Top 93 Countries in AI EXPLAINED: Why America's #1 Ranking Is Misleading
The metrics that matter in the AI race
Here are three questions. Get them all right, and you understand the AI race.
Get them wrong... well, you’re about to find out why everyone’s been looking at the wrong scoreboard.
Question 1: Which country has the most coherent national AI strategy? A) United States, B) United Kingdom, C) Saudi Arabia
Question 2: Which country’s citizens actually USE AI platforms the most in their daily lives? Is it: A) United States, B) United Arab Emirates, C) United Kingdom
Question 3: Which countries get the most AI ‘bang for their buck’? A) United States, B) Singapore, C) Israel
Type your answer in the comment now before we continue.
If you said A, A, and A... you just failed the test that 93 countries are actually taking.
By the end of this article, you’ll know the answers to these questions, and they look nothing like the ones you’ve been reading about.
What the US Actually Won
Let’s start with the one the US is actually winning.
Who can build the best AI?
According to the Observer Global AI Index, which has been ranking countries on AI readiness for 6 years.
The United States is ranked #1 in the world for overall AI capacity.
By throwing trillions at the problem, American companies built the best labs, the best researchers, and the world-class cloud infrastructure to match.
ChatGPT. Claude. Gemini. The three biggest generative AI providers — with over 90% of global AI usage share.
But building the best AI and winning the AI race on all fronts are not the same thing.
Because here’s where the anomalies start to show.
US-Specific Anomalies
1. The AI Adoption Paradox
Somehow, despite owning those market shares, fewer Americans proportionally are using their own Gen AI apps.
The number here is ironic.
While being #1 in overall AI capacity, only 25% of Americans regularly use AI platforms like ChatGPT or Claude.
Compared to the UAE and Singapore, the countries ranked significantly lower in overall capacity, with over 50% of their populations using AI platforms regularly.
The main reason, also quite a simple one, is the income and education inequality.
These two factors are nearly a deadly spiral on their own, especially for Americans living in rural areas.
They have lower incomes and limited access to the internet, let alone AI. They also aren’t educated enough, or they’re mostly blue-collar workers for whom using AI makes no sense at all (in the short term, at least). But this also means their next generations are likely to be blue-collar workers with low income, as their parents were.
The report also discussed how Americans have less trust and a lot of anxiety about using AI, which I believe is biased and has a narrow view of this matter. I will explain more about this at the end, when I discuss the data source for this ranking.
Regardless of the reasons, the response to that adoption gap isn’t to fix the conditions that caused it. It’s to spend more.
2. The Cash Flow Sucked Dry by AI
Amazon, Microsoft, Google, and Meta combined are set to spend $665 billion on AI infrastructure in 2026.
Mostly on GPUs. Data centers. The physical infrastructure to train and run large language models.
Let’s put that in perspective: $665 billion is eight times the entire US federal education budget ($79B). This isn’t a diversified investment, but an all-in bet on one technology, which they pray will deliver returns.
However, what will happen with AI capex consuming 90%+ of these companies’ operating cash flows?
Take Amazon, for example, which is projected to hit a negative free cash flow of $17 billion in 2026. The most cash-efficient company in tech history is about to burn through cash.
This term, "Capex Cliff," is the latest theory about how AI-related earnings don’t match the spending. Their bet fails if Big Tech companies (Microsoft, Google, Meta, Amazon) don't see a sign of profits from AI features by early 2027.
And what if it failed?
It'd start when enough companies decided AI's cost simply wasn't worth it, dumping their subscriptions and cratering revenue for Salesforce and Adobe alike. The data centers become ghost factories. Only the giants with other cash cows, Google, Meta, and Microsoft, survive the wreckage, subsidizing their AI losses while investors lose faith anyway.
That's the financial risk. But there's another layer underneath it: the US is making a $665 billion bet in arguably the worst-prepared environment for it to pay off.
3. Bad Operating Environment and No Government Strategy
Despite ranking #1 overall, in Government Strategy, the US fares only marginally better at #3 — held back not by lack of spending, but by the nature of it.
The US commits enormous capital, but almost entirely toward frontier models and AGI, with no coherent national strategy signed at the head-of-state level and no measurable public KPIs.
The US builds the world’s most powerful AI despite a poor government environment for it.
But what’s worse is that the US sits at #45 in Operating Environment, which measures the social and regulatory soil in which AI actually grows, like public trust, citizen attitudes, and how quickly a country lets skilled AI workers through its doors.
Singapore (#1), South Korea (#2), and the UAE (#3) all knew the best approach isn’t by building the most powerful models.
They got there by building the conditions for AI to be accepted by the business and the public.
The US has no federal AI law, only a revolving door of executive orders that change with each administration. Its citizens are more nervous about AI than excited by it. And the world's best AI researchers enter the country through the same visa lottery as everyone else, capped, employer-tied, and years-long (your author here has been waiting in line for approval for 2 years now)!
Singapore, on the other hand, hands (no pun intended) senior tech talent a visa with no employer required, and the UAE offers a decade-long residency specifically for AI specialists.
The US built the world's most powerful AI despite its own conditions, not because of them.
Who’s Catching Up to the US and How
1. UAE, Buying Influence, Not Just Technology
In 2017, the UAE appointed the world’s first AI minister. At the time, it read as a branding exercise. Eight years later, the appointment looks prescient.
The UAE jumped 10 places in a single year to break into the top 10 of the Observer Global AI Index, and an impressive #2 in Government Strategy.
There are many signs of their rise in AI.
