The news hit the markets like a thunderclap: Anthropic, the AI safety darling of Silicon Valley, has crossed the $1 trillion valuation mark. For those of us who track these things, it’s a validation of the thesis that intelligence is the new oil. But there is a darker current beneath the headline, one that should chill every policymaker in London. Without a sovereign chip strategy, Britain will not merely fall behind. It will become a digital vassal state.
Anthropic’s rise is emblematic of a deeper structural shift. The company’s success is not just about clever algorithms. It is about access to the world’s most advanced chips. The new generation of AI models requires clusters of thousands of specialised processors, known as GPUs and TPUs, to train and run. These chips are the picks and shovels of the AI gold rush. And currently, the supply is controlled by a handful of actors, most of them outside the UK.
The valuation itself is staggering when you consider that Anthropic was founded just three years ago. It has leapfrogged decades-old industrial giants. But the real story is not the number. It is the infrastructure that makes the number possible. The company’s latest model, Claude 4, is rumoured to have been trained on a cluster of over 100,000 Nvidia H100 chips. That is a cluster the size of a small data centre, consuming enough electricity to power a town.
Britain’s position in this new landscape is precarious. We have world-class AI research labs: DeepMind in London, the Alan Turing Institute, and a thriving startup scene. But we do not have the chips. Not the advanced ones needed for frontier training. Our sovereign compute capacity is a fraction of what the US or China can deploy. And geopolitical tensions mean that access to US-made chips is not guaranteed.
The Government’s recent announcement of a £100 million sovereign AI compute fund is welcome, but it is pocket change. To build a cluster comparable to what Anthropic uses would cost billions. And it is not just about money. It is about supply chains, energy infrastructure, and talent. The UK has the talent but lacks the hardware ecosystem. We do not have a domestic chip fabrication plant capable of making advanced AI processors. The few fabs we have are focused on legacy nodes.
Some will argue that we do not need to build our own chips. We can rent them from cloud providers. But that is like arguing that Britain did not need its own coal mines during the Industrial Revolution because we could import from Belgium. It misses the point of sovereignty. If we cannot train our own frontier models on our own hardware, we will be forever dependent on decisions made in Washington or Beijing. And those decisions will not always align with our interests.
There is also the question of privacy and security. The UK government wants to use AI for everything from healthcare to defence. Do we really want patient records or military plans being processed on chips controlled by a foreign power? The logic of data sovereignty is inescapable. It applies to the silicon as much as to the software.
The good news is that there are paths forward. The UK can invest in new chip architectures, such as those from Graphcore or Cerebras, which are less reliant on the dominant Nvidia ecosystem. It can form alliances with allied nations, like Japan or South Korea, to build shared fabrication capacity. It can also rethink its approach to energy, building small modular reactors to power the data centres of the future. But all of these require action now, not after the next election cycle.
The Anthropic valuation is a wake-up call. It signals that the AI race is entering a new phase, one where hardware dominance will determine software outcomes. Britain has a proud history of innovation, from Babbage to the World Wide Web. But that legacy will count for nothing if we fail to secure the silicon that underpins the future. The time for committee reports and pilot projects is over. We need a sovereign chip strategy, and we need it yesterday.
