A quick note: I researched this newsletter using six AI models. Perplexity, ChatGPT, Gemini, and Claude are American. Kimi K3 is Chinese. Mistral is French. I did all of it from an iPhone on a plane at 35,000 feet, connected through Starlink.

That's a strange sentence to be able to write, and it's part of why I started writing this week’s newsletter.

Here's what started it. On July 16 Moonshot AI released Kimi K3, a 2.8 trillion parameter model they're calling the first open 3T-class system. On the Artificial Analysis Intelligence Index it landed at number four, ahead of Claude Opus 4.8, behind Claude Fable 5 and GPT-5.6 Sol. Moonshot's own benchmarks show it beating models that were frontier a few weeks ago while losing to the newest ones. Worth noting those numbers are mostly self-reported, and the full weights aren't out yet. They're promised by July 27.

So it's not quite the frontier. But it's close, it's Chinese, and it's about to be free to download. Free to run is a different story. The weights are around 1.4 terabytes and Moonshot recommends 64 or more accelerators, so nobody's putting this on a laptop.

Every AI news article I read that week covered the same thing. American model, Chinese model, who's ahead, what it means for chips, etc. Which made me wonder about everyone else. Is France working on this? India? Are there twenty countries quietly building their own systems while we all watch two of them?

Turns out yes. And they're not doing what I expected.

The full-stack assumption

The thing I had wrong was the assumption that every country wants the full American setup. Chips, cloud platforms, research labs, consumer apps, venture money, all of it. Build the whole stack or stay dependent.

That's not what's happening. Most governments looked at the full stack, did the math, and picked the two or three pieces they could actually win. Then they built around those.

In many ways it’s similar to a car company. You need factories, parts suppliers, engineers, fuel, dealerships, and customers. The US and China have close to all of it. Most countries have two or three pieces and are missing the rest, especially the chips. So instead of pretending otherwise, they're building the parts of the machine they're good at and buying the rest.

France bet on electricity

France isn't going to make its own chips at scale, and it knows that. What it has is nuclear power, strong technical universities, and Mistral, which is the closest thing Europe has to a frontier lab.

The French play is energy plus government support plus open models. Macron committed roughly 109 billion euros to the sector last year, most of it aimed at data centers and compute. Cheap low-carbon electricity makes those cheaper to run, and open models mean French institutions aren't renting their AI from an American company. The chips still come from Nvidia, which is a real limitation. But "we have the power and the models" is a coherent strategy in a way that "we will build our own everything" would not be.

The EU is pooling

Zoom out to the whole European Union and the approach shifts to shared capacity. Member countries are combining money and infrastructure to build supercomputers no single country would fund alone, backed by good research universities and one big common market. The InvestAI initiative is aiming to mobilize 200 billion euros, with a chunk of that going to AI gigafactories. Privacy and data control are the differentiators, which is a real competitive position and not just regulation for its own sake.

The problem is fragmentation. Dozens of languages, mismatched capital markets, national priorities pointing different directions. Europe is building real compute. Whether that becomes an AI economy or just a lot of parallel projects is still open.

India went after languages

India's approach is shared compute plus local languages.

Rather than expecting every startup to build its own data center, the IndiaAI Mission is trying to make common infrastructure available to researchers and companies that could never fund it alone. Meanwhile India has hundreds of languages and dialects that global models handle badly, a huge technical workforce, and a mature software services industry. That combination points somewhere obvious. Get very good at building affordable systems for multilingual populations and public services.

India may never own a frontier model. It could still end up with the playbook for serving the several billion people that frontier models serve poorly.

Korea and Japan went physical

South Korea and Japan both wrapped AI around industries they already dominate.

Korea is tying AI to semiconductors, electronics, vehicles, and factories, with Samsung and SK Hynix sitting in critical parts of the chip supply chain. The advantage isn't software. It's putting AI directly into physical products at high volume.

Japan is going after robotics and physical AI, systems that run robots, vehicles, factories, and medical equipment, backed by decades of automation experience and government money for domestic compute like the ABCI supercomputer.

Neither is likely to produce a famous chatbot. Both could end up in the hardware everyone else depends on, which is a durable spot to occupy.

