Stanford's AI Index Proves the US Can't Buy Its Way to an AI Lead

The Stanford HAI AI Index 2026 landed last week with 423 pages and a finding that should stop every AI investment committee in its tracks: the performance gap between the best American and Chinese AI models has collapsed to 2.7 percent. Down from 17.5 to 31.6 percentage points in May 2023. The United States invests 23.1 times more in private AI than China does — $285.9 billion to $12.4 billion. And the gap is still closing.

This is not an AI research paper. It is the most comprehensive annual audit of the global AI landscape, covering investment, performance, talent, adoption, safety, and policy across 15 chapters. And its central message is uncomfortable: the dominant American strategy of outspending the competition is producing diminishing returns at every level — from national competitiveness to enterprise deployment.

The spending gap that produced a 2.7% lead

The numbers are staggering and worth sitting with. US private AI investment in 2025 reached $285.9 billion. China's was $12.4 billion. Global corporate AI investment hit $581.7 billion, up 130 percent year over year. Generative AI investment alone surged nearly fivefold to $170.9 billion.

Big Tech capital expenditure has nearly tripled from $162 billion in 2022 to $448 billion in 2025, and 2026 guidance from the hyperscalers totals roughly $635 to $665 billion. Meta alone plans to spend $115 to $135 billion this year on AI infrastructure. Amazon is guiding toward $200 billion. Alphabet, $175 to $185 billion.

And what has all of that money produced in competitive terms? A 2.7 percent edge. US and Chinese models have traded first place multiple times since early 2025. In February 2025, DeepSeek-R1 briefly matched the top US model before being surpassed. The lead, when it exists at all, is measured in months.

The report's co-chairs, Yolanda Gil and Raymond Perrault, put it directly: "The data does not point in a single direction. It reveals a field that is scaling faster than the systems around it can adapt."

That framing applies to national competitiveness as much as it does to enterprise adoption.

China is winning the efficiency war

China is not just closing the gap. It is doing so at a fraction of the cost, under active US export controls on advanced semiconductors, and with an entirely different strategic playbook.

DeepSeek's V4 model, released in March 2026, delivers benchmark scores comparable to frontier US models at roughly 50 times lower API cost. A complex task that costs $15 on GPT-5 runs about $0.50 on DeepSeek. The architecture uses mixture-of-experts: 671 billion total parameters but only 37 billion activated per token, requiring approximately 250 GFLOPs — nearly 10 times less computation than dense models of equivalent capability.

The broader picture is harder to dismiss. China now holds 69.7 percent of global AI patent filings. Chinese researchers produce 23.2 percent of global AI publications and 20.6 percent of citations, compared with 12.6 percent for the United States. China installed 295,000 industrial robots in the most recent reporting period — nearly nine times the US rate of 34,200.

And the Stanford report flags a structural vulnerability that pure spending cannot fix: China channels AI resources through government guidance funds — state-initiated investment vehicles that deployed an estimated $912 billion across industries between 2000 and 2023, with approximately $184 billion directed toward AI companies. Private investment comparisons, the report notes, likely understate the amount China is directing toward AI.

The American advantages in chips and capital are real. But they depend on export controls holding, TSMC remaining aligned, and talent continuing to flow westward — all three of which are shakier than they were two years ago.

The enterprise spending black hole

The national competitiveness data is alarming. The enterprise data is worse.

According to the Stanford Index, 88 percent of organizations now use AI for at least one business function. Seventy percent have deployed generative AI in at least one function, up from 33 percent in 2023. But only 6 percent are actually capturing meaningful enterprise value from their AI investments.

A 2025 MIT study found that 95 percent of corporate AI projects fail to create measurable value. The PwC 2026 Global CEO Survey reports that 56 percent of CEOs say they are getting "nothing" from their AI adoption efforts. Fifty-four percent of C-suite executives admit adopting AI is "tearing their company apart."

None of this is slowing the spending. The AI consulting market is growing at 35.8 percent annually. Companies plan to double their AI spend from 0.8 percent to 1.7 percent of revenue in 2026. Accenture booked $3.6 billion in AI services. IBM reported $6 billion. BCG claims $2.7 billion in annual AI revenue, 20 percent of its total.

Gartner places AI squarely in the "Trough of Disillusionment" throughout 2026. That assessment is generous. What the data actually shows is not a trough — it is a pattern of institutional failure to convert technology investment into operational value. The same pattern, at the national level, that the US-China performance gap reveals.

Money in. Very little out.

The talent pipeline is the real crisis

The finding that should concern American policymakers more than any benchmark score: AI talent migration to the United States has dropped 89 percent since 2017. That decline accelerated 80 percent in the last year alone, attributed partly to visa restrictions under the current administration.

Switzerland — not the United States — now ranks first globally for AI researchers and developers per capita. Singapore leads global AI adoption at 61 percent. The UAE is at 54 percent. The United States sits 24th at 28.3 percent.

The workforce disruption is already visible inside the US. American software developers aged 22 to 25 have seen employment fall nearly 20 percent since late 2022 — the exact moment generative AI tools entered mainstream use. Developers aged 30 and older saw employment grow 6 to 12 percent over the same period. The pattern repeats in customer service, accounting, and administration. It does not appear in low-AI-exposure jobs.

