Top Jurisdictions for AI Companies: Where the Industry Is Actually Registered

Delaware is where a huge share of the AI industry actually lives, including a lot of companies that don't sound American.

The pattern is consistent. A company is founded somewhere in Europe, builds a serious research team, raises a round from American investors, and creates a US parent company as part of that round. The original entity becomes a wholly-owned subsidiary. Branding, hiring and regulatory obligations stay national. Ownership doesn't.

You can see this most clearly with a well-known British text-to-image company. Its UK filing lists a single controlling entity — and that entity's governing law is Delaware. A German image-generation lab runs the same way: a German operating company under an American parent, which reporting has described as a deliberate hedge against EU regulation. A voice-synthesis company with Polish founders and offices in London and New York names a Delaware corporation as the contracting party in its terms of service, including the European version.

Delaware doesn't require public disclosure of owners. It offers a specialised business court and a century of predictable case law, which is what investors are actually buying.

• ownership of frontier language model labs
• agent startups
• voice and audio companies
• image and video generators
• AI infrastructure
• robotics

Almost none of the work happens in the state.

Nevada, the alternative

At least one very large American frontier lab registered in Nevada instead, which offers stronger protection for directors against shareholder suits.

There's an unexpected consequence. Nevada requires public benefit corporations to file their benefit reports publicly, whereas Delaware only requires reporting to shareholders. That transparency is how outside researchers were able to establish from public records that the company appeared to have quietly dropped its benefit-corporation status — while still being described as one in litigation. If it had registered in Delaware, nobody would have been able to check.

Chinese AI companies register as domestic limited or joint-stock companies, and the public credit system discloses registered capital, legal representative, business scope and shareholders.

The business scope field is worth watching. When a lab adds a licence category like internet information services, that's a visible signal a consumer product launch is coming.

The strangest structure in the industry belongs to a Hangzhou lab known for efficiency research and long-context models. At founding it was a wholly-owned subsidiary of its own Beijing subsidiary — which was in turn wholly owned by it. A closed loop, legally impossible to unwind from inside. Someone noticed within a couple of weeks, and the equity was transferred out. Ownership today runs through the founder personally at one percent and a limited partnership holding the rest, with the partnership vehicles registered in a coastal district chosen largely for tax treatment. Behind it all is a quantitative hedge fund.

The key thing about that structure is that it's entirely domestic. No offshore shell, no foreign investors. Which is exactly why a lab like that can't easily list abroad even if it wanted to.

Another well-known Beijing assistant developer carries registered capital of about one million yuan — roughly $140,000 — on a company that has raised well over a billion dollars. The dollars come in through an offshore preferred-share structure that never touches the domestic capital account.

• frontier general models
• mixture-of-experts efficiency research
• long-context work
• video generation
• agent products
• enterprise deployment into state-owned enterprises and government

In January 2026, two Chinese AI companies listed in Hong Kong days apart — the first pure-play large language model companies anywhere to go public. They took opposite routes, which makes them a useful pair to compare without naming either.

The first is a Beijing company with deep university roots, incorporated in 2019 and converted to joint-stock form in early 2025 specifically to list. It's a domestic Chinese company — no offshore shell anywhere in the structure. It priced at around HK$116 and was valued near $7 billion. It has also been on the US Entity List since January 2025.

The second is a Shanghai company founded in late 2021 by a former computer-vision executive. It's Cayman-incorporated, listed under Hong Kong's regime for pre-profit specialist technology companies, with weighted voting rights giving the founder control. It doubled on its first day of trading.

Cayman is where nothing is built and everything is owned. Multiple share classes, no local tax layer, accepted by the major exchanges. For an Asian AI company raising US dollars with an eventual overseas listing in mind, it's the standard parent jurisdiction.

Hong Kong's role is different — it's a disclosure venue. Its listing regime admits companies on R&D spending and third-party investment rather than earnings, which is built for exactly this industry. The prospectuses those two companies filed forced research-stage businesses to publish revenue composition, customer concentration and full ownership. They remain the richest public documents anyone has on Chinese AI.

France has produced the most valuable AI company in Europe, and it's French all the way up — a société par actions simplifiée on the Paris commercial register, with no American holding company above it. Among large Western labs, that's genuinely unusual.

It didn't happen by accident. The French state has used procurement, compute subsidy and sustained political pressure to keep companies domiciled locally instead of flipping.

The French register produces one odd artefact. Share capital has to be re-filed every time it changes, and employee option exercises change it constantly — so the public record shows a trail of tiny increases: a few thousand euros, then two hundred, then seventy-four, then seventy-four cents, then two euros fifty-nine. Total registered capital sits in the low tens of thousands on a company worth billions. Legally meaningful, economically meaningless.

