Which Companies Are Building Superintelligence? (2026)

As of September 23, 2026, only two companies state superintelligence as their own goal on an official page: Meta and Safe Superintelligence Inc. (SSI). OpenAI's stated mission is AGI, not superintelligence — the word appears in its safety writing. Anthropic, Google DeepMind and xAI (now SpaceXAI) barely use the word at all on their official sites.

We opened all six official pages and quote only what each company says in its own words. The useful part is not the list — it is the split: who treats superintelligence as a goal, and who treats it as a risk.

The short answer: two goal-setters, one risk-framer, three abstainers

If you searched for "which companies are building superintelligence," here is the sorted answer based on official pages rather than press coverage.

• Superintelligence as the goal: Meta ("personal superintelligence") and SSI (the goal is the company name)
• Superintelligence as a risk to be aligned: OpenAI (mission is AGI)
• Different vocabulary entirely: Anthropic ("powerful AI"), Google DeepMind ("AGI"), xAI/SpaceXAI ("accelerate human scientific discovery")

One caveat that trips people up: not using the word does not mean not building it. Anthropic and Google DeepMind both design safety systems around models that exceed human performance. What differs is the banner they chose, not the ambition. For the definition of the term itself, see our breakdown of what SI and superintelligence actually mean.

Stance map: goal vs risk, in each company's own words

Every quote below is verbatim from the official page listed in the last column, retrieved September 23, 2026.

Read the table vertically and a pattern shows up. "Superintelligence" behaves less like a technical milestone and more like a recruiting and fundraising banner. When you see the word in a deck, the first question worth asking is which column the source sits in.

Meta: personal superintelligence, for everyone

Meta publishes a letter from Mark Zuckerberg dated July 30, 2025. It opens with "Over the last few months we have begun to see glimpses of our AI systems improving themselves," and states plainly: "Developing superintelligence is now in sight." (Meta — Personal Superintelligence)

Meta's differentiator is distribution, not capability. The same letter says its approach "is distinct from others in the industry who believe superintelligence should be directed centrally towards automating all valuable work, and then humanity will live on a dole of its output." It also argues that glasses — devices that "see what we see, hear what we hear" — will become the primary computing device for that personal superintelligence.

The lab name is confirmed on Meta's leadership page: Alexandr Wang joined in June 2025 as Meta's first-ever Chief AI Officer and leads Meta Superintelligence Labs (MSL) (Meta — Alexandr Wang, Chief AI Officer). He founded Scale AI in 2016.

Worth noting for anyone citing MSL: Meta's AI blog URL for "meta-superintelligence-labs" returned a 404 on September 23, 2026. The leadership page is currently the most reliable first-party source for the org's name.

Safe Superintelligence Inc.: one goal, one product

SSI's entire website is one page. It begins "Superintelligence is within reach." and continues: "We have started the world's first straight-shot SSI lab, with one goal and one product: a safe superintelligence." (Safe Superintelligence Inc.)

The no-intermediate-products strategy is stated as deliberate: "our business model means safety, security, and progress are all insulated from short-term commercial pressures." Offices are listed in Palo Alto and Tel Aviv.

Founder Ilya Sutskever is described by OpenAI's own Superalignment post as "cofounder and Chief Scientist of OpenAI." In other words, the person OpenAI put in charge of superintelligence alignment left and built a company around it.

SSI's official Updates page lists exactly three events (SSI — Updates):

• September 4, 2024: raised $1B from NFDG, a16z, Sequoia, DST Global and SV Angel
• July 3, 2025: Daniel Gross departed June 29; Sutskever became CEO and Daniel Levy President
• July 26, 2026: long-term strategic partnership with NVIDIA to scale compute 10x

The 2025 note also addresses acquisition rumors directly: "We are flattered by their attention but are focused on seeing our work through."

OpenAI: the mission is AGI, superintelligence is the risk case

The OpenAI Charter defines the mission around AGI: "highly autonomous systems that outperform humans at most economically valuable work" (OpenAI Charter). Superintelligence is not in the mission statement.

It is, however, all over the safety writing. "Introducing Superalignment" announced a team co-led by Ilya Sutskever and Jan Leike, with 20% of compute secured to date dedicated to the problem and a four-year target (OpenAI — Introducing Superalignment). A footnote makes the distinction explicit: "Here we focus on superintelligence rather than AGI to stress a much higher capability level."

