The bank’s June 2026 Eye on the Market report, titled “Semiquincententacles,” tracks AI readiness across four separate indices drawn from Stanford University, Tortoise Media, Oxford Insights and the International Monetary Fund. Each index measures something different, and together they offer a more complete picture of a country’s AI standing than any single ranking can.
Where does India stand
India ranks third on the Stanford AI Vibrancy Index from November 2024, behind the US and China. That index measures research and development, economic readiness, infrastructure, policy and governance, responsible AI practices, education, public opinion and diversity.
But the other three indices tell a different story. On the Tortoise Global AI Index from September 2024, which measures practical implementation, innovation and investment, India falls to tenth. On the Oxford Government AI Readiness Index from 2025, which tracks how prepared governments are to deploy AI in public services, India ranks fourteenth. And on the IMF’s AI Preparedness Index from 2023, which covers digital infrastructure, human capital, innovation and regulation, India places last among the fifteen countries listed, behind China, the UAE, Spain and France.
The spread of years across the four indices is worth noting. The IMF measure, now three years old, predates India’s more recent push on digital public infrastructure and AI policy frameworks, and may understate where the country stands today.
But even allowing for that, the pattern across all four is consistent enough to draw a broader conclusion. India has the raw material for AI: the people, the interest, the policy intent, and enough of a research base to register strongly on the Stanford index. What it does not yet have in sufficient measure is the implementation layer: cloud infrastructure, business adoption, government deployment capability and the digital foundations that convert AI potential into economic output.
What the US-China contest looks like from above
JP Morgan’s core finding on AI is that the US remains the dominant player by almost every significant measure, but China is closing the gap faster than most observers acknowledge.
Across all four indices, the US leads comprehensively. It tops the Stanford vibrancy index, the Tortoise Global AI Index and the Oxford Government AI Readiness Index, and places second on the IMF Preparedness Index, behind Singapore. China ranks second on the first two measures, seventh on the government readiness index and thirteenth on the IMF index, suggesting that China’s AI strength is concentrated in private sector and research activity rather than government deployment capability, a mirror image of the US pattern in that last measure.
But that dominance comes with a structural vulnerability that JP Morgan describes as the United States’ Taiwanese Achilles heel. Eight of the ten largest companies in the world by market capitalisation depend significantly on supply from TSMC, the Taiwanese chipmaker, with more than one third of their combined two trillion dollars in revenue coming from hardware that uses TSMC products.
Global trade in semiconductors and integrated circuits has grown steadily and recently exceeded global trade in crude oil. The US is working to reduce that dependence, with Commerce Secretary Howard Lutnick having set a goal for the country to onshore 40 per cent of semiconductor demand by the end of Trump’s term in 2028. Even if that target is met, JP Morgan estimates the US would still be only 30 to 35 per cent self-sufficient in advanced node production by the end of the decade, leaving it heavily reliant on an island that China encircles with naval assets on what the report describes as a routine basis.
Frontier model development
JP Morgan data, sourced from independent AI benchmark entity Artificial Analysis, shows US and Chinese frontier model scores converging. As recently as 2022, there was a visible gap. By mid-2026, Chinese models are within a few points of their American counterparts on composite benchmarks covering mathematics, science, coding and reasoning. When the comparison shifts to intelligence per dollar, Chinese models dominate, with DeepSeek, MiniMax, Kimi, Xiaomi and Alibaba sitting firmly in what JP Morgan calls the green quadrant, defined as maximising intelligence per dollar. Most US frontier models from Anthropic, OpenAI and Google score higher but at significantly higher operating costs.
The shift to open models
One of the more consequential observations in the report concerns what happens when frontier model access becomes too expensive for ordinary enterprise use, and the evidence that this tipping point may already be arriving.
Frontier labs are shifting to usage-based token pricing, driven by extraordinary capital commitments. Anthropic contracted for 8.5 gigawatts of computing capacity in April 2026 alone. The cost is being passed on: OpenAI doubled token prices between two consecutive model generations, Microsoft raised Copilot prices from June 1, and some enterprise users reportedly saw token costs rise by as much as 100 times.
The response has been predictable. JP Morgan cites the founder of Lindy AI, who moved the company’s entire AI service from Claude to DeepSeek, claiming savings of millions of dollars. Brian Armstrong of Coinbase publicly stated that 80 per cent of enterprise workloads would shift to significantly cheaper models within a year.
The report also flags a more fundamental point. A smaller open model trained on a company’s own proprietary data may outperform a frontier model for that company’s specific tasks. Ramp, a financial software company, demonstrated that a Chinese model with 35 billion parameters outperformed Anthropic’s Opus 4.6 on its specific financial data tasks at a fraction of the cost.
This is arguably the most directly actionable insight in the report for Indian enterprises. The assumption that serious AI adoption requires expensive frontier model access has been one reason many Indian companies have treated large-scale AI deployment as a future aspiration.
A well-resourced Indian enterprise sitting on years of proprietary operational data, whether a bank, a hospital network, an insurer or a logistics company, may not need to wait for frontier model costs to fall. It may already have what it needs to build something competitive on open-weight infrastructure, at a fraction of what frontier access would cost.
What else does the report cover
China’s growing chip self-sufficiency
China’s GPU self-sufficiency has risen from 10 per cent in 2021 to 40 per cent today, with projections reaching approximately 80 per cent by 2030, according to Morgan Stanley Asia Technology data cited in the report. JP Morgan flags that Huawei may have found a way to scale compute density through vertical chip logic design rather than shrinking transistors, allowing it to remain competitive despite lacking access to ASML’s extreme ultraviolet lithography tools. If that claim holds up, China’s AI hardware trajectory looks considerably less constrained than western export controls were intended to ensure.
The Nvidia question
Nvidia’s share of AI accelerator revenue has declined from 85 per cent in 2023 to an estimated 75 per cent in 2026, according to Silicon Analysts data. That is still an extraordinary degree of market dominance, but the direction matters. Custom chips from Google, Amazon, Meta and Microsoft are eating into that share, with hyperscalers reporting total cost of ownership reductions of 30 to 40 per cent compared to merchant Nvidia GPU fleets. JP Morgan cites Anthropic’s decision to run Claude on Amazon’s Trainium chips for the next decade as the strongest third-party endorsement of custom chip economics to date.
For India, this has specific relevance. Indian enterprises and startups building AI products currently depend almost entirely on Nvidia hardware. The cost structure of that dependence is significant, and for smaller companies trying to build AI-native products, it remains a genuine constraint.
