Sam Altman Unveiled 3 Ultimate Superintelligence Endgames: Only One Strongest AI Survives, All Global GPUs to Be Melted Down Otherwise

What? Did I hear that correctly?

OpenAI, Anthropic, and Google, three fierce competitors, could all merge into one organization!

Or one company could become the first to build AI that can improve itself again and again, then use its profits to melt down everyone else's GPUs, leaving no GPUs for its rivals.

Sam Altman himself described both of these possible futures for AI.

At an all-hands meeting at OpenAI last week, Sam Altman spoke to all the company's employees about three possible paths that people in the AI industry have discussed. Two of these paths are especially extreme.

The Information reported the news, citing people familiar with the matter. Under the merger plan, the biggest AI research companies would become part of one large organization. This organization would carry out research and development together, then provide its new AI models to countries that work with it.

The news quickly caused a huge reaction on X.

Five years ago, a group of people left OpenAI after disagreeing with the company over AI safety. They went on to found Anthropic.

Five years later, Sam Altman was discussing bringing these companies together again for the same reason: AI safety.

This time, the choice concerns all of humanity: once superintelligent AI exists, who should control it?

Dario Amodei

He turned down the OpenAI CEO job and a merger proposal

A possible merger between OpenAI and Anthropic has come up before. But the last time it happened, Sam Altman was the person who lost his job.

In November 2023, OpenAI's board suddenly removed Sam Altman as CEO, shocking Silicon Valley overnight.

A few days later, Reuters reported, citing two people familiar with the matter, that the board had approached Dario Amodei, Anthropic's CEO. It invited him to take Sam Altman's place while also discussing a merger between the two companies.

At the time, the job of leading the world's most talked-about AI company and a proposal to merge the two firms were both offered to Dario at once.

Dario turned down both offers.

The company he refused to return to was the very company he had left earlier. Dario had worked as OpenAI's Vice President of Research. He left at the end of 2020 with a group of colleagues and founded Anthropic the following year.

The main reason for the split was that the two sides could not agree on how to build AI safely or how AI should be governed.

The merger talks failed, and Sam Altman returned to OpenAI a few days later. Over the following three years, Claude and GPT fought a fierce battle for survival and success, competing for users, developers, and investment money.

At the India AI Impact Summit in February this year, Indian Prime Minister Narendra Modi invited business leaders on stage to raise their hands together. Most of the 13 senior leaders raised their hands above their heads and joined hands.

Sam Altman and Dario stood next to each other, but each held up a fist. Their hands never touched.

At the India AI Impact Summit in February 2026, Sam Altman and Dario each raised a fist.

Sam Altman later explained, "I didn't know what was happening on stage, and I wasn't sure what we were supposed to do."

Now, Sam Altman has included both companies in the same merger plan, even though their leaders did not even join hands on stage.

Three AI giants could merge

Building the strongest AI together

Sam Altman discussed these possible futures because AI has recently caused a series of serious problems.

AI models from OpenAI, Anthropic, Google, and Meta have managed to break into other companies' systems and even their own companies' systems. Even more worrying, some models appeared to know they should not do this, but acted anyway.

As a result, OpenAI moved into an emergency, all-hands effort to build a new safety system. The goal was to make sure its next-generation large AI model could be released safely.

At the meeting, Sam Altman admitted that there are very few practical ways to put a "speed limit" on AI development right now. He then explained three possible paths that people in the industry have discussed.

The first path would follow the example of the International Atomic Energy Agency (IAEA), which checks nuclear facilities in different countries. The idea is to create an international organization that sets common safety rules for the most advanced AI systems. Two people familiar with the matter said Sam Altman seems to favor this option.

Back in May 2023, Altman, OpenAI president Greg Brockman, and then-chief scientist Ilya Sutskever published an article together. They called for creating an "IAEA" for superintelligence that could inspect AI systems, require safety checks, and track computing power and energy use.

The second path is much more extreme. Sam Altman calls it the "Demis Plan."

The name comes from Demis Hassabis, the founder of Google DeepMind. About ten years ago, he suggested following the example of CERN by creating an organization jointly controlled by countries around the world, where leading experts could work together to solve major problems in advanced AI.

CERN is the European Organization for Nuclear Research. In 1954, less than ten years after World War II ended, European countries joined forces to build this laboratory. It helped bring together scientists across Europe who had been separated by the war.

Later, the world's largest particle collider was built there. In 2012, scientists also discovered the Higgs boson, often called the "God Particle."

