Nikesh Arora: Mythos is Real, Analytical SaaS is Dead, and Google can be a $10T company

The segment opens by introducing Palo Alto Networks as an outperformer in cybersecurity, led by CEO Nikesh Arora. The hosts note that he is coming up on eight years as CEO. When he started, by their account, the company had a market cap of $17,000,000,000; on the morning of the recording it stood at 238,000,000,000. One host adds, half-jokingly, that once a company passes 100 it is more likely to 10x again, because the first 10x is much harder, so the company is "on your way to a trillion dollars."

The core idea of the conversation is set out in the opening montage and repeated later. Google search democratized information, and by analogy, "AI is democratizing intelligence." The speaker in that clip mentions ten years at Google and says money "is a way to keep track" and is not the goal.

The idea is then made operational. A company with 250 marketing staff producing varied output can now get 90% of that output to be consistent. A company with 5,000 customer-facing employees has a failure mode where customers ask for one person who knows how to solve the problem and avoid another who doesn't. AI could make those 5,000 people act almost consistently. The speaker calls the likely impact on how businesses run "phenomenal." They also note that the last roughly twenty-four months have been exciting, and that a previous speaker, Sarah, had been right to anticipate how much compute would be needed.

The hosts say Palo Alto Networks was one of the few organizations given early access to Mythos. In the transcript the model's name also appears as "MITSOS" and "Methos." The claim is that Mythos showed AI can assess the bad code humans have written over the last fifty years and expose its vulnerabilities. The headline number is the time compression:

In six weeks, we found vulnerabilities which would have normally taken us five to seven years to find.

Asked where the vulnerabilities were, the answer indicates they were in the company's own code base. That prompts the conclusion that Mythos "was not oversold." The speaker also describes a more advanced mode. An "ultra mode" with persistent thinking keeps trying until it gets an answer, and it can "daisy chain vulnerabilities," which means finding new attack paths by linking flaws together. They stress that Palo Alto Networks, being a cybersecurity company, considers itself among the top percentile of companies at testing its own code. The implication is that companies with weaker testing, and the 10,000,000 developers writing code worldwide, are exposed to findings "which would have taken us ten years to find."

On cost, a host asks whether the token bill was $100,000,000 or $10,000,000. The answer is that it was "in the low millions." The speaker expects the cost curve to fall, noting that OpenAI has a cheaper and more consistent model and that Anthropic (transcribed as "Anthropie") has released another model. Pressed on whether this is hype, they insist "The capabilities are real." As further evidence of how seriously the problem is being taken, they cite IBM announcing a $5,000,000,000 project to fix open source, which another participant calls "the biggest problem."