# AI Foundation Labs Pivot to Profitability and Public Markets

**Podcast:** The AI Daily Brief (Formerly The AI Breakdown): Artificial Intelligence News and Analysis
**Published:** 2026-05-21

## Transcript

Today on the AI Daily Brief, huge news out of Anthropic as the company hires a former OpenAI co-founder and has their first ever quarter of profitability.
Before that in the headlines, OpenAI is on the precipice of filing for an IPO.
The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.
All right, friends, quick announcements before we get going.
First of all, thank you to today's sponsors, KPMG, Robots and Pencils, Assembly and Section.
To get an ad-free version of the show, go to patreon.com slash ai daily brief, or you can subscribe on Apple podcasts.
If you want to learn more about sponsoring the show, send us a note at sponsors at ai daily brief dot AI and ai daily brief dot AI is also where you can find out about everything else going on in the broader open AI ecosystem.
We got some juice today, man.
One of the biggest things that markets are watching is the moves that the big labs are making vis-a-vis going public.
And the Wall Street Journal reports that OpenAI has engaged investment bankers and expects to file IPO paperwork as soon as Friday.
They will file confidentially, meaning that full financial disclosure won't be required until much later.
Now, to be clear, if you're not familiar with this process, filing is only the first step in an overall process that will take several months.
SpaceX, for example, filed at the beginning of April and are expected to begin trading in mid-June, and their 10-week sprint is absolutely on the faster end of IPO listings.
Sources said that OpenAI has set the goal of being ready to IPO by September.
The journal noted that the resolution in the Elon Musk lawsuit has cleared the way.
Although unlikely, it was possible that that lawsuit could have unwound OpenAI's for-profit conversion, sending them back to the drawing board on the IPO.
OpenAI's newfound haste could change the IPO race pretty significantly.
Up until now, it seemed like Anthropic might get out first with their reported target of October, but Anthropic is said to be putting together one last private round.
meaning it's unlikely they'll be able to move up their timeline to match OpenAI.
When I was doing my 2026 predictions, one that very clearly I will get wrong was I thought that ultimately Anthropic and OpenAI would not go public this year based on just what a pain public markets can be when you're trying to move fast, as well as the seemingly endless amount of funding in the private markets.
But the circumstances have clearly changed, the constraints on compute are even greater than I had imagined, and so public market access is looking even more important than it might have just a couple of months ago.
Now, there's an interesting question on whether the sequence actually makes sense.
I tend to be in the camp that there is going to be effectively infinite bids for both of these companies.
But the one interesting wrinkle is that if you think these IPOs could actually stretch public market liquidity, which isn't totally inconceivable given the trajectory that they're on, there could be some disadvantage in going last.
Frankly, the fact that the SpaceX IPO is now not a referendum on Grok.
but is going to be much more about Elon settling into this new role as the hereditary Earl of Compute, I actually think makes it much more exciting for public market investors than it was just literally two weeks ago.
Bloomberg's Conor Sen wrote, the same day that it's reported OpenAI wants to IPO as soon as September, we get a Wall Street Journal story with updated impressive anthropic financials.
Doesn't seem like a coincidence.
I like the game theory of three $1-2 trillion companies all about to IPO, trying to beat each other to the liquidity window, and public market investors thinking about the impact on their holdings.
Should make for an interesting summer.
Over in policy land, the major labs have been read in on a new AI executive order that could arrive by the end of the week.
The information reports that the White House's Office of the National Cyber Director held a briefing on Tuesday to discuss the contents of the order.
Sources said that CEOs from OpenAI, Anthropic, and Google were among those in attendance.
Sources also said that the president could sign the order as soon as Thursday.
There's no news as of the time of recording, but of course I'll cover it on tomorrow's show if that signing happens.
Now, there has been a lot of back and forth around what exactly a new AI executive order would include.
This one is reported to be aiming at establishing a voluntary framework for disclosure and testing of new advanced models.
The White House's proposal appears to be that models should be shared with the government 90 days ahead of release.
