nvidia to local suppliers.
Can the ¥2 trillion data-center plan sustain this?
Beijing plans to invest ¥2 trillion (~$295 billion) in data centers over five years, with at least 80% of AI chips and related tech sourced from Huawei and other domestic vendors.
Morgan Stanley estimates this will lift China's AI chip self-sufficiency rate from 42% in 2025 to 70% by decade's end.
In plain terms = this is not a one-off policy boost — it is a five-year order book with explicit domestic-sourcing quotas baked in.
Each chip trails Nvidia — so how do they compensate?
At Shanghai's top AI summit last month, Huawei, Moore Threads, and Metax showcased server racks integrating tens of thousands of chips, using cluster scale to offset the single-chip performance gap.
Morningstar analyst Phelix Lee: "The message they're trying to send is — we have enough chips and the infrastructure to connect them all, so we can outscale the competition."
In plain terms = if one chip can't outrun the rival, wire tens of thousands together — provided the networking and software stack can keep up.
Why is the inference market the new battleground?
Morgan Stanley analyst Charlie Chan notes that customers are shifting from training AI models to deploying AI in business workflows. Chinese chipmakers are winning orders with competitive inference pricing.
"Purchasing decisions are increasingly driven by deployment economics, not peak single-chip performance."
This means → when buyers care about cost-per-inference rather than benchmark scores, domestic chips' price-performance edge converts directly into orders.
Where is the biggest risk?
Chinese chips still trail Nvidia by several years in single-chip performance, largely because U.S.-led export controls block ASML's EUV lithography machines — the extreme-ultraviolet tools needed to etch the most advanced circuits — from entering China.
TSMC relies on exactly these EUV tools to build the world's most powerful AI accelerators; Chinese fabs cannot match that process capability.
This reflects a core open question: can domestic substitution achieve scale deployment before the performance gap narrows? The window defined by speed and scale is what determines whether this thesis pays off.
Content is for reference only, not financial advice.