Stanford AI Index 2026: China Matches US Models, But 95% of Enterprise AI Investment Returns Zero

Last week, Stanford’s Human-Centered AI Institute (HAI) released the 2026 AI Index Report, a sprawling 423 pages.

Honestly, what struck me most wasn’t a specific number, but two contradictory conclusions:

First, China’s LLM capabilities have caught up to the US.

Second, 95% of enterprise AI investments generated zero returns.

These two numbers together are quite interesting.

China matching the US — what does it mean?

The report states: Chinese models have reached or are approaching US model levels across multiple benchmarks.

Here’s a nuance worth noting: Stanford uses US benchmarks. Measuring Chinese models against a US ruler, and the result shows China has caught up. What does that tell us?

It tells us China’s AI capabilities have genuinely risen. Whether it’s datasets, algorithms, or compute investment, domestic LLMs now have the ability to compete directly with top international players.

But “capability” and “application” are two different things.

95% zero ROI — why?

This is the most jarring number in the report. 95%. This figure exceeded even my expectations.

Possible reasons:

First, most enterprise AI applications are still stuck in the pilot phase. They haven’t truly penetrated business processes, so they haven’t generated quantifiable returns.

Second, there’s a massive gap between AI vendors’ over-promises and enterprises’ expectation management. Companies thought AI was a cure-all, only to discover AI is just a tool — one that requires complementary process redesign and data governance.

Third, AI investment effects take time to materialize. Not seeing returns in the short term doesn’t mean there’s no long-term value.

But 95% is still way too high. This tells us the AI industry as a whole is still in early stages — the technology exists, but application scenarios, business models, and organizational adaptation haven’t caught up.

Why is China progressing faster than the US in some areas?

Let me share an observation that might seem counter-intuitive.

US AI startups generally focus more on improving foundation model capabilities, but commercialization actually lags behind China’s pace. Chinese AI companies focus more on scenario deployment — though their model capabilities might be slightly behind, their execution at the application layer is stronger.

This might explain why China achieved commercialization earlier than the US in certain vertical domains. In areas like facial recognition, voice recognition, and intelligent customer service, China’s deployment speed is much faster.

My conclusion

My takeaway from this report: The AI industry as a whole is in a state of “capability surplus, application deficit.” Technology runs ahead too fast, and commercialization can’t keep up.

The good news: China has proven it’s not a follower in AI, but a player that can compete head-on. The bad news: in both China and the US, most AI investments haven’t generated their deserved returns.

For practitioners, this means — AI competition has shifted from “technology races” to “application races.” Whoever can truly embed AI technology into business processes will laugh last.

Having technology isn’t enough. You have to use it.