More data center spending and an IPO!
Welcome to Cautious Optimism, a newsletter on tech, business and power.
Tuesday! We’re shaking off a weird start to the week, and digging back into our bread and butter: The volume of money flying around the technology world today. That means earnings, startups, more data center spending and an IPO. Let’s get busy! — Alex
📈 Trending Up: eToro, after earnings … companies using AI to help companies find places to use AI … Vision Fund earnings … Oura’s revenue prospects … IP rights … the GOP healthcare plan? … our collective advertising load …
📉 Trending Down: Gemini, after earnings … our national political process … paying your critics … Intel’s AI team …
Startup of the Day: Neros, which is helping Ukraine fend off Russia and build American drone supply lines … investor enthusiasm for AI investment …
CoreWeave’s earnings: On the heels of hyperscaler earnings reports — mostly positive, recall — neocloud CoreWeave reported its own third-quarter results yesterday after the bell. Unsurprisingly, the company said that it operates “in a highly supply-constrained environment where the demand for CoreWeave best-in-class AI cloud platform far exceeds available capacity,” leading to $1.4 billion worth of revenue in the quarter (+134% YoY), and revenue backlog growth of more than $25 billion in the quarter.
The neocloud reported improving revenue diversification — “the number of customers that exceeded $100 million of revenue over the last 12 months tripled year-over-year” — at the cost of rising debt costs. Shares of CoreWeave are off 9% after the company’s Q4 guidance came in light, partially driven by delays at a data center buildout that would have come online in the current quarter.
The most interesting data point that CoreWeave shared went like this:
Demand for AI cloud technology remains robust across generations of GPUs. For example, in Q3, we saw our first 10,000-plus H100 contract approaching expiration. Two quarters in advance, the customer proactively re-contracted for the infrastructure at a price within 5% of the original agreement. This is a powerful indicator of customer satisfaction as well as the long-term utility and differentiated value of the GPUs run on CoreWeave’s platform.
Why do we care about recontracted H100 supply? Because those chips are now a generation dated; that they are contracting for effectively a flat price may imply that the revenue-generating life of prior-generation chips is longer (and therefore more lucrative) than expected. Hyperscaler and neocloud economics could be better than expected, in other words. Not bad!
A $35 billion South Korean data center will be both designed and run by AI
Meta bought another $3 billion worth of compute, this time with Nebius, which reported Q3 earnings this morning, reporting that it was “sold out of available capacity” in the quarter.
Fighting back: Clio, a legaltech software company investing in AI, raised a $500 million Series G with another $350 million worth of debt to help finance its purchase of vLex, a rival company that offers — you guessed it — AI-powered legaltech software.
With AI shaking up the legal technology space quickly, seeing incumbent software companies race to adopt and deploy AI solutions makes sense; you don’t want to lose an entire march to your smaller rivals.
Expect more startups working to build software for lawyers and law firms to see more consolidation in 2026.
Two houses, both alike in dignity: It’s trendy to scorn OpenAI’s business prospects these days. Folks are worried about its compute costs, its revenue growth, its gross margins; the list goes on and on. Anthropic, on the other hand? There’s much less hand-wringing.
That’s because the companies really are quite different. So when the WSJ reports that Anthropic “expects to break even for the first time in 2028” per secured documents, while OpenAI “forecasts its operating losses that year to swell to about $74 billion,” you think that the latter company is a massive dog, especially compared to its rival.
Maybe, maybe not. OpenAI expects to burn $9 billion on $13 billion worth of revenue this year (perhaps closing 2025 on a $20 billion run rate), while Anthropic expects to burn $3 billion against $4.2 billion in sales (perhaps closing 2025 on a $9 billion run rate). Their paths to those numbers diverge:
OpenAI wants to control its own compute future, raising and spending and contracting for north of a trillion dollars in silicon grunt. It also builds expensive video and image-generating tools and serves a massive (often non-paying) consumer user base. It’s aiming to be the global household AI name, while also taking a good-sized chunk of the enterprise AI market.
Anthropic is content to — largely — use investor compute from Google Cloud and AWS while serving a mostly business customer base, and avoiding the GPU strain that comes with Sora 2, for example. It wants as much of the enterprise AI market that it can secure, and a far smaller piece of the consumer game.
Place your bets. It’s fun to compare the companies head-to-head, but the two leading American frontier AI labs are very different beasts, and we’d do well to keep that in mind.
Klook’s going public!
Founded in 2014, Hong Kong-based Klook started life as an experience and travel booking service focused on the Asian market. It later expanded to a more global purview, helping it raise $200 million in a Series D back in 2018, and $200 million more in 2021. By 2023, when Klook raised a $210 million round led by Bessemer, it reported reaching profitability as it climbed out of a COVID-19-impacted travel market.
Now, Klook is going public here in the United States. SoftBank, which took part in its 2021 round, is a major shareholder.
Klook’s business is simple: It generates “the substantial majority” of its revenues from “sales of experiences offerings” that its technology facilitates. Gross transaction value (GTV) is therefore a key metric for the company. Let’s examine its GTV growth, and the rate at which it can tear gross profit (net revenue minus COGS) from that transaction value: