Which companies are building superintelligence? (2026)

Clerion Research Team22 October 2026Updated 25 October 202613 min read

As of October 2026, the companies building superintelligence by name are OpenAI, Meta Superintelligence Labs, Safe Superintelligence (SSI) and Microsoft's MAI Superintelligence Team. Google DeepMind and Anthropic chase the same frontier as AGI or "powerful AI". None has shipped a superintelligent system.

Blue and near-black swirling blurred gradient banner reading Who builds superintelligence

As of October 2026, the companies building superintelligence by name are OpenAI, Meta (Meta Superintelligence Labs), Safe Superintelligence Inc. (SSI) and Microsoft (the MAI Superintelligence Team). Google DeepMind and Anthropic chase the same frontier but call it AGI or "powerful AI". None of them has shipped a system that is superintelligent.

That last sentence is the one most lists skip. Search the question and you get a roll call of logos with no line between "says it is building superintelligence" and "has built it". The gap matters to you, because the label moved fast this autumn and the models you call through an API did not.

Superintelligence is now the name of a team, not a description of a model.

Here is who is in the race, what each one has shipped, what they spend, and what it changes for a founder building on top of them.

Which companies are building superintelligence right now?

Four companies say, in their own words, that this is the goal: OpenAI, Meta, SSI and Microsoft. Google DeepMind and Anthropic work right at the same frontier but give it other names. A third group uses the word for a narrower job, such as CrowdStrike's Cyber Superintelligence Lab, launched on 1 September 2026 for security work.

Here's how each puts it:

  • OpenAI. In "The Gentle Singularity", published in June 2025, Sam Altman wrote that OpenAI is "a superintelligence research company" before anything else, and that "humanity is close to building digital superintelligence".
  • Meta. Mark Zuckerberg's letter of 30 July 2025 sets the goal as "everyone having a personal superintelligence that helps you achieve your goals". The work sits in Meta Superintelligence Labs (MSL), run by Alexandr Wang.
  • Safe Superintelligence Inc. SSI's site calls it a "straight-shot" lab with one goal and one product: safe superintelligence. It has offices in Palo Alto and Tel Aviv and says "superintelligence is within reach".
  • Microsoft. On 6 November 2025 Mustafa Suleyman announced the MAI Superintelligence Team and a goal he calls "humanist superintelligence": systems built for specific domains, with limited autonomy.

The second group matters just as much. Demis Hassabis at Google DeepMind talks about AGI, and Dario Amodei at Anthropic avoids both words. More on them below.

Which labs call it superintelligence, and which avoid the word?

The word splits the field almost down the middle. OpenAI, Meta, SSI and Microsoft use "superintelligence" in public goals. Google DeepMind frames the target as AGI. Anthropic prefers "powerful AI". So the word tells you how a company pitches itself to hires and investors. It says nothing about what the models can do.

Company Term it uses Stated aim Latest dated move Can you build on it today?
OpenAI Superintelligence "A superintelligence research company" (Altman, June 2025) Altman, 5 Oct 2026: "We have not solved alignment" Yes, public API
Meta (MSL) Personal superintelligence A personal superintelligence for everyone (30 Jul 2025) Muse Wonder models, announced week of 12 Oct 2026 Consumer app and paid tiers; check Meta's terms for developer access
Safe Superintelligence Inc. Safe superintelligence One goal, one product NVIDIA partnership, 27 Jul 2026 No. No product released
Microsoft (MAI) Humanist superintelligence Domain-specific systems, limited autonomy Team formed 6 Nov 2025 Through Microsoft products
Google DeepMind AGI AGI, timeline centred on about 2030 Hassabis at Google I/O, May 2026 Yes, Gemini API
Anthropic Powerful AI "A country of geniuses in a datacenter" (Oct 2024 essay) Accord signatory, 29 Sep 2026 Yes, Claude API

Two notes on the table. Every entry comes from the company's own page or a named report opened for this article, linked inline. And "can you build on it today" is the column most lists leave out, even though it is the only one that changes your roadmap.

What has each superintelligence lab actually shipped?

What they've shipped varies a lot. OpenAI, Google and Anthropic sell models through public APIs. Meta's lab shipped its first models and a chart-topping consumer app in 2026. SSI has released nothing, and says it won't until its one product is ready. Microsoft's team works through Microsoft's own products.

