The sovereignty trap: how emerging market AI infrastructure can mislead you

Sarvam launched Token Factory as India’s sovereign AI answer, but 6 of its 12 models are Chinese. The paradox matters far beyond India: European SMBs evaluating “regional” AI infrastructure face the identical trap—geographic independence is not the same as strategic independence.

Sarvam, the Bengaluru AI startup positioning itself as India’s sovereign AI champion, launched Token Factory on November 16 as an inference platform running on locally-procured GPUs. Smart move on paper. And then it turned out that 6 of its 12 available models were Chinese—including DeepSeek, Qwen, Kimi, and GLM.

The backlash was predictable. Sarvam co-founder Vivek Raghavan had spent months warning that Big Tech was “dumping” free models on the market. Now his company was reselling DeepSeek—which Token Factory users are running at a scale that dwarfs Sarvam’s own model, with DeepSeek handling 450+ billion tokens daily. The irony is so clean it stings.

But here’s the uncomfortable part: Sarvam is not wrong about Token Factory being sovereign in a narrow sense. Data stays in India. The GPUs are independently procured—no government subsidies (those ended in March 2026). The infrastructure is real, domestic, and competitive on cost. The company isn’t lying about data residency or technical sovereignty.

They’re just revealing that geographic sovereignty and strategic sovereignty are not the same thing.

Why this applies to your vendor choices

If you work at a 40-person distributor, a 200-person manufacturer, or any mid-market operation in Europe evaluating alternatives to US cloud vendors, Sarvam’s contradiction should scare you more than it scares Indian startups. Because you face the exact same trap, except the pressure is higher and the marketing is slicker.

A vendor pitches you a “European alternative” to AWS or Azure. Data stays in Frankfurt. The company is founded in Amsterdam. You feel like you’re taking control. What you might actually be doing is swapping one dependency (US compute, US SaaS) for another (European infrastructure running OpenAI or Claude models). You’ve gained data residency. You haven’t gained independence.

The question that matters is not “where is my data?” It’s “whose models am I building my business around, and what happens if that relationship changes?” Geography answers only the first part.

Sarvam had to resell Chinese models because—as co-founder Pratyush Kumar explained—revenue from inference services is now funding their own model development after subsidies dried up. It’s economically rational. It’s also strategically hollow for Indian sovereignty. And it’ll be strategically hollow for you if you pick a “regional” infrastructure provider without understanding whether that provider is building toward independence or just serving as a margin-thin middleman for existing US or Chinese models.

This is especially true in Europe, where you have the added layer of GDPR friction. A German SMB customer of yours asks if their data could end up in a US system. A “European” infrastructure provider says no—it stays local. That’s true. But if that platform is running GPT-4o and feeding usage telemetry back to OpenAI, you’ve solved the residency problem while creating a new one: your customer’s operational data is in Germany, but their strategic model dependency is in Redmond.

The pattern emerging markets reveal about infrastructure plays

What Sarvam is experiencing is the model-compute asymmetry that characterizes emerging market AI infrastructure. India has strong systems talent, can build competitive inference platforms, and can operate infrastructure as cheaply as anyone. What India does not yet have—and what Reliance and others are still chasing—is competitive frontier models. The gap between “we can run inference on GPUs in Delhi” and “we control the model that matters” is wider than any CEO wants to admit.

That gap is real for European SMBs too, just inverted. You have access to the best frontier models in the world (OpenAI, Anthropic, DeepSeek, etc.). What you don’t have is leverage over where they run or how they’re governed. A “regional” provider can promise you the second without delivering the first.

Your move, if you’re serious about independence, is to verify three things before signing: (1) Are my queries and data actually staying within the geography I need? (2) Which models does this platform commit to supporting, and how locked am I into that set? (3) If the provider’s underlying model partners change—say, OpenAI pulls its API access—can I migrate my workloads in 30 days or is it a six-month rebuild?

Sarvam would have been on firmer ground saying plainly: “We’re building domestic inference infrastructure. We’ll host any model that runs. We’re funding our own model development. When we have competitive models, we’ll deprioritize reselling others.” That’s honest. What they can’t do—and what most “regional” vendors are doing to you—is claim sovereignty while outsourcing the thing that actually matters.

What European SMBs should actually verify

Before you pick a “regional” AI infrastructure vendor over a US one, ask yourself: Am I buying geography, or am I buying strategic independence? Because Sarvam just proved you can win at the first and still lose at the second.

The honest answer for most European mid-market firms is that you’re buying geography and regulatory compliance, which is valuable and real. But call it what it is. Don’t let a vendor’s sovereignty language obscure the fact that you’re still strategically dependent on whoever trained the models you’re running. Data residency is a solved problem. Model independence is not. And the moment you treat them as the same thing, you’ve bought into the same paradox that Sarvam is now publicly wrestling with.

Share on LinkedIn


0 Comments

Leave a Reply

Avatar placeholder

Your email address will not be published. Required fields are marked *

MARKET INSIGHTS

Explore All Our AI & Market News

Daily insights on AI and global business — what’s changing, and what it means for your growth.

See All Articles →
GET IN TOUCH

Ready to Grow Demand or Save Time?

Tell us what you’re trying to grow, or what’s slowing you down. We’ll respond within 48 hours with a clear plan of action.

No commitment. No pitch deck. Just a direct conversation about what we can build together.

LOCATION
Paris, France – Poznan, Poland
Paid Media
Search Display Social Ads Reporting Tracking Strategy
Business Solutions
Business Solutions (RDZ Systems)