The CoreWeave-OpenAI-Microsoft drama explained
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
📈 Trending Up: Tariffs … jobs anxiety … bribes … confusing the scrapers … cleanup at Apple … digital euros? … Claude … breaking the law …
📉 Trending Down: Heathrow throughput … the Department of Education … crypto regulation … Nike … Accenture …
The DoorDash-Klarna brouhaha isn’t actually very funny
There’s a great meme on Twitter that makes fun of people who complain about the rising cost of living while making interesting financial choices. Summarized:
Person 1: Inflation is bad.
Person 2: Inflation is bad, or you ordered a private taxi for your burrito?
Person 1, with shame: Private taxi for my burrito.
The gist is that if you are hiring a restaurant to operate and prepare food for your consumption, a car, and driver to prepare you a single meal and deliver it to your house, it’s not going to be cheap. The era of subsidized food delivery is behind us, and Uber Eats makes lots of money now (on an adjusted EBITDA basis, but you do a better job sorting out Uber’s segment results.)
You can understand the discourse that cropped up concerning DoorDash (a food delivery giant) and Klarna (a consumer lending company) joining forces. The tweets write themselves: Now you can finance your private burrito taxi! Or, How long until someone creates synthetic burrito CDOs? You get the idea.
Sure, it’s weird that people might turn to a buy now, pay later vendor to finance a dinner delivery, but it’s hardly a unique concept. To wit:
Instacart and Klarna have a deal that allows customers to spend up to $1,000 on the grocery delivery service using the BNPL service as creditor.
Affirm recently lost a deal to provide Walmart customers with credit to Klarna.
You can use EBT to order groceries on Uber Eats
I could go on, but you get the point.
What matters is that consumer debt accretion is a worrying trend:
Credit card writeoffs are around their highest level since 2011
While American household debt reached a record of more than $18 trillion earlier this year
You can find countervailing data, but with businesses like Delta, Target, and other companies carping about consumer stress have led to some worrying that everyday folks might be “maxed out.”
Sure, we can yell about burrito taxis and how some folks might not be making great financial choices. I think the more interesting take is that we’re seeing a large number of people discover that the fully-loaded cost of existing as a middle-to-upper-middle-class household is probably in excess of most people’s income.
Musk-China-Trump
Lord save us all.
The latest 47 mess goes like this: The New York Times reported that the “Pentagon was scheduled on Friday to brief Elon Musk on the U.S. military’s plan for any war that might break out with China.” The Times clearly stated that it had two sources confirming the meeting and its focus, another source confirming the meeting and its theme (China) but not its details (war prep), and a final source that merely confirmed that Musk was heading to the Pentagon.
On one hand, given that Elon Musk has seeming carte blanche to do whatever he wants — including dropping in on Presidential calls to other premiers — why wouldn’t he go collect a briefing on possible war plans with a trading partner the nation is currently at odds with?
On the other hand, Musk has huge business ties to China, is careful not to criticize China’s government, and we’re watching China ratchet up its saber-rattling in the South China Sea.
Musk, who is currently working to keep his car company’s share price afloat while also gutting domestic government, was quick to lash out at the report. Trump weighed in, too. The Secretary of Defense also pushed back. Then, the Times writes, “Roughly 30 minutes after that social media post, The Wall Street Journal confirmed that Mr. Musk had been scheduled to be briefed on the war planning for China.” (Link added for your benefit.)
Well!
You can carp about the Times all you want, but a second major newspaper in quick succession with multiple sources confirming the same thing is signal. The big paper reporters do not fuck around. The entire saga could hinge on how we define briefing, but more importantly, Musk’s wings are getting toasty:
An all-hands at Telsa didn’t move the needle much.
The Trump administration is publicly telling people to buy Elon’s cars, and his stock.
And now, his business in China — where Telsa makes about half its cars — is at risk because no matter how much of a pass Musk has received in the nation, if he gets into a national security spat between Beijing and Washington, it won’t turn out well for the businessman’s business interests.
The mixing of business and politics is odious. Pretending that one is not mixing business and politics and acting shocked when your actions have consequences is even more whiffy.
Perplexity’s big round comes into focus
A little while ago, The Information reported that AI search upstart Perplexity could raise new capital at a $15 billion valuation. More recently, Bloomberg put the figure at $18 billion, implying winsome investor interest in the investment.
With a potential $500 million in the offing, how big is Perplexity today? Bloomberg reports that the “company’s current annual recurring revenue is nearly $100 million.” That’s very impressive in raw terms, but not that much when we do simple math and realize that Perplexity could be about to raise at 180x ARR.
That’s expensive, even for the AI era. It’s a multiple greater than what OpenAI or Anthropic command today, and is in excess of what Cursor is looking at.
Recent reporting pegs the Anysphere, the company behind Cursor, as looking for nine-figures at a $10 billion valuation after reaching $100 million worth of ARR in what could be all-time record speed.
A good question to ask is why Perplexity is able to command such a massive revenue multiple premium to other standout companies in the current market. I think the answer is search. Here’s one way to do the math:
Google has around 87% of the US search market, and just over 90% on a global basis.
That was worth $54.0 billion worth of revenue in 2024.
Using some very crude math, each 1% of the global search market is worth $600 million in yearly revenue.
If you think that Perplexity has a shot at even a few points of global search market share in time, the company should scale to billions worth of revenue, making its potential $18 billion valuation seem slight. (Keep in mind that Perplexity expects healthy margins in the not-too-distant future.)
From that viewpoint, what matters more than Perplexity’s near-term monetization (ARR growth) is its user and query metrics.
January, 2024: “Since publicly launching the world’s first fully functional conversational answer engine a year ago, we’ve grown to 10 million monthly active users and have served over half a billion queries in 2023.”
October, 2024: “Aravind Srinivas, the CEO of Perplexity, says that the AI-powered search engine is now performing 100 million queries each week. Extrapolated out to a month, that’s roughly 400 million queries — up from 250 million queries in July.”
If internal data shows consistent user and query growth, then you can draw a line to multi-percentage points of search market share for Perplexity. That’s not just enough for an IPO — it’s enough for a huge company that could challenge one of the Mag7. An $18 billion valuation might wind up more albatross than wings for Perplexity’s future fundraising efforts, but if you are going to make money in venture you have to swing hard at the pitches that might yield four runs, right?
Ah, so that’s how OpenAI got so much CoreWeave capacity
I have to go prepare TWiST’s show docket, so we’ll have to be brief here at the close.
When CoreWeave filed its S-1, the world noticed that it was pretty heavily indexed on Microsoft. Later, reporting came out that Microsoft was backing out of some CoreWeave deals. CoreWeave disputed that. Then, OpenAI snapped up a huge chunk of CoreWeave’s future capacity.
I should have put the pieces together. Heres’s Semafor:
Microsoft chose not to exercise a nearly $12 billion option to buy more data-center capacity from CoreWeave […] CoreWeave, which is readying for the year’s most closely watched IPO, quickly found another buyer — OpenAI snapped up the contract last week.”
Duh.