Cohere CEO Warns Renting Frontier AI Equals Digital Serfdom After U.S. Blocks Anthropic Models

Aidan Gomez saw it coming. For years the Cohere CEO prepared for the moment a foreign government would yank access to powerful AI systems. That moment arrived earlier this month when the U.S. Commerce Department ordered Anthropic to cut off foreign nationals from its latest models, Fable 5 and Mythos 5. The company complied immediately. Services vanished for customers worldwide. Governments and enterprises scrambled.

"This sentiment of renting AI from someone rather than owning it is a national security risk," Gomez told Fortune. "You need to fully control it."

The incident exposed a vulnerability many leaders had downplayed. Nations built critical systems on American APIs. Hospitals. Defense contractors. Financial institutions. One policy shift and the plug pulled. Suddenly the conversation shifted from abstract digital strategy to immediate operational risk. And Cohere, the Toronto-based company Gomez co-founded in 2019, found itself flooded with interest.

Its chief AI officer, Joelle Pineau, described a surge. "Huge number of inbounds" from business customers seeking model diversification. Governments outside the U.S. and China expressed particular alarm over future access. They wanted predictable technology stacks. They wanted deployment inside their own walls. BetaKit reported the details just days after the order.

But the stakes run deeper than one export control. The U.S. and China together command 90% of global AI compute, per Epoch AI data cited in the same Fortune piece. Middle powers face a stark choice. Accept dependence. Or invest in genuine alternatives. Gomez frames it bluntly in his own commentary. Nations stand at the edge of an AI industrial revolution yet remain dangerously dependent on a handful of big tech companies. Centralized control creates brittle systems. A monopoly of intelligence is inherently fragile.

The Wake-Up Call From Anthropic's Shutdown

The U.S. action wasn't about a specific jailbreak or safety failure, though Anthropic had applied extra constraints to Fable 5. It was about national security and export controls. Foreign nationals, even Anthropic employees, lost access. The company disabled the models entirely to comply. Other Claude versions stayed available. The message landed anyway. No one enjoys immunity from sudden revocation.

Gomez had warned G7 leaders days earlier. At their summit he pushed for coordinated action among democracies. True digital sovereignty demands choice and control. Who sees your data. Who modifies your systems. Who holds the off switch. Without those, countries slide into digital serfdom. His Fortune op-ed laid out three practical tests. Control over model quality and governance. Strict limits on third-party data access. Operational autonomy across environments, including air-gapped ones.

Yet full independence proves elusive. Frontier training demands enormous capital, energy and chips. Most nations cannot replicate the entire stack. Gomez advocates alliances instead. Back a few trusted champions. Share capacity. Avoid spreading resources too thin. "It's not the case that every country is going to have each layer of the AI stack inside their own country," he explained. "Instead they need to find multiple partners for those layers."

Canada and Germany already tested this approach. Their joint declaration on AI preceded Cohere's planned acquisition of Germany's Aleph Alpha. The deal, valued near $20 billion according to the Financial Times, aims to create a transatlantic player anchored in both nations. Infrastructure matters first, Gomez insists. Each country needs a domestic champion controlling chips, power and data centers. Only then can governments trust that no foreign regulator can intervene.

Recent developments reinforce the momentum. Today, HIVE Digital Technologies announced a major partnership with Bell Canada and Cohere to build Canada's sovereign AI infrastructure platform, combining national networks, enterprise models and domestically built GPU servers. The deal underscores how quickly governments and companies move once risks crystallize. Similar conversations accelerate in Europe, Japan and India. Japan's Sakana AI pushes its Fugu model as a direct challenge to Anthropic's Mythos while advancing local capabilities. Europe hears repeated warnings about technological irrelevance without independent systems.

Analysts at Stanford's Human-Centered AI institute note the definitional challenge. Sovereign AI means different things to different capitals. Some emphasize data localization. Others focus on cultural alignment or regulatory oversight. National security remains the constant thread. Protecting sensitive information. Ensuring defense and intelligence access. Avoiding single points of failure in critical infrastructure.

Cohere positions itself as the practical option. Customers download models and run them on their own hardware. No data leaves the premises. Cohere cannot revoke access even if it wanted to. This on-premises approach contrasts sharply with API-only providers. It introduces friction, yes. Deployment requires expertise. Yet it delivers the control governments now demand. Enterprise buyers noticed the difference after the Anthropic disruption. Inquiries poured in for orchestration platforms alongside the models themselves.

The broader pattern mirrors historical technology races. Nations once competed over railroads, telecommunications, semiconductors. AI adds a new layer. Intelligence itself becomes strategic infrastructure. Dependence on foreign models creates asymmetric power. A U.S. administration can restrict allies as easily as adversaries when priorities shift. The Anthropic order, tied to earlier Pentagon contracting disputes and national security reviews, illustrates how quickly commercial tools turn geopolitical.

But. Alliances introduce their own complexities. Trust between partners must run deep. Technical standards need alignment. Compute sharing requires governance that survives political changes. Gomez acknowledges the difficulty. He draws from his computer science roots. Avoid single points of failure. Design distributed systems. Apply the same principle to international AI cooperation.

So far the response mixes urgency with hesitation. Some governments announce grand sovereign AI initiatives. Others quietly audit dependencies. Companies like Cohere and Mistral gain from the uncertainty. They offer non-U.S., non-Chinese alternatives with strong enterprise focus. Their success will test whether middle powers can sustain competitive frontier capabilities without matching American scale.

The Anthropic episode serves as both warning and opportunity. It woke leaders to structural risks. It boosted demand for controllable systems. Whether that translates into sustained investment remains uncertain. Capital requirements run into tens of billions. Energy constraints bite harder each year. Talent stays concentrated in a few hubs.

Gomez pushes for realism mixed with ambition. Democracies should coordinate. They should back champions. They should treat AI infrastructure with the seriousness once reserved for energy grids or defense networks. "Centralized dependence on a single entity is a structural risk," he repeated. The lesson applies equally to code and to policy. Nations ignore it at their peril.

Today's announcements suggest the message resonates. Canada's latest infrastructure push with Cohere and Bell signals concrete follow-through. Similar moves will likely follow in allied capitals. The race isn't simply about building the smartest model. It's about ensuring your models cannot be switched off when you need them most. Control, after all, defines power in the intelligence age.