The UAE is buying its way in

The Emirates are running on capital, energy, and partnerships. Big investment funds, available land and power, a central location, close ties to American tech companies. That combination lets them approve and finance infrastructure at a speed most countries can't match.

Building it is slower than announcing it. Stargate UAE was announced in May 2025 as a 1-gigawatt cluster inside a larger 5-gigawatt campus outside Abu Dhabi. The first 200 megawatts is expected sometime this quarter, more than a year after the announcement. That's still fast for infrastructure at that scale, and the campus is enormous, but it's a reminder that data centers move at the speed of concrete and transformers rather than press releases.

Most of the core technology also comes from foreign partners, which is the obvious constraint. Nvidia chips, OpenAI and Oracle operating it. What the UAE brings is the money, the power, and the permission to build.

The pattern

Every one of these countries is missing pieces of the stack. None of them are trying to close every gap. They're building around what they already have, whether that's electricity, language, manufacturing, capital, or regulation.

Which raises the question I'll take up next week. The US and China aren't in the lead because their engineers are smarter. They're in the lead because they control most of the supply chain, and the reasons why are more interesting than the venture capital charts suggest.

Part 2 next week.

---

Sources

- Kimi K3 release, benchmarks, and weight timeline: [Artificial Analysis — Kimi K3](https://artificialanalysis.ai/models/kimi-k3); [Simon Willison — This is Kimi K3](https://simonwillison.net/2026/Jul/16/kimi-k3/); [Kimi API Platform — K3 quickstart](https://platform.kimi.ai/docs/guide/kimi-k3-quickstart).

- France's AI investment commitments and "AI powerhouse" strategy: [Élysée Palace — Make France an AI powerhouse](https://www.elysee.fr/en/emmanuel-macron/2025/02/11/make-france-an-ai-powerhouse); [Reuters — Macron signals investments of 109 billion euros in French AI sector](https://www.reuters.com/technology/artificial-intelligence/france-invest-109-billion-euros-ai-macron-announces-2025-02-09/); [TechCrunch — Macron unveils $112B AI investment package, France's answer to Stargate](https://techcrunch.com/2025/02/10/macron-unveils-a-112b-ai-investment-package-as-frances-answer-to-stargate/).

- European Union shared capacity and AI gigafactories: [European Commission — EU launches InvestAI initiative to mobilise EUR200 billion investment in artificial intelligence](https://luxembourg.representation.ec.europa.eu/actualites-et-evenements/actualites/eu-launches-investai-initiative-mobilise-eur200-billion-investment-artificial-intelligence-2025-02-11_en); [European Commission — AI Continent Action Plan](https://ec.europa.eu/commission/presscorner/detail/en/ip_25_1013).

- India's shared-compute and multilingual strategy: [Government of India — Cabinet approves IndiaAI Mission](https://www.pib.gov.in/PressReleasePage.aspx?PRID=2015261); [Government of India — Democratising AI in India](https://www.pib.gov.in/PressReleasePage.aspx?PRID=2225781).

- South Korea and Japan's industrial AI focus: [Ministry of Science and ICT, Republic of Korea — Korea AI computing infrastructure plan](https://www.msit.go.kr/eng/bbs/view.do?bbsSeqNo=42&mId=4&mPid=2&nttSeqNo=1092); [AIST — ABCI 3.0 announcement](https://www.aist.go.jp/aist_j/news/pr20241010.html); [ABCI official site](https://abci.ai/en/).

- UAE AI infrastructure and Stargate UAE timeline: [OpenAI — Introducing Stargate UAE](https://openai.com/index/introducing-stargate-uae/); [Reuters — 'Stargate UAE' AI datacenter to begin operation in 2026](https://www.reuters.com/business/media-telecom/stargate-uae-ai-datacenter-begin-operation-2026-2025-05-22/); [The National — Stargate UAE's first phase to be completed in third quarter of 2026](https://www.thenationalnews.com/business/2025/12/05/stargate-uaes-first-phase-to-be-completed-in-third-quarter-of-2026/).

// community highlights

Know someone who’d enjoy this?

Don’t miss the next issue.