Nvidia’s $164.5 Billion Backstop Is Demand Finance

1. The $164.5 billion backstop is a credit instrument, not a buildout

The $164.5 billion of data-centre guarantees and commitments on Nvidia's books represents vendor-financed demand creation rather than a funded Indian buildout pipeline. It shifts counterparty exposure onto Nvidia while preserving the flow of GPU orders. At the top end, $105 billion backs lease obligations for the Ohio campus that SB Energy is constructing for OpenAI; should OpenAI miss rent, the landlord can seek payment from Nvidia. The remainder consists of $36 billion in commitments to purchase unsold cloud capacity, $20 billion in direct data-centre leases that are passed on to customers, and $3.5 billion in guarantees covering cloud-provider leases. Nvidia has also committed more than $7 billion directly to AI cloud providers through investments or pledges. For lenders, the relevant credit is increasingly Nvidia itself rather than the ultimate operator.

The same pattern is already playing out in India. Yotta requested 80,000 GPUs from Nvidia on 11 September, split across two tranches in a $12 billion order, and may enter Nvidia's AI Compute Partnership, which pays a predetermined rate for unused GPUs and thereby gives financiers a revenue floor. Most of this capital originates abroad. Mubadala and wealthy individuals support Yotta; AM Intelligence's $1.5 billion chip purchase is financed by Japanese bank loans. Indian lenders typically extend credit against land and power assets, not against GPUs. Backstops do not eliminate credit pressure. CoreWeave debt has traded at distressed levels, Microsoft terminated an Nscale site deal, and Firmus withdrew its IPO. A guarantee can get a financing round done. It cannot show that an end customer will pay inference prices that recover the capital.

2. Delhi-NCR's first binding constraint is nodal power, not national generation

India's installed data-centre power draw has expanded from 375 MW in 2020 to 1,575 MW, prompting the Central Electricity Authority to set up a committee on grid stability because these sites impose large, round-the-clock loads. The signal is genuine demand, not merely a project blueprint. One CEA estimate puts data-centre demand at about 17 GW by 2031-32; later capacity projections reach 24-26 GW, and S&P Global reports a parliamentary figure of 26.3 GW for FY2031-32. Announced investment cannot be treated by planners as simultaneous load at a specific node. National resource adequacy does not resolve local transmission and distribution constraints. Several hundred megawatts at a single location poses a different set of problems than the same consumption dispersed across millions of retail users.

Delhi-NCR makes that abstraction concrete. Adani is pursuing a distribution licence for its 80 MW Noida data-centre park, which would let it pick off a dependable, high-value customer from a discom already carrying heavy debt. Securing a dedicated high-tension grid hook-up can require 12-24 months, so power procurement discussions now begin years ahead of construction. Power has overtaken land as the decisive siting factor. The central question is not whether India has enough aggregate generation. It is whether several hundred megawatts can be delivered at the correct node, at the correct moment, under a regulatory structure that still treats distribution as a public duty.

3. Water, heat, and tax treatment decide whether Delhi-NCR clears the hurdle rate

A given Delhi site can move between viable and unviable even if AI demand does not change. With the cloud-services tax holiday introduced in the Union Budget 2026-27, the five leading hub states all show a positive baseline net present value when discounted at 9 per cent, and three clear the 12 per cent infrastructure hurdle. Under ordinary corporate tax treatment, no state meets the hurdle, and Delhi-NCR goes negative. That shift is fiscal, not physical. Monetising water costs and adding climate and regulatory risk premia lowers baseline NPV by 11-47 per cent in four states, pushes Delhi-NCR into unviability, and rearranges the hub rankings. The tax holiday can make a poor site look attractive on a spreadsheet-that is exactly its purpose, and its risk.