• open-weight language models
• code generation models
• document AI and OCR
• speech
• non-profit audio research labs

Germany's flagship language model company is registered as a GmbH in Heidelberg. Its research work sits in a completely separate company, registered in a different court district entirely, and wholly owned by a charitable non-profit.

That structure isn't available in the US or UK in the same form. It lets research capacity sit under a charitable parent, insulated from investor return requirements, with a commercial sister company alongside. If you care about how AI research can be governed differently, it's the most interesting legal shape in Europe.

In 2026, that company agreed to be acquired by a Canadian AI firm — which quietly moves Germany's national champion under foreign ownership.

Germany also hosts one of the leading image and video generation labs, which runs a German operating company under a Delaware parent.

• diffusion-based image and video generation
• sovereign and public-sector language models
• industrial AI

These aren't places anything gets built. They're where the holding company sits.

The best illustration is a European AI infrastructure business that today runs large GPU clusters and trades on Nasdaq. It exists in its current form only because its Dutch holding company — set up nearly two decades ago for a completely different business — survived the divestment of its Russian operations. The holding jurisdiction is what made a clean separation legally possible.

That's the whole argument for these jurisdictions in one example. It's insurance that pays out once, and enormously.

• AI cloud and compute providers
• group holding vehicles
• autonomous driving subsidiaries
• the European service entities of foreign labs

Singapore has become the pressure valve of the AI industry, and the reason is geopolitical rather than fiscal.

The clearest case is a Chinese agent company whose product went viral in 2025. It had incorporated a Singapore entity back in 2023, and in mid-2025 it moved there in earnest: cut its mainland team from over 120 people to about 40, relocated those 40, cleared its Chinese social accounts, and blocked Chinese IP addresses from its site. The original Chinese entity kept existing. By late 2025 it looked like a Singapore company.

The driver was capital. Since October 2024, US rules have restricted American funds from investing in China's AI sector, and Singapore became the workaround — a practice the financial press now calls "Singapore washing."

It didn't fully work. A US technology giant announced a $2 billion acquisition at the end of 2025; Chinese regulators opened a review in January 2026 and the deal was blocked, with domestic commentary framing the relocation as crossing lines on technology and data sovereignty.

The lesson generalises. Redomiciliation isn't laundering — acquirers and screening regulators look through to founders, staff and technology origin. And the country you left may object as loudly as the one you were trying to satisfy.

• AI agents
• consumer AI apps
• relocated model developers
• regional headquarters

Canada's leading enterprise AI company is a federal corporation under the Canada Business Corporations Act, with a Toronto registered office and no Delaware parent. For a company that has raised heavily from international investors, that's rare.

It's also now the buyer of Germany's flagship model developer, which makes Canada a bigger presence on this map than most people would guess.

• enterprise AI for regulated industries — finance, healthcare, energy, public sector
• multilingual open models

Israeli AI companies register with a national registrar and get a nine-digit company number, but the Israeli entity is usually only half the picture. US customers, US investors and a US-registered parent or sibling are close to standard from the very beginning.

One of the most watched superintelligence research labs in the world is an American corporation operating across California and Tel Aviv simultaneously.

• enterprise language models
• retrieval and RAG
• computer vision
• autonomous driving
• defence AI
• alignment research

Japan gives every entity a permanent 13-digit corporate number, published free by the tax authority and cross-referenced across government datasets. Confirming a Japanese company takes about ten seconds. Almost nowhere else is that easy.

The country's most prominent AI research startup is a kabushiki kaisha registered in central Tokyo, which relocated its registered office to a newer tower as it grew — a move visible in the public record.

• novel architectures and evolutionary methods
• Japanese-language models
• public sector and defence work
• robotics

Indian AI companies register with the national corporate affairs ministry, and the identifier is unusually informative — the string itself encodes industry code, state, year of incorporation, company class and sequence number before you open a single filing.

One prominent Indian model developer, backed by a well-known mobility founder, carries authorised and paid-up capital of ₹100,000 — about $1,150. Another registers at an address in a rural district outside Bangalore, separate from its working office in the city.

India also ties eligibility for national compute and language-model programmes to domestic incorporation, which gives founders a real reason not to flip offshore.

• Indic-language models
• speech interfaces for low-literacy users
• document AI
• government language missions

Some of the strongest open-weight models in the world come from an Abu Dhabi research institute that isn't a company.

It sits under a government research council established by an emirate-level law as an independent legal entity affiliated to the executive council. No shareholders. No share capital. No registrar entry, no cap table, no acquisition risk. It answers to a state rather than to investors.

Alongside it, the Gulf financial free zones run their own common-law courts and registries inside a civil-law country, which gives conventional companies English-language commercial law and full foreign ownership.

• open-weight frontier models
• Arabic-language AI
• national compute
• sovereign AI cloud