Byron Deeter says the frontier labs will keep the leadership and most of the economics, and that the application layer built on top of them is where the next set of large companies gets made.
The reflex read of a lab calling for a slowdown is that its investors must be unhappy. Deeter, who has money in Anthropic, argued the opposite on air: a managed pace is worth more to a shareholder than a sprint that ends in a forced stop.
"I think the worst thing would be to sprint into this and then have to do a hard stop, because things get out of control."
Deeter is a partner at Bessemer Venture Partners and an investor in Anthropic, so he is being asked to defend a position that costs his own portfolio company short-term revenue.
The full segment is covered here so you can skip it.
Here are the 6 arguments that matter.
👤 Guest: Byron Deeter, Partner at Bessemer Venture Partners and an investor in Anthropic
🎙️ Hosts: Carl Quintanilla, David Faber and Sara Eisen, the anchors of CNBC's Squawk on the Street
👥 Also on: SambaNova co-founder and CEO Rodrigo Liang, in a separate segment of the same hour
📰 Published: 14 September 2026 on the Squawk on the Street podcast feed (CNBC)
🟣 Apple Podcasts | 🔗 Segment page | ⏱️ length not available
Key Takeaways
The word the industry has settled on is pacing, and Deeter says it is not the same as a pause
He puts inspectors and transparency in the same category, and "pause and criminalize and halt" in another
An Anthropic investor is arguing that giving up short-term economics is the shareholder-friendly choice
He expects government oversight and more regulation, but not government ownership
The globalization pillar of Amodei's essay is the one he thinks will not land
He does not expect China to voluntarily stop trying to leapfrog the US
Going public increases oversight, so using safety to argue against an IPO has the logic backwards
The frontier will not consolidate to two or three model providers, which is where his portfolio sits
He points to the software index recovering as evidence the application layer was oversold
1. Pacing, Not a Pause
The desk set the segment up with a prediction-market number: the odds of an Anthropic IPO announcement before 1 November had fallen to 54% on Kalshi, from about 75 a week earlier. The first question was simply what Deeter made of the weekend.
"It's been wildly entertaining," he said, before arguing there was a real conversation underneath the noise. Everyone has anxiety about the rate and pace of the technology, he said, alongside genuine excitement about 200-year-old math problems being solved, incurable diseases being addressed, and the economy being built.
"But fundamentally, we need to be responsible in how this technology is rolled out."
The distinction he drew is a vocabulary one, and he thinks the industry has now settled on it: "And I think the critical terminology that the industry is adopting is this notion of pacing and the idea of inspectors and transparency."
"And I think that those actions are very different from some of the extreme notions of pause and criminalize and halt and these sorts of things, which would be bad policy both politically and economically."
Asked whether he finds the pacing language legitimate and sincere, he said he does, and read it as the industry converging: the essay from Dario Amodei, then "Sam Altman at OpenAI, Elon Musk at SpaceX," and "Demis at Google, we hope and expect, you know, someone like a Sundar would come in shortly on behalf of Google as well." Microsoft, he said, has said positive things on some of these themes.
His precedent is the public-private partnership rather than anything new. "We see these public private partnerships a lot. We saw it in COVID with the pharma companies." He named defense technology and oil companies now, and banks in the financial crisis: "Back in the global financial crisis, we saw it with banks where the industry can come together in important critical times, work with the government in constructive ways, keep the pace, do what's needed, but make sure first and foremost that the public is safe."
2. Oversight, Not Ownership
A host pointed out that the precedents he had just cited generally involved the government taking a stake, and asked whether that is what he envisions.
"I don't envision government ownership. I envision government oversight and deep engagement, and we're going to see more regulation here."
The line he does not want crossed is competitive rather than regulatory. What the country cannot do, he said — and here he agreed with the president's statements over the weekend — is yield national leadership in AI, inside private industry and globally, for national-security reasons. "It is imperative."
"And so those are the bounds we need to play with it. Don't put the public at risk, but continue to push the frontier for all these positives and for our own defense."
3. The Investor's Trade-off
The sharpest question of the interview was put to him as a conflict: as an Anthropic investor, is what Amodei is leading not directly contrary to what a shareholder wants from an IPO and from growth?
"I think it's a long term view versus a short term view, which is that long term responsible pacing will yield great outcomes" and will do it in a way that avoids fits and starts.
"I think the worst thing would be to sprint into this and then have to do a hard stop, because things get out of control."
His image for it is a driving one: "If we tap the brakes, this is the car going into a corner where you tap the brakes, you're able to enter in a responsible way and accelerate faster on the back end." That, he said, is what the industry is doing now.
The reason it has to be collective is that a single firm slowing alone is just losing. The industry has to work together, he said, because adversaries will not embrace it the same way.
