🧠 Food for thought

Implications, context, and why it matters.

The B200 tender rate sits near older hardware in India

• IndiaAI set a benchmark of 290.7 rupees (US$3.1) an hour for a new B200 graphics processing unit (GPU). That is only a little above the Indian market rate for the older H100 at about 249 rupees (US$2.6) an hour 1.
• The benchmark also comes in below the current on-demand rate for the H200, which rents for about 300.14 rupees (US$3.2) an hour 2.
• Questions about long-term viability stem from steep purchase costs in India. Import duties add a 25-30% premium, which lifts the price of one H100 to 30 rupees (US$0.32)-40 lakh 1.
• The tender matters for local supply because Amazon Web Services (AWS), Microsoft Azure, plus Google Cloud do not offer newer GPUs such as the H100 in their Indian data center regions 1.

Standard pricing gives AI compute a utility-style setup

• India’s plan sets one hourly rate and removes network ingress plus egress fees, which makes AI compute work more like a utility.
• Large cloud providers often charge more for GPUs to support their wider platforms. Network use and data transfer can also raise the bill, while this tender sets both ingress plus egress at zero 3.
• For Indian AI startups, GPU spending can take 40-60% of an early technical team’s budget. Lower, predictable compute costs can ease entry and let teams spend more time on product work instead of fundraising 3.
• If the model works, this state-backed, multi-vendor system could offer a template for countries that want local AI capacity without relying on the pricing power of large US cloud companies.