Money: Who Is Spending, and How the Numbers Should Be Read
The single most reliable cross-country source on AI spending is Stanford University's AI Index. Its 2026 edition reports that private investment in AI in the United States reached about 285.9 billion dollars in 2025, roughly 23 times China's reported private figure of 12.4 billion dollars. Taken at face value, the gap is humungous. But the same report warns plainly against reading it that way.
China's spending runs mostly through the state, not private venture funds. Stanford highlighted that Chinese government guidance funds invested an estimated 184 billion dollars into AI firms between 2000 and 2023 alone, which does not appear in the private tally. Bank of America estimated China's own 2025 AI capital spending at 84 to 98 billion dollars for that year.
So, the honest framing is this: the US leads decisively on private risk capital. China competes through directed state money and corporate build-out. The EU is mobilising public funds to pull in private investment. Historically, India has seen low investments. But, this has recently started to turn around with both government and private capital expenditure picking up in the sector.
India's flagship government programme, the IndiaAI Mission, carries an outlay of Rs. 10,371.92 crore, roughly 1.25 billion dollars, over a five-year period. The EU's InvestAI initiative aims to mobilise 200 billion euros. China has drafted a plan to spend around 2 trillion yuan (about 295 billion dollars) on data centres over five years. The US government barely needs to intervene because its private sector alone is set to spend about 725 billion dollars on AI in 2026. India's government money is modest, but it has been used as seed capital to pull in private players rather than to do the heavy lifting itself, a deliberate public-private design.
When the Race Began, and Where the Research Sits
The US has funded AI research since the 1950s and its commercial boom dates back to 2012. China started with the July 2017 Next Generation AI Development Plan, aiming to lead the world by 2030. Europe's coordinated push followed in 2018, with France and Germany committing significantly.
India published its National Strategy for AI through NITI Aayog in 2018. But the serious money and compute arrived only with the IndiaAI Mission, approved in March 2024. By that measure India is several years behind on state-backed effort.
On research output, China now leads the world in the sheer volume of AI papers, citations and patents, while the US still produces the highest-impact research and the most top-tier models. Stanford counts 59 notable AI models from the US in 2025 and 35 from China; Europe produced a handful.
India's research strength is real but narrower. Its premier institutes, the IITs and the IISc, are active contributors. IIT Bombay leads the government-funded BharatGen consortium that built Param-2, a multilingual model. India is the second-largest contributor to AI projects on the GitHub platform, and the Stanford AI Index places it among the top four countries for AI skills, talent and policy. The gap lies not in brains; it is in compute, capital and frontier-scale models.
Sovereign AI, Data Centres and the Chip Question
Sovereign AI, the ability to build and run advanced AI on your own soil with your own infrastructure, is where India has moved the fastest. The IndiaAI Mission set a target of 10,000 graphics processing units (GPU), the specialised chips that train AI. It has already deployed over 38,000, offered to startups and researchers at a subsidised rate of Rs. 65/hour.
In February 2026, at the India AI Impact Summit, the Bengaluru startup Sarvam AI released two home-grown large language models (LLMs) of 30 and 105 billion parameters, trained on Indian compute and tuned for 22 Indian languages. It marked India's formal entry into the foundation-model club.
The private sector has committed to making substantial investments in data centres. Reliance has pledged Rs. 10 lakh crore, about 110 billion dollars, over seven years. Adani Group has pledged 100 billion dollars by 2035, alongside Google, Microsoft and an OpenAI-Tata tie-up.
The one stubborn dependency is silicon. India's AI today runs almost entirely on Nvidia chips imported from the United States. The risk of that dependency was made vivid on 12 June 2026, when Washington directed Anthropic to suspend its most advanced Fable and Mythos models for all foreign nationals overnight. If software access can be cut that quickly, hardware can be too.
India's answer lies in its Semiconductor Mission, now in its second phase, plus indigenous chip efforts: ISRO's Vikram processor for space, IIT Madras's SHAKTI family built on the open RISC-V design, and a clutch of startups working towards AI-specific chips for data centres. None yet rivals Nvidia, and officials concede indigenous AI-grade GPUs are still a few years away. But, the efforts matter. The Ministry of Electronics and Information Technology (MeitY) is the engine behind all of this, running the compute roll-out, the foundation-model selections, the skilling programmes and the chip mission together.
The Gap, the Speed, and the Bubble Worry
How far has India come? Or, how behind is India? The gap between what others are doing and what India has undertaken remains significant when it comes to government spending. But, India's lean model narrows its practical effect.
On private capital the gap is the largest, though the corporate pledges in 2026 are closing it fast. On compute and applied models the lag is perhaps shrinking quickly. On frontier research and indigenous chips, it is wider – five years or more. India's distinctive advantages, language models, low-cost engineering, and a digital public infrastructure of 1.4 billion Aadhaar identities and over 12 billion monthly UPI transactions, give it a route that does not require matching Silicon Valley.
Finally, the bubble question. Some respected analysts warn that US and Chinese AI spending has outrun real returns. An oft-cited finding suggests roughly 95 per cent of company AI pilots show no clear profit yet, and AI capital spending as a share of sales has passed dot-com-era peaks.
Others, including Goldman Sachs and Wall Street desks, point to genuine cloud-revenue growth and large order backlogs as evidence that the returns are real but simply lag the spending.
The truth sits in between. The productivity gains are measurable but modest so far. One has to closely watch how today's enormous build-out pays. For India, arriving later may prove a quiet advantage, building capacity as costs fall rather than at the peak of the frenzy.
No comments yet. Be the first to share your thoughts!