As Britain's business leaders grapple with the escalating costs and growing unease over data control, they're increasingly prioritising 'AI sovereignty' – a concept that's as simple to understand as it is vital for their bottom line. In essence, AI sovereignty refers to the degree of control an organisation has over its own artificial intelligence destiny. And right now, that control is under threat from the prohibitive costs and data security concerns surrounding proprietary AI models.
The financial burden of these high-stakes systems is becoming increasingly unsustainable. Reports suggest some large tech companies have blown through their AI budgets by hundreds of millions in a single quarter without seeing any tangible business benefits. It's not that AI itself is useless – it's the economic model that doesn't quite add up, with hefty upfront costs and the need for extensive human oversight outweighing the value generated.
The rising costs are pushing businesses towards open-source models, particularly those from China. According to data from OpenRouter, an AI model access platform, there's been a seismic shift in usage patterns: Chinese models were used one-third as much as US models back in January, but by July they'd surged to five times higher. The main driver of this change? Cost – with Chinese models often being more than ten times cheaper. Take the recently released open-source Kimi-3 model from Chinese AI company Moonshot, for example: not only is it free, but it's also reportedly outperforming Anthropic's latest Fable-5 model across numerous benchmarks – a model that's been branded 'too dangerous' for widespread use and is costly to run.
But it's not just the wallet that's driving this shift; concerns around data sovereignty are equally pressing. Distrust of major US frontier AI labs such as Anthropic and OpenAI has long been present in China and Europe, but now this sentiment is spreading rapidly among US enterprises too. Business leaders are questioning the credibility of contractual clauses promising 'no training on client data', especially given the widely held belief that generative AI models were initially built through extensive use of copyrighted material. This erosion of trust is significant – with some executives fearing these labs might be 'stealing their alpha' – a term used to describe proprietary business insights.
The situation for companies like Anthropic has become more complex following allegations, albeit geopolitically charged, that its models contained surveillance-related backdoors. While the veracity of such claims remains difficult to confirm, the damage to trust appears to be done. For UK businesses, this means a heightened awareness of where their data resides and how it's used – and a growing recognition that AI sovereignty is no longer just a nicety, but a necessity.