Ali Ghodsi, chief executive of the data analytics firm Databricks, has sent shockwaves through the technology sector by declaring that the industry is 'running out of GPUs everywhere'. Speaking at a conference in San Francisco on Monday, Ghodsi attributed the crunch to the relentless expansion of artificial intelligence workloads, which require vast numbers of graphics processing units (GPUs) — particularly Nvidia's H100 and forthcoming Blackwell chips.
The warning comes as global demand for AI infrastructure continues to outstrip supply, despite Nvidia's efforts to ramp up production. Ghodsi noted that even large cloud providers are struggling to secure enough hardware, a situation he described as 'unprecedented in its scale and duration'. For UK businesses, the shortage threatens to delay AI deployment and inflate costs for cloud computing services.
London's FTSE 100 slipped 0.3% to 8,214 on Tuesday, with technology-heavy indices feeling the pressure. Shares in UK-listed chip designer Arm Holdings fell 1.8%, while the FTSE 350 Technology Index dropped 0.9%. Analysts at Peel Hunt commented that the GPU scarcity 'creates a bottleneck for the entire AI value chain', potentially slowing innovation and raising barriers for startups.
The shortage has also hit UK pension funds with exposure to AI stocks. The Legal & General Future World Fund, which holds significant positions in Nvidia and Microsoft, saw its value dip as investors reassessed growth expectations. 'The GPU crunch is a double-edged sword,' said Sophie Lund-Yates, an equity analyst at Hargreaves Lansdown. 'It benefits Nvidia's pricing power but risks capping the pace of AI adoption across sectors.'
Databricks itself, which recently raised $10bn in a funding round, relies on GPU clusters to power its data lakehouse platform. Ghodsi's comments underscore the strategic importance of chip supply for AI firms, many of which are based in the UK. The British government has identified semiconductor sovereignty as a national priority, but domestic production remains minimal.