The headline number from the India AI Impact Summit is large enough to be almost unusable on its own: pledged AI infrastructure investment from Reliance, Adani, Microsoft, Google, and others exceeded $260 billion, spanning commitments through 2035. Adani Group's piece alone is $100 billion over the next decade, aimed at renewable-powered data centers built specifically for AI workloads -- a pledge, not yet a built facility, and pledges at this scale have a long history of slipping or rescoping.
The more trackable number sits underneath the summit headline: between March 2025 and April 2026, operators announced roughly 30 large data center projects across India, adding about 3.5GW of planned capacity. Andhra Pradesh and Telangana alone account for over 2GW of that, concentrated in AI-focused campuses in Visakhapatnam and Hyderabad -- a real geographic concentration bet, not a diffuse national rollout.
On the sovereign-compute side, the IndiaAI Mission has already moved past the announcement stage: over 38,000 GPUs onboarded through a subsidized national compute facility, with a further 20,000 planned and a stated target of 100,000 public GPUs by December 2026, backed by a ₹10,371.92 crore (~$1.25B) government outlay. That's a concrete, dated target with a number already delivered against it, which is a meaningfully different kind of claim than a summit pledge with a 2035 horizon.
CBRE's own projection -- data center investment commitments exceeding $100 billion by 2027 -- sits between the two: bigger than what's built, smaller than what's pledged. The pattern worth watching isn't the $260B topline, it's whether the GPU-onboarding number and the gigawatt-under-construction number keep tracking close to their stated near-term targets (100K GPUs by December, specifically) -- that's the actual leading indicator for whether the rest of the pledge converts to built capacity or joins the long list of AI infrastructure announcements that outran execution.
A $260B summit pledge and a 100,000-GPU-by-December target are different kinds of claims -- the pledge has a 2035 horizon and no enforcement mechanism, while the GPU target is dated, government-owned, and already has 38,000+ delivered against it. Track the near-term, government-owned number as the real signal; treat the multi-decade corporate pledges as directional, not scheduled.