The labs are reportedly pushing for a much shorter timeline.
with one source stating that the labs want to share their models just 14 days prior to release.
A longer timeline could obviously be an issue for the labs.
It's been less than two months since the first rumors of Mythos emerged, and that feels like an eternity in this industry.
A 90-day review period could pretty significantly slow down the release cadence for the major labs, which has implications for their ability to iterate quickly, correct issues, etc.
Now, currently the rumors suggest that the government review will be voluntary and a relatively minor part of the plan.
But the executive order is also expected to instruct the Pentagon to harden critical systems within 30 days.
In addition, the Treasury will be instructed to establish an AI clearinghouse, which partners Frontier Labs together with critical industries to find and patch vulnerabilities ahead of new model releases.
Details of how the vetting process will work are far more vague.
It seems like about half a dozen agencies are going to be collaborating to create a definition of Frontier models as well as to create a classified benchmarking process.
Rumors state the NSA will be given responsibility for model testing.
placing the framework squarely within the national security portfolio.
Preparation of the framework will be given a 60-day time horizon, so we're unlikely to see model releases going through the new process until the end of the summer.
Based on the rumors, it seems to me that the administration is less concerned with delaying or restricting model releases, instead focused on establishing formal protocols for preparedness.
Functionally, it feels a lot like Anthropics Project Glasswing, i.e.
the way that they've rolled out Mythos, is just becoming a formal part of government policy.
Now, of course, sources noted that the situation is still fluid and the draft order is subject to change.
They also emphasize strong collaboration from the AI labs in dialing in the details, with one source telling Politico, everybody's involved.
That's why it's been on the table and off the table and on the table and off.
It does seem like it's getting more real, however.
Reports are that basically every big tech CEO has been invited to the White House for the signing, so we'll see who actually shows up and if it actually happens.
One interesting story around the compute crunch.
I mentioned this briefly before, but a new offering from OpenAI promises to lock in the supply of AI for rapidly scaling enterprise customers.
The program is called OpenAI Guaranteed Capacity, and it allows firms to make one to three-year commitments in return for discounts and certainty for critical workflows.
The new structure makes AI billing look a lot closer to cloud than to SaaS.
Enterprises will commit to long-term budgets which can be drawn down across various services rather than topping up accounts at the end of the month.
This obviously helps firms get a better handle on their AI budgets, which has become one of the biggest challenges for executives this year.
In April, Uber's CTO made headlines when they said they had burned their entire annual token budget in just four months.
Meanwhile, Box's Aaron Levy says that the problem has been kicked to the CIOs and CFOs, with token strategies becoming a hot topic.
There is also an implicit benefit here if OpenAI hits capacity constraints.
Enterprise customers will presumably be able to nominate critical workflows to receive guaranteed service without interruptions, which feels close to uptime guarantees provided by cloud services.
Wrote Sam Altman, Customers are increasingly asking us for certainty on capacity.
As models get better, we expect that the world will be capacity constrained for some time.
It also helps us plan, so hopefully a big win-win.
For now, OpenAI seems to have enough GPUs, but as I've said before, that does not strike me as a likely permanent state.
And by the way, as a side benefit, This gives some pretty rock-solid ARR numbers heading into an IPO.
Now, one of the ways that companies are looking to handle compute constraints is more efficient models.
And I actually wanted to circle back to Cursor's new Composer 2.5 model as something that has made some major gains on that front.
According to Artificial Analysis, it is now in third place on their Coding Agent Index.
It's behind Opus 4.7 on Max and GPT-5.5 on Extra High, but ahead of both Opus 4.7 and GPT-5.5 on their Medium settings.
What's more, according to artificial analysis, it is 10 to 60x lower cost than the higher effort of 4.7 and 5.5, which could make it a serious contender as cost and token efficiency become one of the biggest constraints for the enterprise.
And one last story on compute, OpenAI is now offering 2 million in tokens to every Y Combinator startup in the current batch in exchange for equity, which I think is less like free AWS credits, and at this point is actually more like headcount cash.
Tyler Bosmini writes, I can't wait to see what's unlocked when you let the most driven, creative, and formidable founders token max.
