Arm chief executive Rene Haas says the computing industry’s ambitions for AI in medicine are running into a more immediate constraint: chip supply.
In an interview with the BBC, Haas said AI could eventually help solve biological problems—such as modelling cells, people and the effects of cancer on DNA markers—that remain beyond the reach of current systems. He predicted that AI would help cure cancer within his lifetime. That is a long-range view from an executive whose company designs processor architectures used across phones, devices and, increasingly, data centres—not a clinical forecast or evidence of a new treatment.
The more concrete business point is that demand for AI infrastructure is outpacing the available supply of chips and fab capacity. “We are absolutely in a supply-constrained environment,” Haas said, pointing to planned multi-gigawatt data centres in the US and France.
Compute is becoming a delivery constraint
For companies building frontier models, the shortage is not simply about buying a few more accelerators. It affects the ability to secure systems, power, networking, packaging and manufacturing slots on schedules that support data-centre expansion. More chip factories are needed before some of the sector’s more speculative infrastructure proposals—such as data centres in space—become feasible, Haas said.
Arm has a direct stake in that buildout. Haas said its power-efficient technology is now used in half of AI data centres globally. The company has also moved further into selling its own chips; he said demand for an Arm AGI chip developed for Meta had exceeded $2 billion since its March launch.
For operators, that combination of supply limits and demand concentration raises the value of infrastructure flexibility. Hardware roadmaps, cloud capacity commitments and model-development plans increasingly need to be managed together. A product strategy that assumes unconstrained access to leading-edge compute may become expensive—or simply delayed.
Bigger clusters are not the whole medical-AI story
Haas frames future advances in cancer research as a function of better models and more sophisticated computers. But Chris Bakal, a professor at the Institute of Cancer Research in London and chief executive of Sentinal4D, offered a useful qualification: medical AI’s advantage will depend on the measurements and patient data supplied to it, rather than solely on the biggest available data centre.
Bakal said his lab trains models on data generated from patient samples, rather than data scraped from the internet, and that such work does not necessarily require giant data centres. Better prediction could shorten treatment-development timelines, he said.
That distinction matters for founders and healthcare leaders. The bottleneck in a medical-AI program may be access to high-quality, appropriately governed clinical data, experimental design and validation—not raw GPU capacity. Larger models can improve some capabilities, but they do not replace the evidence required to turn a prediction into a safe diagnostic, drug candidate or treatment decision.
A supply-chain question for the UK
Haas also rejected the idea that the UK needs to build leading-edge chip fabs. Such facilities are expensive, labour-intensive and resource-heavy, he said, and depend on a broad specialised ecosystem. Manufacturing remains heavily concentrated around Taiwan’s TSMC, even as governments seek more resilient semiconductor supply chains.
Arm’s position illustrates a different UK technology playbook: retain high-value design, research and engineering talent rather than replicate every part of semiconductor manufacturing. Haas said half of Arm’s employees remain in the UK and that it is Cambridge’s largest employer.
What to watch
The next test is whether supply constraints ease quickly enough to meet the AI industry’s data-centre plans. But in healthcare, the more meaningful signal will be whether AI systems trained on strong patient and laboratory data can produce validated results that improve development timelines or patient outcomes. Compute matters. In medicine, it is only one part of the stack.