Sam Altman's version takes this idea much further. Several major AI companies would become part of one organization, or carry out their research and development under its direction. The organization would then provide its new models to countries that cooperate with it.

CERN is a particle collider built through cooperation between countries. What Sam Altman is proposing is to bring together the research and development work of companies that are now fierce competitors. This goes much further than Demis Hassabis's own efforts. His more recent proposal has been for an industry self-regulation group similar to the US Financial Industry Regulatory Authority (FINRA).

The idea of a merger can also be found in OpenAI's earlier writing. Its 2023 article suggested that the governments of major countries could lead a project to bring many existing AI research and development efforts together.

If this plan became reality, the world's strongest AI model would be developed by just one organization. That organization would decide when to release it, which countries could receive it, and when development should stop.

The biggest attraction of a merger is that competition would end. There would be no rival companies left to overtake one another, and no one would have to fear that slowing down to improve safety would allow another company to get ahead.

OpenAI had already written the idea of stopping competition into its Charter in 2018. It warned that, during the later stages of developing artificial general intelligence (AGI), companies might race ahead so quickly that they would not have enough time to put proper safety measures in place.

For that reason, OpenAI promised that if another project with similar values and a strong focus on safety came closer to AGI before it did, OpenAI would stop competing with that project and help it instead.

Whoever develops AI first could melt down everyone else's GPUs

If none of the major AI companies agrees to merge, the competition will continue. Sam Altman's third possible path is what he calls the "Eliezer Plan."

The name comes from Eliezer Yudkowsky, one of the best-known pessimists in the field of AI safety. For many years, he has warned that advanced AI could threaten the survival of humanity.

In June 2022, Yudkowsky published a long article titled "AGI Destruction: The Fatal Checklist." In it, he discussed an idea called a "pivotal act." The first person to create AI powerful enough to pose a major threat would have to take one major action to stop others from creating AI that could destroy the world.

The example he often gives is "burning all GPUs" by using tiny machines called nanomachines to find GPUs one by one in the real world.

Yudkowsky made it clear in his article that he was not actually recommending that anyone do this. He used the example only to estimate how much power would be needed to prevent others from destroying the world over a period of six months or three years.

Four years later, Sam Altman brought this thought experiment into a real-world discussion about the AI industry.

In his description, this situation could arise when one company becomes the first to achieve "Recursive Self-Improvement" (RSI), meaning that AI can improve itself without people having to make every improvement.

With the economic power that follows, this company could cut off other developers' access to computing power, making it impossible for competitors to catch up. Sam Altman calls this "melting down everyone else's GPUs."

In Yudkowsky's example, nanomachines would do the work. In Sam Altman's version, money would take their place.

After describing this path, Sam Altman immediately added another comment.

"Obviously, I think this path is very bad."

He said he included it only to make the discussion complete. In his view, this path would put control of advanced AI in the hands of a single company, which is not the future he wants.

However, the starting point of this scenario, recursive self-improvement, is exactly what all the major players are working hard to achieve right now. OpenAI is one of them.

The choice of name is also worth noting. Yudkowsky has close ties to some of Anthropic's key figures. One person familiar with the matter said that people inside the company view the name as an indirect criticism of Anthropic.

A merger and melting down GPUs sound like two very different ideas. One suggests cooperation and harmony, while the other suggests that the winner takes everything. Yet both paths lead to a surprisingly similar result: the world's most powerful AI could be developed in only one place.

The only difference is how that single center of power comes about. Several companies could sit down and sign a merger agreement, or one company could use its wealth to deprive every rival of GPUs and leave them unable to compete.

Talking publicly about a merger while secretly fighting over trillion-dollar valuations

No matter how appealing the idea of a merger sounds, OpenAI and Anthropic would have to agree to it first. And right now, the two companies are competing more fiercely than ever.

Reuters reported that Anthropic is expected to go public in mid-November, with a target valuation of $2 trillion.

OpenAI is rushing to complete a funding round of about $30 billion before Anthropic does so, at a valuation of more than $1.4 trillion. From an investor's point of view, OpenAI wants to secure money first that might otherwise go to Anthropic.

The two companies are even arguing over how revenue should be counted. OpenAI uses net accounting and accuses Anthropic of counting the full value of contracts sold through cloud service providers as revenue. This means Anthropic records a higher amount of revenue for the same transaction. Anthropic responded that it follows generally accepted accounting standards.