Allman seemed to agree, writing, I'm excited to see what will happen with token maxing startups, both how they work internally and the products they can build.
Good time to be NYC, but for now, that's going to do it for the headlines.
Next up, the main episode.
One of the most important AI questions right now isn't who's using AI, it's who's using it well.
KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising.
The highest impact users aren't better prompt engineers.
They treat AI like a reasoning partner.
They frame problems, guide thinking, iterate, and push for better answers.
And the good news?
These behaviors are teachable at scale.
If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time.
Learn more at kpmg.com slash US slash sophisticated.
That's kpmg.com slash US slash sophisticated.
Today's episode is brought to you by Robots & Pencils, a company that is growing fast.
Their work as a high-growth AWS and Databricks partner means that they're looking for elite talent ready to create real impact at velocity.
Their teams are made up of AI-native engineers, strategists, and designers who love solving hard problems and pushing how AI shows up in real products.
They move quickly using RoboWorks, their agentic acceleration platform, so teams can deliver meaningful outcomes in weeks, not months.
They don't build big teams.
They build high-impact, nimble ones.
The people there are wicked smart with patents, published research, and work that's helped shape entire categories.
They work in velocity pods and studios that stay focused and move with intent.
If you're ready for career-defining work with peers who challenge you and have your back, Robots & Pencils is the place.
Explore open roles at robotsandpencils.com slash careers.
That's robotsandpencils.com slash careers.
You know Assembly AI for having the most accurate streaming speech-to-text out there.
But they just went a step further and launched a full voice agent API.
The idea is simple.
One connection and they handle everything.
The listening, the thinking, the speaking.
You just stream audio in and get your agent's voice response back.
We're talking about things like outbound sales calls that actually qualify leads.
Here's a harsh truth.
Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized.
Half of companies have AI tools, but only 12% use them for business value.
Most employees are still using AI to summarize meeting notes.
If you're the one responsible for AI adoption at your company, you need Section.
Section is a platform that helps you manage AI transformation across your entire organization.
It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value.
The result?
You go from rolling out tools to driving measurable AI value.
Your employees move from meeting summaries to solving actual business problems.
And you can prove the ROI.
Stop guessing if your AI investment is working.
Check out Section at sectionai.com.
That's S-E-C-T-I-O-N-A-I dot com.
Welcome back to the AI Daily Brief.
Today we are discussing a set of news that I think in many people's estimation has actually reset their sense of where we are when it comes to AI.
Anthropic has been surging for a while now, but after this week's news, it is not unreasonable to ask whether we've entered some new endgame.
The first story dropped on the same day as Google I.O.
and was certainly, for the AI builder and insider community, much more significant than anything that happened at the event.
On Tuesday morning, renowned AI researcher Andre Karpathy, one of the founding team members of OpenAI, announced that after a couple of years of being solo, he would be joining an AI lab once again.
Except this time, it was Anthropic.
Personal update, Andre tweeted, I've joined Anthropic.
I think the next few years at the frontier of LLMs will be especially formative.
I'm very excited to join the team here and get back to R&D.
I remain deeply passionate about education and plan to resume my work on it in time.
So for those who aren't familiar, let's do a quick background on Andre.
Like I said, he was initially one of the founding members of OpenAI way back in the day when it was just a non-profit with a dream.
A couple years later though, he was lured away by none other than Elon Musk.
In recently released documents that were part of the court case, we get an email from Elon to Tesla's Jim Keller that reads, Just talked to Andre and he accepted joining as director of Tesla Vision, so anyone working on neural net software would report to him.
Andre is arguably the number two guy in the world in computer vision after Ilya.
The OpenAI guys are going to want to kill me, but it had to be done.
Then in 2022, however, Andre came back to OpenAI and was there through the pivotal early chat GPT and GPT-4 period, finally leaving again in mid-2024.
as part of the executive exodus that happened in the wake of the Sam Altman firing and rehiring.
At the time, Andre announced that he was going to be starting an AI education company called Eureka Labs.
