Blackstone just dropped $1.2 billion into India’s biggest AI infrastructure bet — and half of it is borrowed money. The timing wasn’t subtle. The announcement landed February 16, 2026, the opening day of India’s AI Impact Summit, as the country tries to build sovereign compute independence while running on Wall Street debt.
The deal splits evenly: $600 million in equity from Blackstone and co-investors, plus $600 million in planned debt, valuing Neysa at $1.4 billion. That’s a 24x jump from the company’s $50 million prior raise. The capital funds a 16.6x GPU scale-up — from 1,200 live units to over 20,000 GPUs across Indian data centers within 9 to 12 months. That’s not growth. That’s a construction project financed like a leveraged buyout.
Here’s the tension: India wants AI independence, but it just took $1.2 billion from a US private equity giant to build it. And the buildout only works if chip supply chains cooperate.
India’s sovereign AI bet runs on Wall Street debt
Most US hyperscalers fund GPU deployments with operating cash flow. Neysa is doing it with a 50/50 equity-debt structure. If global chip shortages delay the buildout — and Neysa CEO Sharad Sanghi told TechCrunch the company needs “supply chain resilience” to access chips quickly — the debt service starts before the GPUs arrive.
Translation: they don’t control the supply chain.
Blackstone’s Ganesh Mani called digital infrastructure “one of our highest conviction investment themes globally” in a statement. But conviction doesn’t ship Nvidia H100s. And India’s current GPU count sits at just 60,000 units, with projections to hit over 2 million short-term, according to Economic Times. That’s a 33x infrastructure gap. The market opportunity is real. The execution risk is enormous.
Blackstone’s Amit Dixit framed Neysa as “picks and shovels” for India’s AI buildout. But picks and shovels still need raw materials. If Nvidia’s GPU allocation favors hyperscalers and sovereign projects with government backing — as Nvidia’s infrastructure priorities show — allocation doesn’t guarantee delivery timelines that match debt covenants.
The sovereign cloud pitch nobody can verify
Neysa’s Velocis platform promises something every Indian enterprise wants: AI compute that stays in India, with data locality guarantees US hyperscalers can’t match. The company’s pitch deck claims cost advantages over AWS, Azure, and Google Cloud. But no named customers. No public before/after migration data. No TCO breakdowns showing power, cooling, or real estate costs per GPU.
India’s push mirrors sovereign AI strategies elsewhere, but most of those bets involve domestic chip production, not imported GPUs on borrowed capital. Neysa serves financial services, technology, healthcare, and public sectors, according to Blackstone’s announcement. But which banks? Which hospitals? The evidence is a press release.
The AI infrastructure race isn’t just about models — it’s about who controls the compute. India just signaled it’s willing to take on debt to compete. Whether that’s strategic vision or financial overreach depends entirely on whether those 20,000 GPUs show up on schedule.
What happens when the chips don’t arrive on time
Neysa’s CPO Karan Kirpalani told the India AI Impact Summit the company is “democratizing AI” for enterprises worried about data sovereignty. That’s the pitch. Here’s the reality: if the GPU buildout slips by six months, debt service starts before revenue scales. Indian enterprises waiting for sovereign compute either stay on AWS and Azure or delay AI projects entirely.
This isn’t FUD. It’s the reality of building data center infrastructure in a GPU-constrained world. Sanghi himself acknowledged the dependency: “supply chain resilience” is the phrase executives use when they mean “we’re at the mercy of Nvidia’s allocation priorities.”
Infrastructure bets assume everything goes right. They rarely do.
Two quotes, no editorial needed. Blackstone’s Mani: “Digital infrastructure is one of our highest conviction investment themes globally.” Neysa’s Sanghi: “We need supply chain resilience to access chips quickly.” The reader can sit with that contradiction.