"Governance of superintelligence" proposes "something like an IAEA for superintelligence efforts" with inspection and audit powers (OpenAI — Governance of superintelligence). The more recent "AI progress and recommendations" states: "no one should deploy superintelligent systems without being able to robustly align and control them" (OpenAI — AI progress and recommendations).

A practical note: none of these OpenAI pages display a publication date (checked September 23, 2026). If you need the date in a citation, verify it against a second source.

Anthropic: "powerful AI," not superintelligence

Anthropic's company page opens with "Anthropic is an AI safety and research company." (Anthropic — Company). No superintelligence goal appears anywhere on it.

Its "Core Views on AI Safety" expects systems "possibly equaling or exceeding human level performance at most intellectual tasks," while stating flatly that "We do not know how to train systems to robustly behave well." (Anthropic — Core Views on AI Safety)

The vocabulary choice is explained by CEO Dario Amodei in "Machines of Loving Grace" (October 2024). He writes "powerful AI (I dislike the term AGI)," adding in a footnote that AGI "has gathered a lot of sci-fi baggage and hype." His preferred image of the end state is "a country of geniuses in a datacenter" (Dario Amodei — Machines of Loving Grace).

In practice, this is the cleanest signal in the whole comparison: some labs inflate the vocabulary to attract capital and talent, and some deliberately deflate it.

Google DeepMind: AGI all the way down

Google DeepMind's responsibility page never says superintelligence. It describes an "AGI Safety Council, led by our Co-Founder and Chief AGI Scientist Shane Legg," working to guard "against extreme risks that could arise from powerful AGI systems in the future," plus a Frontier Safety Framework of protocols it keeps updating (Google DeepMind — Responsibility & Safety). Even the job title is standardized on AGI.

xAI (SpaceXAI): not a superintelligence claim

xAI's About page states the mission as "Accelerate human scientific discovery." — no mention of superintelligence (xAI (SpaceXAI) — About, checked September 23, 2026). Its stated principles are reasoning from first principles, no goal being too ambitious, and moving quickly.

The company also announced that SpaceX acquired xAI on February 2, 2026, and the site now brands itself SpaceXAI (xAI — xAI joins SpaceX). It is frequently grouped with "superintelligence companies" in coverage, but that framing is not supported by its own pages.

Why the US government's new "SI" label is a different thing

On September 22, 2026, President Trump announced at the UN General Assembly that US government documents would refer to AI as "Super Intelligence (SI)." That applies to US government documents and carries no obligation for companies or other countries — see our breakdown of the AI-to-SI rename and what it actually changes.

The two uses are not the same. The government label renames today's AI. The labs' "superintelligence" describes a system that does not exist yet. SSI says it is "within reach"; Meta says it is "now in sight." Neither calls a shipping model superintelligent. For timeline estimates, see our roundup of when superintelligence might actually arrive.

Which of these are public companies?

A common follow-up: can you buy any of them? Here is the ownership picture as of September 23, 2026.

Of the two companies that name superintelligence as their goal, one (Meta) is listed and one (SSI) is private. OpenAI, Anthropic and SSI remain private as of September 23, 2026, while xAI is now reachable only indirectly through SpaceX, much as DeepMind is through Alphabet. This article documents official statements only and is not investment advice.

Three checks before you cite any of this

A common pattern we see in internal AI strategy decks is confident sourcing that falls apart on inspection. Three checks catch most of it.

1. First-party page or press paraphrase? "Meta Superintelligence Labs" is widely reported, yet the dedicated blog page 404s. Go back to leadership bios, charters or IR pages.
2. Mission document or safety document? OpenAI's mission is AGI; its safety posts discuss superintelligence. Which document you quote changes the claim.
3. Is there a date on the page? OpenAI's posts show none. Record a retrieval date, or verify elsewhere.

The third one matters most, because AI strategy documents age in months. Storing "source URL + retrieval date" together saves whoever re-checks the claim later.

Summary

Two companies build toward superintelligence by name: Meta and SSI. OpenAI aims at AGI while treating superintelligence as the catastrophic-risk case. Anthropic, Google DeepMind and xAI use other vocabulary for similar ambitions. The word tells you about positioning, not capability.

Mihata builds custom AI tools and runs monthly AI strategy sessions for teams trying to separate signal from vocabulary in exactly this space. If that sounds useful, we would be glad to talk.