Writing that with the progress that had recently come in generative AI, the ability to design a totally different type of educational experience was right there waiting to be seized.
Over the last couple years, though, we didn't really hear much about Eureka Labs, but Andre never left the conversation.
You might have heard about him on this show as the person who coined the term vibe coding back in February of 2025.
There's a new kind of coding I call vibe coding, he wrote, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.
Now, recently, there were some interesting indications that Andre might be getting antsy about his position outside of the arena.
When VC Sarah Guo asked what she should ask Andre on her upcoming podcast conversation with him back in March, OpenAI's Noam Brown wrote, Why is he not at a frontier AI lab at the most pivotal time in human history since at least the Industrial Revolution?
And one of the things he said in that conversation was basically that anyone outside the labs would inevitably start to drift away from the frontier.
So with those tea leaves, perhaps the move was in retrospect telegraphed, but the choice of Anthropic hit people with the force of a freight train.
By the way, Noam Brown, to his great credit, classily retweeted Andre's announcement post saying, I would have loved for him to rejoin OpenAI, but I'm happy he's at any frontier lab pushing the field forward.
It's easy to frame this as zero-sum among the labs, but in truth, we're collectively advancing the most important tech of our era.
However, for lots of folks, that was not the main takeaway, and the zero-sum thinking was exactly where they were headed.
Capturing my feelings, Signal wrote, This is bigger news than Google I.O.
Aish wrote, Worth noting, Jan Leakey left open AI for Anthropic, John Shulman left open AI for Anthropic, now Carpathy.
That's a serious pattern.
Anthropic isn't just building good AI products anymore, they're assembling an absurd research bench.
Signal again wrote, This is like a leading franchise recruiting someone who's simultaneously the best player and the league's best broadcaster and its most-watched developmental coach all in one.
In terms of what Andre is doing, Nicholas Joseph from Anthropic wrote, Now what makes this interesting is that this idea of recursive research is something that Andre has been talking about a lot recently.
Back in March, we talked about his approach to auto-research, where AI agents do a research run in a continuous loop, as something that was both demonstrative of as well as advancing.
this loop-type pattern that we were starting to see spring up in agent work all over the place.
And for some of the more sophisticated observers, this was less about any horse race dynamics, and much more a statement about how close we were to what's referred to as recursive self-intelligence, where the AI actually does the research that improves the AI.
TMT Longshort wrote, Seeing concern now that this carpathy move indicates anthropic already won and is therefore bearish for the other labs.
My take is this indicates that we are close to RSI and therefore an acceleration in model IQ increases.
He then went on to talk about some of the market implications.
In that scenario, the value of compute is going to explode as supply chain scale-ups are linear while demand creation is non-linear.
Anyone with compute is sitting pretty regardless of lab talent.
Full stop.
Every GPU will explode in value.
If you can run a million von Neumann in a data center, we will quickly have AI inventing use cases for token consumption faster than we can supply them.
New fields of science, reverse aging, the Gunasphere.
It all gets pulled forward and it will all require compute.
Shanu Matthew wrote, I had a similar initial read.
This is a move you do to try and accelerate to the endgame at this point in the cycle.
RSI must be in sight for the labs, at least they believe, so they are locking in Talon and compute as necessary to take that final sprint.
Altimeter's Frida Dwan had an interesting take about whether we might start to see a break in the pattern where each of the labs takes its turn being at the state-of-the-art for a few weeks before the next lab comes in and pushes things forward again.
Frida wrote, We're used to state-of-the-art rotating every few months.
One lab pulls ahead and another catches up.
Gemini was state-of-the-art literally two quarters ago.
Feels like ages.
But RSI, recursive self-improvement, could change the competitive dynamics.
When Anthropics Dario Amadei was on Dwarkesh's podcast back in February, Dwarkesh challenged him.
If recursive improvement is real, why does state-of-the-art keep rotating between labs?
Dario's answer was that, until very recently, the compounding advantage from AI-assisted AI research was still too small to matter, but it's changing.
Since January and February, I've heard more and more researchers talk about RSI.
The idea is simple.