For instance, in Spring 2024, Microsoft invested $1.5 billion into the UAE’s flagship AI company. It now has preferential access to US-origin AI hardware and a direct pipeline into one of the world’s largest technology ecosystems.
With 64% of its working-age population already using AI, making it the most AI-adopted country on earth by that measure.
But the more consequential play is what the UAE is doing with that position. The UAE has signed 20 memoranda of understanding with African governments for AI-related projects since late 2022, pledging $3 billion+ to build data centres and cloud computing infrastructure across the continent. Partnering with countries like Ghana, Rwanda, and Zimbabwe.
Of course, this isn’t pure philanthropy.
The UAE’s bilateral trade with Africa reached $107 billion in 2024 — a 28% increase year-on-year. AI infrastructure is the new entry point for long-term economic relationships.
To them, it isn’t about building the best model, but more importantly, the alliance and the conditions under which its model of AI governance becomes the one that other countries adopt.
2. South Korea’s Sovereignty AI, From Hardware-to-Models.
South Korea ranks #5 overall. They’ve surpassed France and are breathing down the UK’s neck.
Besides being the nation of K-pop and the Squid Game, they are attempting something almost no other country can credibly try: full-stack sovereignty.
Meaning, owning every layer of the AI value chain:
Data, with an enormous, high-quality Korean-language dataset
Compute with the world’s leading high-bandwidth memory chips; Rebellions is building AI-specific NPUs domestically
Infrastructure, leveraging its telecom as the network for local access
A government-backed consortium, involving Naver, LG, and SK Telecom (think Google, IBM, AT&T), is building a sovereign open-source LLM.
Full sovereignty isn’t quite there yet at the hardware layer. But that’s the direction of travel, and it’s deliberate.
South Korea’s Science Minister estimated (optimistically) at the end of 2025 that the gap between Korean models and the US frontier is now 5.9 months.
Korea chose control over speed.
3. China’s Open-Source Takeover
In early 2024, the US accounted for over 70% of global open-weight AI adoption by model origin. Meta had Llama. Europe had Mistral. While China was barely a rounding error — under 5% of global usage ran on Chinese models.
With just a year apart… In late 2025, that figure had inverted.
China’s share of global open-weight AI usage had crossed 50%. The US had fallen to around 25%. The country that wasn’t supposed to have the firepower to compete, being cut off from advanced Nvidia chips, had become the one the open community was switching to.
The adoption is, ironically, happening within Silicon Valley.
Companies like Airbnb have shifted production workloads to Chinese models like DeepSeek or Alibaba’s Qwen, which offer performance that rivals frontier US models at a fraction of the inference cost.
You can also see in the Google Trends screenshot below of how often American users are searching for Chinese models.
Chinese developers, denied access to the best compute, successfully engineered a workaround. Then released the results under permissive open-source licences. Anyone can take them. Anyone can deploy them.
The US has now registered this as a strategic problem. The White House AI Action Plan explicitly names open-source as a matter of “geostrategic importance”. Recent months have seen open-weight releases from US labs that signal a genuine change in posture.
But posture isn’t position.
If the US continues to focus on immediate commercial revenue and leaves open source as is, the US will soon lose in innovation.
Which country got the most AI bang for its buck?
So, after all the discussion, it brings us to the final reveal:
Which country got the most AI bang for its buck?
To answer that, we need to look at intensity: AI capability per person—researchers per capita, infrastructure per capita, investment, innovation per capita, and all the factors mentioned in this report.
#1 in the world: Singapore.
Singapore's governance frameworks, so their businesses can plan around, a national reskilling infrastructure that reaches every worker, and coordinated public-private investment aligned behind a single strategy.
While the US has thrown trillions at the problem, resulting in half the efficiency.
If you’re an American, you’re statistically less likely to be near cutting-edge AI research, work in the sector, or benefit from the infrastructure than if you’re Singaporean.
Closing. The Methodology Problem in the Observer Global AI Index
These rankings from the Observer Global AI Index are useful.
But, a reminder they’re not gospel.
The index’s deeper problem is how it’s measured and what it can’t see.
They themselves admit that a big part of their data relies on self-reporting information. For example, if someone on LinkedIn crowned themselves an AI marketing expert, it counts as an AI talent.
By anchoring its Commercial scoring on GitHub commit histories or Crunchbase-style databases that structurally undercount non-Anglophone ecosystems, the index mistakes visibility data from Western platforms for actual global AI capacity.
This is most damaging for China, which operates an entirely parallel software and professional ecosystem — Zhipin rather than LinkedIn or Gitee rather than GitHub — none of which feed into the index’s indicators.
The result is a ranking that systematically rewards countries whose AI activity happens to be legible to Anglo-American data infrastructure, while penalising those whose equivalent activity occurs elsewhere.
One other glaring example that sits right in the data.
The index uses public trust surveys to explain why the US underperforms on adoption. The implicit diagnosis: Americans are nervous about AI.
That framing entirely misses something… I’d say, basic.
Around 80% of persistent poverty counties in America are rural. These are communities where the internet is unreliable or unaffordable, let alone a ChatGPT subscription.
Low adoption in those communities isn’t anxiety. It’s the absence of a broadband connection.
When an index measures “trust in AI” as a proxy for adoption readiness, it’s a very European perspective.
Not to mention their very subjective and arbitrary weighting method.
For an index that presents itself as a definitive global benchmark, the unexamined circularity of its sources — OECD drawing on LinkedIn drawing on self-reported job titles — should give you pause, especially if you're a policymaker shaping decisions based on this ranking.
If this made you think differently about the AI race, share it with someone who needs to see it.