The US government’s decision to cut off foreign access to Anthropic’s most advanced models, Mythos and Fable, has triggered a sharp reaction from Zoho founder Sridhar Vembu, who says the episode confirms two things he has believed for a while now: that technology has become the central battleground of national power, and that globalization as the world knew it is finished.

Anthropic disabled both models worldwide after the US Commerce Department issued an export control directive citing national security concerns, barring access to foreign nationals “whether inside or outside the United States.” Anthropic complied within hours, taking the models offline for everyone, including paying customers in India, where Claude has been one of the fastest growing AI products in the country.

Vembu posted his reaction on X soon after the news broke, framing it not as an isolated regulatory hiccup but as a structural shift that Indian policymakers and businesses need to internalise immediately.

“Technology is the ultimate weapon,” Vembu wrote. “National sovereignty, national security, all of it is now about technology.” He followed this with an even blunter line for anyone still hoping things might go back to how they were: “Globalization is dead and Bharat must find her own way ahead.”

What Vembu thinks India should do right now

Vembu’s prescription for the Indian government is fairly direct. He wants Indian organisations to start embracing smaller models, both Indian-built and Chinese open source ones, arguing that with some engineering effort these models can be made to work for most practical purposes. His reasoning is also pointed: why continue paying for access to companies that, as this episode shows, can switch off that access overnight regardless of whether you’re a paying customer.

This isn’t a new theme for Vembu. He has spoken before about India’s software industry facing a reckoning as AI reshapes what software actually costs to build, and he stepped down as Zoho’s CEO last year specifically to focus on AI R&D as the company’s chief scientist. The Mythos and Fable ban, in his framing, is simply the latest data point validating that pivot.

The money problem, and why he isn’t asking Delhi for $100 billion

Where Vembu’s post gets more interesting is in how he addresses the elephant in the room: India simply doesn’t have the money or the hardware to compete at the frontier. He’s upfront about this. Training the kind of models Anthropic, OpenAI and Google are building costs upwards of $100 billion to even be in contention, and on top of that, the high-end GPUs needed for this scale of training are themselves subject to export restrictions.

Vembu’s response to this isn’t to demand the government find tens of billions of dollars anyway. He explicitly says he wouldn’t want that money diverted from other priorities, arguing it has far better uses elsewhere. Instead, he points to the approach Zoho itself has been pursuing: cheaper, alternative R&D paths that don’t require matching the GPU budgets of US labs dollar for dollar. He acknowledges this kind of frontier research takes time, but says he’s confident India will get there.

This lines up with what Sarvam AI, the Bengaluru-based lab backed under the IndiaAI Mission, has also been attempting. Sarvam’s 105B parameter open source model currently ranks sixth globally by country on the Artificial Analysis Index, a result Vembu would likely point to as proof that smaller, focused efforts can still produce something usable, even if they’re nowhere near matching Mythos or Fable on raw capability.

A message for the “globalization will return” crowd

Vembu closed his post with a line aimed squarely at anyone in India still operating under the assumption that the old order of cheap, frictionless access to the best American technology will resume once the current turbulence passes. Anyone harbouring such delusions, he said, should wake up now.

The timing makes the message land harder than it might have a few months ago. Anthropic had been aggressively expanding in India, opening a Bengaluru office and calling India its second-largest market for Claude, with CEO Dario Amodei visiting the country as recently as February to talk up partnerships with companies like Infosys. For a government directive in Washington to be able to instantly switch off access to a company’s flagship models for an entire country’s worth of users, paying customers included, is exactly the kind of dependency Vembu has been warning about.

Whether Indian organisations actually start shifting toward smaller open source models in any meaningful numbers remains to be seen. But Vembu’s post adds to a growing chorus of voices, including Yann LeCun’s recent suggestion that India might be better served exploring approaches beyond the standard LLM race, arguing that the country’s AI strategy can’t simply be a smaller, slower copy of what the US labs are doing. The ground has shifted, and Vembu wants India to notice before it’s forced to.