You build a better AI, that AI helps you build the next better AI, and the loop starts compounding.
It also explains part of the current token maxing dynamic.
Big companies spending billions of dollars on cloud code just to keep up.
And it helps explain the rumors that OpenAI and Google are reorganized around coding as the top priority.
Once that loop crosses a certain threshold, it starts to look like a true industrial revolution.
Horses were never going to catch up with cars.
And for some, this shift is now a foregone conclusion.
Latent Space's Swix writes, RSI is here.
Jesus.
Anthropic, though, was not done shocking the industry.
Wall Street got their hands on recent financials that were given to investors in the current fundraising round that show not only that Anthropic's numbers continue to grow, but that they're actually expecting to turn a profit this quarter.
Anthropic is currently forecasting $10.9 billion in revenue for Q2 and an annualized rate of $44 billion.
Last quarter, Anthropic closed the books with $4.8 billion in revenue.
They also expect a small operating profit of $559 million.
Now, this isn't just a first for Anthropic.
It's the first profitable quarter for any foundation lab.
Coming into this year, Anthropic had forecast that they wouldn't be profitable until 2029, while OpenAI wasn't expecting profitability until 2030.
Now, there are a few caveats and provisos here.
Anthropic, for example, has a slightly different accounting process than, for example, OpenAI, where reports are that they count top-line revenue before partner shares.
as Anthropic revenue, which potentially inflates the total number that actually makes it into Anthropic's coffers.
And honestly, maybe the bigger asterisk is, they probably wouldn't be profitable if they had their druthers.
In other words, part of why they are profitable is that Compute is so sold out that they couldn't spend more even if they wanted to.
Still, frankly, this pretty dramatically challenges the math once again for AI skeptics.
This year's shifted goalpost as revenue has grown was that even if revenue was growing, profitability at this scale was basically impossible.
One of the loudest AI skeptics literally wrote a post to this effect this week.
For others who aren't pot committed to their skeptical position, however, this moment is a bit of a reset in expectations.
Journalist Derek Thompson wrote, Mr.
Radable, in an ironic tweet, pointed out that it absolutely in no way makes sense to think about Anthropic as some startup anymore, that these numbers are massive not on a startup scale, but on a company scale in general.
He tweeted, How is it possible for Anthropic to be profitable despite only having more revenue than Workday, ServiceNow, Palantir, and Snowflake combined?
On a recent podcast appearance, Gavin Baker made the point that an uncompute-constrained Anthropic could do even more than $100 billion.
Reflecting on the difference between the telco bubble back in the late 90s, and the infrastructure build-out today, Gavin wrote, Back then, supply could largely keep up with demand because there was massive underutilized wafer capacity coming out of the 1998 Asian crisis that could ramp quickly.
The relatively quick supply response led to an overbuild, which caused the crash as the overbuild was largely debt-funded, which required an immediate ROI for the clerks to service their debt.
There is no comparable slack in the system today.
Leading-edge wafers in power are both structurally constrained, and neither can be turned on in a matter of months.
That's the core difference in my opinion.
And obviously, the largest buyers of Compute will have no trouble servicing their debt anytime in the near future.
Now, coming in over the top to reinforce these points was NVIDIA's Wednesday night earnings, which delivered a record quarter and beats across the board.
Revenue came in at $81.6 billion, beating estimates of $78.9 billion.
And earnings per share also outperformed, coming in at $187 per share ahead of estimates of $167.
Data center revenue grew at a 92% pace, which was up 21% on the quarter.
This is the first quarter we've seen Blackwell revenue firing on all cylinders, and it's a return to form for NVIDIA.
This earnings call felt similar to the mind-boggling period in 2024, when Jensen Huang presented earnings to a stadium full of investors.
It's also the first time that NVIDIA has separated out data center revenue between hyperscalers and other customers.
46% of revenue came from hyperscalers, growing 12% quarter over quarter, with NVIDIA also saying that they are gaining market share among the hyperscalers, putting to bed the idea at least for now that Google's TPUs are somehow eating into NVIDIA's lead.
