Europe’s chip sovereignty claim rests on illusions

Europe’s first home-designed supercomputer chip has reached testing, but it will not free the continent from dependence on foreign technology: the hardware is still assembled at TSMC’s factory in Taiwan, and European chipmaking capacity is unlikely to grow significantly soon, ERR’s R2 Portal reports in a commentary by Kristjan Port.

The Rhea1 samples entered testing in late September. The processor, described as the most complex server chip designed in Europe so far, is intended to power a future expansion of Jupiter, the European Union’s first exascale-capable supercomputer, which began operating in Germany a year ago.

Jupiter ranks fifth in the public supercomputer rankings, but its performance relies on 24,000 GH200 chips made by US company Nvidia. That chip generation is now outdated: its performance is several times below that of the newer GB series and about 25 times below Nvidia’s latest Vera Rubin generation. Port notes that the industry faces two major bottlenecks: producing advanced processors and manufacturing the equipment used to make them.

The scale of US investment highlights the gap. In 2024, Elon Musk established the Colossus artificial intelligence computing centre in Memphis, Tennessee, with 230,000 Nvidia H100, H200 and GB200 processors. A second centre, Colossus 2, opened a year later about 12 kilometres away, with 110,000 GB200 chips and 440,000 GB300 chips.

A planned expansion is expected to add 660,000 GB300 processors by the end of the year. If Musk’s plans succeed, the Memphis complex could eventually run nearly 1.44 million Nvidia AI processors, compared with tens of thousands of Nvidia chips in Jupiter’s Booster system. Musk’s company has rented the capacity of Colossus 1 to AI developer Anthropic for 1.25 billion dollars a month. The companies have agreed that Anthropic will receive about half as much computing capacity again from the new centre.