China’s AI industry is facing a widening gap between demand for computing power and the chips available to meet it, according to a report by Bloomingbit. The report says demand for AI semiconductors could exceed five million units this year, against supply of no more than 3.5 million, as companies turn to Nvidia GPUs and overseas data centres to fill the gap.
Watch Desk analysis
What happened
Bloomingbit reports that the Financial Times put China’s AI-chip demand above five million units this year, with supply capped at 3.5 million, including about one million chips produced overseas. It says demand could top 10 million next year. Those are forecasts relayed by the report, not a final count of chips delivered.
The report says Chinese technology companies bought more than one million Nvidia gaming and workstation GPUs this year. It also describes efforts to build domestic capacity: Huawei is targeting 1.6 million AI chips this year, while Alibaba and Baidu are expected to produce one million combined. Tencent, meanwhile, has signed a deal with Oracle to lease 100,000 GPUs at data centres in Southeast Asia.
The piece also places the shortage alongside rising infrastructure spending. It cites a forecast of about 700 billion yuan in combined capital expenditure this year by Alibaba, ByteDance and Tencent, and a Morgan Stanley projection of US$193 billion in Chinese computing-infrastructure spending.
Why it matters
AI needs computing capacity both to train models and to serve them to users. When suitable chips are scarce, companies may have to rely on less specialised hardware or rent compute abroad. The report says Nvidia GPUs can still handle some early-stage inference tasks, even as they trail Huawei’s Ascend 910C in most areas.
That makes the chip gap more than a manufacturing scoreboard: it can shape which services companies can run, where they run them and how quickly they can expand. The report also describes CUDA, Nvidia’s software platform, as a reason developers may find switching to domestic alternatives difficult.
Our read
The striking figure is not simply how many chips China wants. It is how many workarounds the shortage is prompting, from gaming GPUs to rented capacity overseas, while domestic alternatives scale up. That is a useful snapshot of the infrastructure race, not proof that any one forecast will land exactly as stated. The real test is whether supply and usable software catch up with demand.
What to watch
- Whether reported demand and supply forecasts are revised as the year progresses.
- How quickly domestic chipmakers deliver the output targets described in the report.
- Whether overseas GPU leases and Nvidia purchases continue as local capacity expands.
Discussion spark: Should AI companies prioritise building domestic chip capacity, or is renting compute abroad a sensible long-term part of the supply mix?
Sources and evidence
- DeepSeek Suffers Nearly 5-Hour Outage as China’s AI Boom Fuels Chip Shortage – bloomingbit (16 October 2026, 07:43 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.