Bloomberg presented this as the earnings call that silences skeptics and reinforces that AI is going mainstream.
In his statement, Jensen said, Build out of AI factories, the largest infrastructure expansion in human history, is accelerating at extraordinary speed.
Agentic AI has arrived, doing productive work, generating real value, and scaling rapidly across companies and industries.
NVIDIA is uniquely positioned at the center of this transformation as the only platform that runs in every cloud, plowers every frontier and open-source model, and scales everywhere AI is produced.
Compounding how dramatic the growth is, Huang reinforced that NVIDIA is selling zero chips in China and doesn't expect to return to that market in a meaningful way.
In a side interview with CNBC, he said, the demand in China is quite large.
Huawei is very, very strong.
They had a record year.
They'll likely, very likely have an extraordinary year coming up, and their local ecosystem of chip companies are doing quite well because we've evacuated that market.
We've really largely conceded that market to them.
And although NVIDIA squeezed the spotlight for a moment, Huang couldn't avoid discussing Anthropic.
This year, he said, we had the benefit of winning Anthropic.
We're helping them scale capacity so they can have more reach, more revenues, and grow their company.
With Anthropic, we're scaling very, very quickly.
We've got big plans for them.
Surprisingly, though, the stock plunged by 3% in after-hours trading despite the incredibly strong report.
Analyst Patrick Moorhead had this to say about that.
On the stock reaction, he wrote, The challenge with NVIDIA at a 5 trillion plus market cap is that investors do not know how to deal with it.
If you take Jensen's $1 trillion forward demand pipeline from GTC at face value, this is an $8 or $9 trillion company.
People are afraid to move even on exceptional earnings.
In other words, we are simply in uncharted waters.
And still somehow after all of this, Anthropic was not done for the week.
It was a huge deal recently when a new alliance between Elon Musk's SpaceX and Anthropic made headlines with Elon announcing that Anthropic would be using the Colossus 1 data center to expand Anthropic's compute capabilities.
On Wednesday, however, Anthropic Chief Compute Officer Tom Brown announced that that partnership would be deepening, with Anthropic scaling up GB200 capacity in Colossus 2 throughout June.
Colossus 2 is a much larger data center than the original Colossus, and very quickly it's beginning to look like a significant portion if not all of Elon's short-term compute resources are being poured into scaling Anthropic.
Separately, the SpaceX IPO filing has revealed just how much Anthropic is paying to take over Colossus.
The filing said Anthropic had agreed to pay $45 billion over three years.
That's about $1.25 billion a month or around $15 billion a year.
Capacity will begin ramping over May and June at a reduced fee.
And the deal instantly makes Colossus the biggest revenue generator for SpaceX, overtaking Starlink's $11 billion in 2025.
Just this contract alone adds 80% to SpaceX revenue based on last year's figures.
And that's not even considering the new contract for Colossus 2, which should add substantially more to the top line.
It is entirely possible that Claude will quickly overtake the Tesla Model 3 as the single biggest revenue driver in Elon Inc.
Musk, for his part, clearly understands that he is now in the anthropic business and is settling quite quickly into his role as compute czar.
He tweeted, As the recently expanded partnership with Anthropic demonstrates, SpaceX is offering AI compute as a service at significant scale.
We are in discussions with other companies to do the same.
Over time, especially with orbital data centers, we expect to serve AI at extremely high scale.
Signal reflected on the impact for OpenAI, calling it really, really bad news for them because, quote, Now your largest and now actually profitable competitor with the most momentum is in a deep partnership that empowers them with potentially infinite compute.
And since they make money, they can pay for it directly without using equity.
And Elon can't afford to walk away from this partnership either because as a publicly traded company, that type of revenue fluctuation would have fiduciary impact.
Incredible twist.
Still, for most, it was just the capstone on a massive moment of AI acceleration.
Investor and Bloomberg opinion columnist Connor Sen wrote, the Anthropic IPO won't be for less than $2 trillion.
Just a head spinning amount of news, but that is going to do it for today's AI Daily Brief.
Appreciate you listening or watching as always.
Until next time, peace.