Meta is the newest and loudest. According to Fortune's 17 October 2026 profile of Wang, MSL shipped Muse Spark in April 2026, then the Muse app in the US on 8 September. The app passed 2.5 million downloads in about two weeks, per Sensor Tower data cited by CNN. Then came Muse Wonder, the first model Meta calls frontier-class. By Meta's 21 October 2026 earnings call, Zuckerberg said more than 10 million people in the US and Canada had set up Muse accounts, the AP reported. Euronews reported on 13 October that Meta pitched it at law, finance, health and coding, with subscriptions of $20 to $100 a month.

Independent rankings were cooler than Meta's launch. Fortune notes that on the Artificial Analysis Intelligence Index, the top Muse Wonder model is roughly level with the best OpenAI and Anthropic models and a couple of points behind the leader. Frontier, yes. Superintelligent, no.

SSI is the opposite case. Its 27 July 2026 partnership with NVIDIA gives it the Vera Rubin platform and will raise its compute "by an order of magnitude", along with an NVIDIA investment of undisclosed size. Ilya Sutskever's line in the release is plain: "We have research that is worthy of scaling up." No product, no API, no date. We covered the lab in detail in What is Safe Superintelligence (SSI).

Microsoft points to one result instead of a product launch. Suleyman's post says its MAI-DxO diagnostic system scored about 85% on New England Journal of Medicine case challenges, against roughly 20% for the doctors tested. That is Microsoft's own figure, on its own benchmark, and it is the clearest example of what "domain-specific superintelligence" is meant to mean.

How much money is going into superintelligence?

The biggest disclosed number belongs to Meta. On its earnings call of 21 October 2026, Meta narrowed its 2026 capital spending forecast to $135 billion to $145 billion, from $130 billion to $145 billion. That covers all of Meta's infrastructure, not only the lab, but Meta ties the spending to MSL.

From the Q3 2026 call transcript, Zuckerberg described the goal as "delivering personal superintelligence to everyone". CFO Susan Li said MSL investments are building "the technology and infrastructure needed to unlock significant new revenue opportunities". Meta gave no 2027 figure and said it would wait for the Q4 call.

Fortune adds the context. Meta paid $14.3 billion for 49% of Scale AI to bring Wang in. Its 2025 capital spending was $72.2 billion, so this year's range is close to double. And the core MSL research team is small: fewer than 200 people, "closer to 100", in Wang's words.

OpenAI's numbers come from reporting rather than a filing. Fortune's 5 October 2026 interview puts OpenAI's valuation at $730 billion and its reported Q2 revenue at $6.7 billion, with an operating loss. Altman said an IPO won't happen this year.

SSI and Microsoft have not published what their superintelligence work costs.

Is "superintelligence" a real capability or just a label in 2026?

In 2026 it's mostly a label. No company has shown a system that beats humans across the board, and OpenAI's own CEO says the hardest problem is still open. Since 29 September 2026 the label also has an official use: Executive Order 14434 tells US agencies to say "SI" instead of "AI" in their own documents.

The order changed the words, not the tech. As our guide What is SI? Super intelligence, the new name for AI explains, the order keeps the existing legal definition of AI and covers federal agencies only. The same day, Google, Anthropic, Meta, OpenAI, SpaceXAI and NVIDIA signed a voluntary White House Accord on Super Intelligence. Deepak Gupta, CTO of LoginRadius, wrote on 14 October that builders should treat the word as a label for today's frontier models and plan around their unreliable judgment, not their new name.

The labs' own words back him up. Altman told Fortune: "We have not solved alignment." He added that no lab has. SSI's homepage says superintelligence is "within reach", which is a claim about the future.

Think of "superintelligence" like "organic" on a food label. It tells you what the seller wants you to believe about the jar. It does not tell you what is in it. You still read the ingredients, and for a model the ingredients are evals, price, latency and terms.

How close are these companies to superintelligence?

There's no shared yardstick, so "how close" depends on who you ask. Altman wrote in 2025 that "the takeoff has started". Hassabis puts AGI, a lower bar, at about 2030. SSI says it is "within reach" and gives no date. Treat every timeline as a forecast from someone with something to sell.

Here is where the public statements sit, from most to least aggressive:

  1. OpenAI: "We are past the event horizon; the takeoff has started" (The Gentle Singularity, 2025).
  2. Meta: Zuckerberg wrote that glimpses of its systems improving themselves mean superintelligence is "now in sight" (July 2025 letter).
  3. SSI: "Superintelligence is within reach", with no date.
  4. Google DeepMind: Hassabis, speaking at Google I/O in May 2026, centred on about 2030 for AGI, with 2029 possible, per Axios. He described the moment as the "foothills of the singularity".
  5. Anthropic: Amodei's 2024 essay "Machines of Loving Grace" avoids AGI as a term and describes "powerful AI" as "a country of geniuses in a datacenter", per Sherwood News.

