Just Go Public
And: The Datacenter slump?
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
📈 Trending Up: Model Context Protocol … shade … yet more Chinese AI models … accidentally leaking opsec changes … RNG! … banking …
Stablecoins: On the back of news that Fidelity wants to build a stablecoin, Custodia Bank and Vantage Bank are already in the market with the Avit token.
📉 Trending Down: Open source … OpenAI’s legal battles … Oracle … free media …
Free trade in the United States is dead for the foreseeable future. POTUS’s latest 25% import tax (text here) on automobiles and their parts is a big damn deal. The new duties will take effect alongside other tariffs expected in early April. Europe is doodling up a response. The largest car exporting nations to the United States — Japan, Mexico, Canada, and South Korea — could take a notable whack. American consumers are also likely looking at higher prices for cars in the near-future. Tesla is set to win, thanks to a large domestic manufacturing base, and Bloomberg reports that Ford is also less exposed than other domestic auto giants.
As a modestly inflexible free trader, this is all a big bummer.
Free speech in the United States is dead for the foreseeable future. Watch this video of a Tufts graduate student get disappeared by ICE agents out of uniform on the street. She has not been charged with any crimes, per current reporting. Instead, the F-1 visa holder was yanked off the street and shipped across state lines apparently because she wrote an op-ed in her university paper that criticized Israel’s military actions and her university’s leadership. Regardless of where we stand personally on Israel’s conduct in its war with Hamas, losing legal status to stay in the United States over thoughtcrime is an abdication of the principles that have made America the haven for immigrants and dissent it has long been.
As a rather inflexible free speecher, I am a virulent mixture of distraught, angry and disappointed.
Big Rounds of Note
Mercury raises $300M at a $3.5 billion valuation: Fintech Summer cometh? It sure seems like it. Stablecoins are on fire, Robinhood is firing on all cylinders (see above), crypto is hot again, eToro is going public, and now Mercury, a neobank aimed at upstart businesses is raising nine-figures at a huge valuation jump from its last private round. The Information reported that Mercury was at or around $500 million worth of ARR, giving it about a 7x multiple. That’s cheap in Seed-round-AI terms, but far in excess of the basement-scale price/sales multiples that Block and PayPal exhibit today.
Island raises $250M at a $5 billion valuation: This is a fun one. While the world swoons over upstart AI companies, there’s still juice in the enterprise software game. Island, which makes a secure browser for corporate customers, is on a solid fundraising journey and writes that since its 2022 exit from stealth, “the company has won 450 customers, with annual recurring revenue more than doubling each year following.” That’s super solid. I don’t know off top if I would rather have shares in Mercury at $3.5 billion today or Island at $5.0 billion, but both companies should prove IPO candidates in time.
Speaking of which:
Just Fucking Go Public
Stripe’s finances are about as lovely as you thought. The Information reports that the payment giant saw revenues of $5.1 billion last year, up 28% from the year prior, while doubling “its free cash flow last year to about $2.2 billion.”
Now, FCF is not GAAP net income, but any company that kicks off 43% of its revenue in free cash is a super brilliant company. Period. And Stripe, now worth $91.5 billion, is able to shit cash while sitting atop a private-market multiple of about 18x. Not bad.
But the Adyen comparison loometh. Here’s a rough rundown of the numbers (inspired by Sheel, of course):
Stripe’s 2024: TPV $1.4 trillion (+38% YoY), revenue $5.1 billlion (+28% YoY), FCF $2.2 billion (~+100% YoY), $91.5 billion valuation
Adyen’s 2024: TPB €1.29B ($1.39T at current forex rates, +33% YoY), revenue of €2.0 billion ($2.15 billion at current forex rates, +23% YoY), FCF €859.8 million ($927.6 million at current forex rates, +34% YoY).
Stripe will cost you $91.5 billion, while Adyen is currently valued at about $49 billion after converting from Euros. Both companies are sicker than your local LAG villain hitting a two outer on the river. But, I do wonder if we’re seeing pricing pressure accumulate on each.
Why? Both companies are printing TPV growth at a faster clip than revenue growth, implying either larger customers coming in at lower rates, or falling overall prices for payment processing due to competition.
Regardless, Adyen is doing great as a public shop. Stripe should follow suit and drop the endless parade of secondaries and annual letters and just get on with it.
As should Databricks and Anthropic. The latest had me howling:
Anthropic and Databricks struck a five year, $100 million pact to sell artificial intelligence tools to businesses, targeting those seeking to build their own AI agents.
For reference, Databricks told the world that it expected to cross “$3 billion revenue run-rate and be free cash flow positive in the fourth quarter ending January 31, 2025” back in December. Meanwhile, Anthropic reportedly reached about $80 million worth of monthly revenue in late 2024, and $2 billion to $4 billion worth of top line this year.
Nothing to see here. Just two private-market companies worth enough to instantly land on the S&P500 doing nine-figure deals with one another. You know, normal startup shit.
The Datacenter slump?
The latest mini-trend is data center overcapacity. As OpenAI builds out its own mega-clusters, xAI wants to double its chip base, and all the major tech companies are shelling out tens and hundreds of billions of dollars on new data center buildout, is there simply too much investment racing towards the end of the same rainbow?
The latest is Bloomberg reporting that Microsoft “walked away from new data center projects in the US and Europe that would have amounted to a capacity of about 2 gigawatts of electricity.” That note comes after Alibaba’s chairman said that we could have a bubble forming in AI-tuned compute.
Recall that CO is not worried about hyperscalers having a little bit of spare capacity:
I think [overcapacity] is a reasonable concern if you are a company building out mass-AI compute on credit without having a stable source of internal demand. To make that more concrete, Microsoft spending $80 billion on its own datacenter buildout this fiscal year bothers me not at all. Why? The company is cash rich, enormously cash generative, has a cloud business outside of AI workloads, and has its own internal AI compute needs. If Redmond winds up with a little extra capacity, it’ll grow into the racks soon enough.
The above concern is not idle; it could impact the worth that CoreWeave manages later today when it prices its IPO. Provided that that can doesn’t percussively come together with a boot, of course. But with Microsoft hedging its bets a little, I do wonder if improvements to AI efficiency will partially limit chip demand. Not so much as to change the larger arc of technology investment, but enough to take some of the air out of the current, heady moment.
Microsoft is not alone in worry about spare capacity. Here’s the MIT Technology Review discussing China:
Just months ago, a boom in data center construction was at its height, fueled by both government and private investors. However, many newly built facilities are now sitting empty. According to people on the ground who spoke to MIT Technology Review—including contractors, an executive at a GPU server company, and project managers—most of the companies running these data centers are struggling to stay afloat. The local Chinese outlets Jiazi Guangnian and 36Kr report that up to 80% of China’s newly built computing resources remain unused.
That’s lethal, but also perhaps a statistic to not get too worried about — yet. While overcapacity is miserable for owners of data centers in the near term, from a different perspective China has massive, underutilized AI compute capacity at the same time that its domestic AI companies are releasing new models like they are going out of style, and the nation is working to bake homegrown AI into every product possible. I want to know what the idle number is in six months. If it’s down to 60%, I’ll lose no sleep. If it’s still 80% — or higher — ring the alarm.