And SpaceX IPO updates! Trillion dollar memory companies! Microsoft's upcoming BUILD event! #
- Welcome to *. Cautious Optimism, a newsletter on tech, business, and power. Modestly upbeat
**Tuesday. **In the past, a European startup exiting to an American tech giant for more than $1 billion would be big news. In today’s market, it’s a footnote. Why? Startups are staying private longer, allowing them to grow larger. The bar for what matters is therefore always rising. Good luck turning heads with an exit that is, what, a single AI seed round?
Today we’re looking at **Alphabet’s **new financial planning, trillion-dollar memory companies, **Microsoft’s post- OpenAI **plans, updates from **SpaceX’s **IPO, and questions for **Anthropic’s **IPO filing, now that it has filed a confidential petition to list. To work! — Alex
📈 Trending Up #
Climate-themed venture capital investment…racism in politics…technology export rules in China…anti-AI GOP sentiment? …successful cyber penetration? …code for copper, PVC…the money in the data center networking game…consumer spending? …executive turnover…chances of guillotines……EU membership
AI compute, via Berkshire: **Google **intends to sell $80 billion worth of stock in a multi-part transaction to raise fresh capital to finance AI spending. The plan includes a $10 billion private placement to **Berkshire Hathaway **(at a roughly 6% discount to Alphabet’s current share price), a $30 billion underwritten public offering, and a $40 billion at-the-market sales program (starting Q3 2026). The capital will be used for several purposes, earmarking $40 billion for “general corporate purposes, including capital expenditures to scale AI infrastructure and global compute,” $30 billion “to facilitate, for a period of time, an administrative change to how [Alphabet] meets tax obligations associated with vesting of employee equity awards,” and the last $10 billion left for “general corporate purposes.”
The company reminds investors in a document that it projected 2026 capex between $180 billion and $190 billion, with greater investment in 2027. The issue? Alphabet’s free cash flow was $174 billion in the preceding four quarters. That means the search and cloud giant can no longer afford to finance its capex from cash flow while continuing to pay its dividend and repurchase shares. So, it’s raising capital.
- Alphabet: “Google now has over 8.5 million developers building new experiences with its models monthly and its first party model APIs are processing 19 billion tokens per minute, a 6x increase year-over-year.”
That Berkshire is involved — the investment group is a long-time shareholder of Alphabet — is not a shock; when you need capital, you talk to the cash-rich, and that’s Berkshire today. Famously.
Why not use debt? Stratechery reckons that the equity sale first implies future debt fundraising. I think that’s right; Google’s free cash flow will not scale as quickly as its capex needs, so the company is going to make the rounds with a very large and deep hat. Debt is ~cheaper than selling shares, but I reckon that at today’s rates, selling a few shares when tech stocks are at, or near, record highs is also an efficient fundraising mechanism.
- Gist: The compute buildout is not slowing down. Alphabet is still incredibly bullish.
OpenAI’s Sam Altman spoke to ** CNBC **recently at the site of his company’s
nascent data center in Saline, Michigan. Why is he so confident in future compute demand? Here’s the man himself: “[P]eople are really saying, the future of my company, the future of my scientific research program, it is going to depend on this. Can you promise me compute long into the future?” Who wouldn’t want to say yes?
Trillion-dollar memory companies: Back in my day, a trillion-dollar company was an outlier. Today? There are three trillion-dollar *memory *companies. **Micron **recently crossed the $1 trillion threshold alongside SK hynix, while the more diversified **Samsung **is worth around $1.5 trillion.
What’s causing market hype for the memory-havers? Everyone wants their product, and there isn’t enough of it. Here’s a line from Micron’s (now somewhat dated) most recent earnings report:
Micron set new records across revenue, gross margin, EPS, and free cash flow in fiscal Q2, driven by a strong demand environment, tight industry supply, and our strong execution, and we expect significant records again in fiscal Q3
Even more, Micron said that it is forging “strategic customer agreements” that include “specific commitments over a multi-year time horizon for improved visibility and stability in our business model.” Incredible demand, and a break from the historical memory boom/bust cycle? Up go the valuations!
Microsoft’s AI aspirations: A few days ago, I poked around Microsoft’s first-party AI model family. Redmond’s recent MAI-Image-2.5 model (MAI = Microsoft AI) placed third on the Arena leaderboards, which was impressive. But what about MAI-1-Preview, the company’s first “end-to-end foundational model?” It’s not available.
Perplexing. Why was Microsoft not making the model more available? Here’s the answer, via The Verge’s Tom Warren: “Sources tell me that we’ll hear about new AI models in Windows, a new reasoning model from Microsoft AI, and a Copilot ‘super app'” at Microsoft’s upcoming Build developer event.
Ah, so Microsoft has been in the kitchen all this time. Can MAI-2-Preview, or whatever the new model is called, put real points on the board for Microsoft? Here’s hoping. Nothing would bring me more joy than *more *competition in the AI game; more compeititon is more good, as we like to say.
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html) Trending Down
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html)
Cybersecurity at…Metaslowing the compute rush…my faith in humanity…data portability…childhood’s light…comity in venture-land…the free press…your right to privacy…
SpaceX’s target IPO valuation: Good news, SpaceX’s upcoming IPO is getting discounted. All the way from a whisper-number of $2.0 trillion to $1.8 trillion. The 10% haircut may spur demand for shares in the combined space-launch, social media, cloud computing, and AI conglomerate.
And for investors curious about SpaceX’s growth prospects, the company updated its IPO filing with updated details on its compute deal with Anthropic. Here’s the old language:
[Anthropic] has agreed to pay us $1.25 billion per month through May 2029, with capacity ramping in May and June 2026 at a reduced fee. The agreements may be terminated by either party upon 90 days’ notice.
And the new language:
[Anthropic] has agreed to pay us $1.25 billion per month through May 2029, with capacity ramping in May and June 2026 at a reduced fee. After the initial three-month period, the agreements may be terminated by either party upon 90 days’ notice.
The wording of the S-1 filing now matches what CEO Elon Musk recently said on Twitter. So the deal is a three-month lease, with opt-outs from there on. How you want to weigh that against future revenue is up to you, but at least we now have clarity on the deal’s terms. This close to what could become the largest IPO of all time, details really do matter.
Five questions we can’t wait to answer with Anthropic’s IPO filing #
Anthropic announced yesterday that it had filed IPO documents with the SEC, albeit confidentially. Confidential IPO filings allow companies looking to list to get their docs in shape and gather feedback from regulators before sharing their results publicly and finally listing. Nothing odd here.
The fact that Anthropic is now ready to file indicates we *will *get an Anthropic IPO this year. A third-quarter timing wouldn’t shock me either way.
Sadly for you and me, a confidential IPO filing means that there are numbers out there, but we don’t have access to them. So today, we’re going to outline the things we most want to understand from Anthropic’s eventual public IPO documents:
How does Anthropic count revenue sourced from cloud partners? Anthropic and OpenAI generate lots of revenue by selling access to their AI models on partner clouds, like Microsoft’s Azure or Amazon’s AWS. OpenAIreportedly countsrevenue on a net basis (minus partner share), while Anthropic reports the same revenue on a gross basis (including partner share). As a result, Anthropic will report higher revenue and lower margins (assuming cloud partner revenue share is a cost of revenue rather than a marketing expense). How the company does math will matter. If alotof Anthropic revenue is really revenue share earmarked for cloud partners, investors could be dismayed by its (relative to expectations) real scale.What’s the gross margin profile of AI inference? At a very basic level, when Anthropic sells tokens on its own or via a cloud partner, how profitable is that marginal token? This data point will help us understand the true cost of AI inference today and whether or not the market can bear it.How much money did Anthropic lose offering underpriced subscription plans? That AI companies were subsidizing subscription plans is now common knowledge. But! Will we be able to see, in the filings, just how subsidized those plans were? Will we be able to see gross margin improvements after limits were tightened? Faster upgrades? Revenue growth? And what will the company tell us aboutwhyit made those shifts?How capital-efficient has Anthropic been in terms of compute capex? OpenAI spends more on first-party data centers than Anthropic, but the latter company doesn’t spend zero. So, how much? And how does the company account for its long-term compute deals? Those should be obligations, right? I presume that if we compare Anthropic’s RPO to its long-term leases, it’s gonna be upside down.Does Anthropic look good under normal accounting? All that is fun, but Anthropic is at once an AI labanda normal company. On the normie front, what does Anthropic look like sans nuance? If we just stare at its income statement, does it appear healthy? Yes, the market will be willing to grant the company a little room to maneuver thanks to its growth and prominence, but its IPO valuation will hinge (in part) on its basic, boring, GAAP results. (This is where WeWork hit the skids; it had fancy talk, but terrible normal metrics.)
We have other questions, of course. Which investors own the most, the company’s larger debt profile, its operating cash flow results, and their improvement (or not) over time. The list goes on and on: What does the company say about bringing Mythos to market? Its relationship with the DoD? I could go on all day, but the above questions — and their eventual answers — are enough to form a pretty strong view of Anthropic.
Now, we just need the damn filing.