New AI PhDs in the US and Canada increased 22 percent from 2022 to 2024, but the growth went to academia — not to industry, where the deployment crisis lives.

The Stanford report is blunt about what this means: "These talent patterns represent a fundamental challenge to U.S. technological leadership that export controls and computing investments alone cannot address."

Read that sentence twice. Export controls and computing investments — the two pillars of current US AI policy — "alone cannot address" the talent crisis. Stanford is telling policymakers, in the most diplomatic language available to an academic institution, that the strategy is incomplete.

What actually wins — and it is not compute

The Stanford data, taken together, makes a clear argument: the future of AI competitiveness will not be decided by who spends the most on GPUs.

It will be decided by who builds the deepest talent pipelines. Who funds university research at a level that retains researchers domestically instead of sending them to industry or abroad. Who creates workplace cultures that can actually absorb AI into operations — not just purchase licenses. Who builds social norms around AI adoption that accelerate diffusion rather than resist it.

China's 69.7 percent of global AI patents did not come from outspending the US. It came from a sustained, decades-long investment in research infrastructure, STEM education, and industrial policy that treats AI as a national capability — not a line item on a tech company's capital expenditure guidance.

The United States ranks 24th in AI adoption. Not because it lacks models. Not because it lacks capital. Because the institutions between the models and the people — the enterprises, the agencies, the schools — are not ready. Readiness is the bottleneck, not compute.

The value crisis is solvable — but not with more spending

The larger signal I am tracking in my research into AI value in enterprise deployments is exactly what the Stanford Index confirms at the macro level: spending is not the constraint. Organisational readiness is.

Eighty-eight percent adoption and six percent value capture is not a technology problem. It is a deployment, culture, and change management problem. The 95 percent project failure rate MIT documented is not caused by bad models. It is caused by bad implementation — unclear objectives, misaligned incentives, missing skills, and leadership that treats AI as a procurement decision instead of an operational transformation.

This is why I built AI Value Acceleration. Not to help organisations buy more AI, but to break down the internal barriers — workforce readiness, cultural resistance, process misalignment, measurement gaps — that sit between a signed enterprise license and actual business value. The Stanford data makes the case more urgently than I ever could: the spending is there, the technology is there, and the results are not. Something in between is broken, and fixing it requires a fundamentally different approach than writing bigger cheques.

I am currently building a report on exactly where enterprise AI adoption breaks down and what the organisations that are capturing value are doing differently. If you are an AI leader, a CAIO, a VP of digital transformation, or anyone responsible for scaling AI adoption inside your organisation, I want to hear from you. What is working. What is not. Where the real blockers live. You can reach me at [email protected].

The Stanford AI Index gave us the data. The question now is whether the people making the decisions will read it — or keep writing cheques.

Frequently asked questions

What does the Stanford AI Index 2026 say about US vs China?

The Stanford HAI AI Index 2026 found that the performance gap between the best US and Chinese AI models has collapsed to 2.7 percent, down from 17.5 to 31.6 percentage points in May 2023. This narrowing occurred despite the United States investing 23.1 times more in private AI than China ($285.9 billion vs $12.4 billion). China now holds 69.7 percent of global AI patent filings and produces 23.2 percent of global AI publications. US and Chinese models have traded first place on benchmarks multiple times since early 2025.

How much does the US spend on AI compared to China?

US private AI investment in 2025 was $285.9 billion, compared with China's $12.4 billion — a 23.1x ratio. However, China also channels significant resources through government guidance funds, estimated at $184 billion directed toward AI companies between 2000 and 2023. Global corporate AI investment reached $581.7 billion in 2025, up 130 percent year over year. Big Tech hyperscalers (Amazon, Alphabet, Microsoft, Meta, Oracle) plan to spend roughly $635 to $665 billion combined in 2026.

Why are enterprise AI projects failing?

A 2025 MIT study found that 95 percent of corporate AI projects fail to create measurable value. The Stanford AI Index reports that while 88 percent of organisations now use AI, only 6 percent capture meaningful enterprise value. PwC's 2026 Global CEO Survey found 56 percent of CEOs report getting "nothing" from their AI adoption efforts. The primary causes are not technological — they include unclear objectives, misaligned incentives, workforce readiness gaps, and leadership treating AI as a procurement decision rather than an operational transformation.

What is DeepSeek V4 and why does it matter?

DeepSeek V4 is a large language model released by Chinese AI lab DeepSeek in March 2026. It uses a mixture-of-experts architecture with 671 billion total parameters but activates only 37 billion per token, requiring roughly 10 times less computation than equivalent dense models. It delivers benchmark scores comparable to frontier US models at approximately 50 times lower API cost — a complex task costing $15 on GPT-5 runs about $0.50 on DeepSeek. It demonstrates that model performance is increasingly driven by architectural efficiency, not raw compute spending.

How has AI talent migration to the US changed?

AI talent migration to the United States has dropped 89 percent since 2017, with an 80 percent acceleration in the decline in the past year alone. Switzerland now ranks first globally for AI researchers and developers per capita. The Stanford report states that "these talent patterns represent a fundamental challenge to U.S. technological leadership that export controls and computing investments alone cannot address." Meanwhile, new AI PhD production in the US has increased 22 percent, but the growth has gone to academia rather than industry.