"But I think giving up a little bit of short term economics makes not only prudent sense for humanity, but also long term business sense to do it in a responsible, managed way that enterprise customers are going to trust and that the government will trust."
4. Open Source and China
Asked how open-source models fit, given that plenty of them are Chinese and investors have been trying to work out what that means for Anthropic's and OpenAI's businesses, Deeter split the answer in two.
"Yes, I think we're going to continue to have a world where the frontier models drive the majority of the leadership and capture majority of the economics, but the open source models are going to have a critical role in this ecosystem," as they have in earlier software waves.
His condition is that the US open-source efforts — he named Nvidia's, and many private businesses — sign up to the same rules: "They need to have the same responsible guardrails, and they need to be subject to the same transparency and engagement."
On the part of Amodei's argument he does not expect to work: "It's highly unlikely that in the final pillar of Dario's essay, the globalization goals are met. We do not reasonably expect China to voluntarily back down from their attempts to leapfrog us."
He also thinks the technology polices itself better when there are more of them. "The other thing is that this technology will help to manage and govern and police." "And so having multiple agents, multiple platforms is good not only for competitive markets, but also for competitive insights and for some notion of group policing, which I think will happen in a rational way."
5. Going Public Is the Fix
Sam Altman had said there should be no AI IPOs this year and that one would be ill-advised given safety concerns. A host asked whether an IPO had ever been the plan.
"I don't think many people are surprised by the fact that they're not going public this year. They're not ready for many reasons." "And so it's just it's amusing that he's trying to then throw a wrench in anyone else's plans by extension."
The substantive objection is that the safety argument points the other way: "But the transparency argument is exactly backwards." Going public forces reporting and accountability.
"That's why many companies have stayed private longer is they've wanted to avoid that." Listing, in his framing, is "a positive step in this notion of letting the public see what's happening, letting the government be more involved, being subject to immense reporting, compliance, regulatory oversight."
A host pushed back on the signal it sends to buy shares in a company saying it cannot control its own pace without outside help. Deeter took it as two messages at once. "So I think it's saying both, which is we are open to the transparency of public reporting, which is by definition more oversight."
"And so I think it's openly embracing more oversight in both areas. And I welcome the rest of the industry to do the same."
6. A Thousand Flowers
The last question was whether the past five or six days have changed how he invests. The answer was the closest thing in the segment to a portfolio position.
"Yeah, the competitive dynamic continues to heat up," he said, and put the foundation-model news next to what is happening in the application and platform stack.
"We continue to believe that there will be a thousand flowers that bloom on top of these great foundational layers. The world is not going to consolidate around 2 or 3 model providers."
His market evidence is the recovery in software stocks: "You've seen a little bit of a resurgence in the software index, which shows that people once again, think that systems of record and vertical businesses and api businesses have fundamental value, that they may have been a little aggressive in their overselling."
"We also believe that we're going to see this next wave of AI powered applications that build on top." The agentic version of that, he said, becomes available to businesses and consumers in a powerful way, and creates companies.
Bonus Insights
On the consumer side of the agent question, Deeter said Perplexity is in that world as well, and framed the opportunity as a question rather than a product: "How does everyone have an assistant in their pocket?" He added that "Apple's upcoming release promises to have much more capabilities here."
He tied the agent argument back to employment, saying the leading companies in this area are adding jobs rather than cutting them, because the pairing of a person with the technology is what produces the output. He began to extend the point to medicine as the segment ended.
Deeter's bottom line is that pacing is a shareholder argument, not a concession: a managed slowdown with inspectors and public reporting protects the value of the frontier labs, and the money is in the many companies built on top of them rather than in a market that narrows to two or three model providers.
Products, Companies & Tools Mentioned
Anthropic (The company he invests in, whose chief executive wrote the essay that started the debate; the desk noted Kalshi's odds of an IPO announcement before 1 November had fallen to 54% from about 75 a week earlier)
OpenAI (Sam Altman backed the pacing statements and separately said there should be no AI IPOs this year, which Deeter called amusing)
Google and Microsoft (Deeter expected Google to add its voice shortly and said Microsoft has said positive things on some of these themes)
Nvidia (Named as a US open-source effort that he says must accept the same guardrails and transparency as the frontier labs)
Perplexity (His example of a consumer-facing agent business)
Apple (Its upcoming release, he said, promises much more capability in the assistant-in-your-pocket category)
Bessemer Venture Partners (Deeter's firm, and the source of the "thousand flowers" view of the application layer)
Kalshi (The prediction market the desk used for the Anthropic IPO odds that opened the segment)
Books & Resources Mentioned
We Must Pace the Frontier – Dario Amodei (The essay Deeter says led the cause; he singles out its final, globalization pillar as the one unlikely to be met)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: