NEW DELHI — In four years, the startup in Bengaluru that routes its AI jobs through servers in Frankfurt will be able to run the same workloads on Amazon’s custom AI chips from a building in Mumbai. That is what Andy Jassy flew to New Delhi to announce on Wednesday.
After meeting Prime Minister Narendra Modi at Seva Teerth in the capital, Amazon’s chief executive committed the company to investing $48 billion in India between 2026 and 2030, adding a fresh $13 billion to the $35 billion Amazon had pledged in December. The expansion will bring Amazon Web Services data centers, custom AI accelerators, and managed AI services to India’s two largest technology cities. Since Amazon entered India in 2013, its cumulative commitment to the country will exceed $88 billion by the end of the decade, the company announced.
Modi described the commitment as a historic one. “I welcome Amazon’s record $48 billion investment in India,” he said after the meeting, adding that the investment would “create new opportunities for our youth,” a framing consistent with what India’s government has pitched to every technology executive who has made the trip to Seva Teerth this year.
There have been several of them. In the weeks before Jassy landed in New Delhi, three other companies made comparable trips and comparable pledges. Google committed $15 billion to India’s AI infrastructure. Microsoft’s $17.5 billion investment in Indian data centers is already in its planning cycle. Australian firm AirTrunk, backed by BlackRock, announced $30 billion for five gigawatts of data center capacity by 2030. India is receiving more than $100 billion in hyperscaler investment in a single calendar year, a concentration of infrastructure capital without precedent in any emerging market. The race for India’s cloud market has become something closer to a land rush.
Amazon’s contribution, the largest single commitment, will specifically expand AWS capacity in Mumbai and Hyderabad. The expanded footprint will bring what Jassy described as “custom AI chips, managed AI services, secure and reliable cloud technologies, and developer tools” within reach of Indian startups, enterprises, and government agencies that have previously operated with high-latency access to servers on the other side of the world. The announcement arrived the same week OpenAI and Broadcom unveiled their first custom inference chip, underscoring how contested the market for AI accelerator access has become.
Amazon’s scale ambitions extend beyond cloud computing. By 2030, Jassy said the company plans to support 3.8 million jobs and enable $80 billion in e-commerce exports from India, while extending AI access to 15 million small businesses and 4 million students in government schools. Amazon will open more than 20 new fulfilment centers and more than 100 last-mile delivery stations, expanding its logistics reach into Tier 3 and Tier 4 cities where the next wave of Indian e-commerce demand is concentrated.
“The response has been tremendous,” Jassy said, “with strong growth especially across our ecommerce, AI, and cloud businesses. As we grow Amazon in India, our business priorities continue to align with India’s priorities of democratizing access to AI, digitizing small businesses, creating jobs, and enabling exports. We are investing over $48 billion in the coming five years to meet the strong demand across our business in India.”
What the announcement does not resolve is a structural question India raised with Washington in the same week. Earlier this month, a U.S. Commerce Department directive required Anthropic to shut down its most capable AI models for all foreign users simultaneously, including Pax Silica member countries, without warning. India’s technology ministry raised the episode at the Pax Silica Summit in Washington and secured what its technology secretary called an assurance that access would not be abruptly revoked once granted to trusted partners. As India’s AI access assurance at the Pax Silica Summit showed, that understanding remains verbal, with no written commitment signed.
Amazon’s custom AI chips, the hardware it plans to bring to Mumbai and Hyderabad, fall precisely within the technology categories U.S. export control policy has been tightening around since late 2025. Amazon has not disclosed whether the specific chips and AI services it intends to deploy for Indian government and enterprise customers have received Commerce Department export classification review, or whether any of the models they will run are subject to directives of the kind that cut off Anthropic access last month.
Five-year infrastructure pledges in emerging markets carry a familiar history of revision. Land acquisition delays, power grid constraints, and regulatory frictions that do not appear at prime ministerial announcement ceremonies have slowed data center buildouts in India before. Amazon said it is investing $48 billion over the coming five years. That is a plan, not a deployed facility, and the gap between the two is where most of the interesting questions live.
What is not in doubt is the direction of the bet. When Google, Microsoft, AirTrunk, and Amazon all commit over $100 billion to the same country in a single year, the question of where the world’s next significant AI infrastructure market lies has been settled. Indian developers and government services have been running AI workloads under latency constraints their Silicon Valley counterparts have not faced. Whether the hardware arriving in Mumbai and Hyderabad reaches those users on the terms Jassy advertised, or on terms Washington’s export control system reserves the right to revise, is the test the next four years will run.
