Arm Holdings PLC (NASDAQ: ARM) CEO Rene Haas said artificial intelligence could help find a cure for cancer within our lifetime, but warned that chip shortages are limiting the expansion of AI infrastructure.
Haas said AI could eventually solve complex problems in cancer research that are difficult for humans and current computers to model. He also said AI could accelerate drug development and testing, according to an interview with the BBC published on Tuesday.
In the interview, Haas said modelling cells, humans and the effect of cancer on DNA markers remains too complex today, but advances in computing and AI models could eventually make those problems solvable.
AI Drug Discovery
The computing requirements behind AI-driven drug discovery are already substantial. Generate Biomedicines co-founder and CTO Gevorg Grigoryan said large-scale model training, molecular generation and evaluation require access to a large number of GPUs. He also said computing power alone is not enough, with proprietary biological data and large-scale experimentation needed to turn that infrastructure into a durable platform.
Pharmaceutical companies are also expanding AI infrastructure for research. Bristol Myers Squibb Co (NYSE:BMY) has deployed a new NVIDIA AI computing system to support work across cancer and other therapeutic areas, while using biological AI tools to train models on decades of proprietary research data.
That need for more computing power is running into a broader semiconductor supply constraint.
Chip Bottleneck For AI Data Centers
Haas said the shortage of chips needed for data centers is holding back the further expansion of AI infrastructure. He described the industry as being in an "absolutely supply-constrained environment" and said more chip factories are needed to support increasingly large data centers. Haas also said that more chip factories are needed before a data center can be put in space.
While the AI buildout is driving demand for more chips, Haas said the semiconductor industry cannot add manufacturing capacity overnight. Fabs can cost billions of dollars and require specialized workers and resources, creating a physical limit on how quickly AI infrastructure can expand.
However, Haas does not think that manufacturing — which is dominated by tech giants like Taiwan Semiconductor Manufacturing Company Limited (NYSE:TSM) — could eventually occur in the U.K., in spite of the government carrying out discussions with the industry about the possibility of bringing parts of the physical chip supply chain over there.
The pressure extends to AI memory, where supply may remain tight for years. Industry executives have warned that high-bandwidth memory capacity cannot be expanded quickly enough to match the pace of AI infrastructure investment, with production requiring advanced manufacturing and years of additional capacity.
Haas also said Arm’s power-efficient technology is used in about half of AI data centers worldwide. He said demand for Arm’s AGI chip, developed with Meta Platforms Inc. (NASDAQ: META) , has exceeded $2 billion since its launch.
Image via Shutterstock
Login to comment