DeepSeek’s AI chip is a wake‑up call: the arms race has only just started
DeepSeek’s push to design its own AI inference chip is a clear sign that the real AI arms race is shifting from flashy models to control of the hardware stack.
China’s DeepSeek is reportedly building its own AI inference chip to cut reliance on Nvidia and even domestic champion Huawei. That is not just another chip story; it is a clear signal that the real AI arms race is shifting from flashy models to control of the hardware stack.
DeepSeek made its name with cheap, capable models tuned for Huawei silicon, showing you do not need Silicon Valley‑style budgets to matter in this space. Now it is quietly hiring semiconductor talent and lining up foundries to design its own silicon, targeting the fastest‑growing part of AI demand: inference at scale, not just headline‑grabbing training runs. In a world of tightening US export controls and GPU shortages, this is about sovereignty, not just margins.
In my recent posts on the AI hype cycle and on why open institutions matter, I argued that power is crystallising around the boring bits of AI: infrastructure, governance, and who controls the rails. DeepSeek’s move proves the point. When model labs start designing chips, hardware becomes a geopolitical instrument. Whoever owns the compute can decide who gets to participate, under which rules, and at what price.
Markets clearly get the stakes: news of DeepSeek’s chip plans was enough to spook global chip stocks and raise questions about Nvidia’s long‑term dominance. But the broader lesson for boards and policymakers is blunter. If you’re still treating AI as an “apps and use‑cases” problem, you’re already behind. Strategy now has to extend down the stack to supply‑chain exposure, export‑control risk, and concentration around a tiny handful of fabs and vendors.
We keep talking about the AI arms race as if it is nearing a finish line. DeepSeek just reminded us we have only run the first lap. The next phase will be fought over fabs, export licences, and who can keep the lights on when the geopolitics get rough.
For Australia, DeepSeek’s chip play is another reminder that our AI future will be shaped as much in fabs and data centres as in policy roundtables. The government’s new National AI Plan and its expectations for data centres and AI infrastructure are a start, but they still lean heavily on foreign silicon and hyperscaler capacity. If we want genuine AI sovereignty in the Indo‑Pacific, we need to treat compute, critical minerals, and regional infrastructure as strategic assets - not just back‑office plumbing - and be honest that the hardware battles now unfolding between Washington and Beijing will wash up on our shores whether we are ready or not.