For the definitions behind those words, see SI vs AGI vs ASI: what's the difference? and What is superintelligence, and how close are we?.

What does this mean for a founder building on AI models?

Your stack barely changes. Your traffic does. The models you call are the same ones you called before the rebrand. What is new is that there are more assistants sending people to websites, and Meta's Muse app is the latest one to reach millions of people.

Say you run a small AI writing tool on OpenAI's API. The race doesn't touch your prompts or your bill this month. It does shift three things around you.

Your vendor list will keep moving. Meta now has frontier-class models with paid tiers, and SSI has none. A lab's name tells you nothing about whether you can call it. Check for a public API, pricing and terms before you plan around one.

Your marketing copy has a new word available. Some founders will rebrand to SI. Most searches still say AI, so add the new term next to the old one rather than replacing it. We made the case in Should I rebrand my AI startup to SI?.

Your traffic sources are multiplying. Every new assistant with millions of users is a new place your product can be recommended or ignored. Some assistant apps strip the referrer, so those visits land in "direct": the direct-traffic graveyard, where AI referrals go to be miscounted.

Do this tonight

  1. Write down which model providers your product depends on, and mark which ones have a public API you are actually using. Ignore lab names. That list is your real exposure to this race.
  2. Ask ChatGPT, Claude, Gemini, Perplexity and Meta's Muse the three questions your product should be the answer to. Note whether you appear and which page they cite.
  3. Open your analytics and look for referrals from chatgpt.com, perplexity.ai, gemini.google.com, claude.ai and copilot.microsoft.com over the last 30 days. Then check how much of your traffic is "direct".
  4. If you use "AI" in your page titles, add "SI" to one FAQ answer or meta description, not the H1. Test it for a month before going further.
  5. Check that your robots.txt lets the assistants' search crawlers read your pages.

If step 3 comes back empty, it may be your tooling rather than your reach. Clerion reports ChatGPT, Perplexity, Claude, Gemini and Copilot as their own channel, and its free AI crawler checker covers step 5 in about a minute.

Frequently asked questions

Which company is closest to superintelligence?

Nobody can say, because there is no shared test for it. As of October 2026, OpenAI, Google and Anthropic sell the strongest public models, and Meta says its Muse Wonder models are frontier-class. None has shown a system that beats humans across the board.

Is OpenAI building superintelligence?

Yes, by its CEO's own description. In his 2025 post The Gentle Singularity, Sam Altman called OpenAI a superintelligence research company before anything else. In October 2026 he also told Fortune that alignment is not solved, at OpenAI or any other lab.

What is Meta Superintelligence Labs?

Meta Superintelligence Labs (MSL) is Meta's AI unit, led by Alexandr Wang. It shipped Muse Spark in April 2026, the Muse app in September and the Muse Wonder models in October. Its core research team is under 200 people, per Fortune.

Does Safe Superintelligence Inc. have a product?

No. As of October 2026, SSI has released no product or API. It says it has one goal and one product, safe superintelligence. In July 2026 it signed a partnership with NVIDIA to raise its compute by an order of magnitude.

Is Google DeepMind building superintelligence?

Google DeepMind frames its goal as AGI rather than superintelligence. At Google I/O in May 2026, Demis Hassabis centred his AGI timeline on about 2030, with 2029 possible, according to Axios.

Does Anthropic use the word superintelligence?

Not as its stated goal. Dario Amodei's 2024 essay Machines of Loving Grace avoids AGI as a term and talks about powerful AI, described as a country of geniuses in a datacenter. Anthropic did sign the voluntary White House accord in September 2026.

Does the US SI executive order mean superintelligence exists?

No. Executive Order 14434, signed on 29 September 2026, tells federal agencies to write SI instead of AI in their own documents. It renames existing AI and keeps its legal definition. It does not certify any system as superintelligent.

Sources

companies building superintelligencesuperintelligencesuper intelligencemeta superintelligence labssafe superintelligenceai labs
Updated 25 October 2026 · 13 min read

Clerion is website intelligence for AI founders.

Web analytics, AI traffic, conversions, errors and speed from one script tag.

Start a 14-day trialSee pricing

Keep reading

All writing →

Written from real site data by the team building Clerion. Find out whether ChatGPT, Claude and Perplexity can read your site. Check your site.