Welcome to the $4 trillion club
No, not you. I'm talking about Redmond.
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Thursday! Pour some coffee, there’s a lot to talk about. So much so that we’re going to gloss over Ford’s roughly $800 million profit hit in the second quarter from tariffs, and its anticipated $2.0 billion “tariff-related headwind” for the full year.
No. In order, we’re talking AI revenue, Ambiq’s stellar IPO result, Figma’s IPO pricing, and earnings from Microsoft and Meta. If I manage to leave room, a closing note on media economics. Cool? Let’s go. — Alex
📈 Trending Up: Support for Taiwan … in Congress at least … Robinhood, after reporting glorious results … state control of the economy … AI bullishness … tariff rates, starting tomorrow … government meddling in media … inflation …
Trending flat: Interest rates, after the Fed held steady …
📉 Trending Down: Work to limit White House corruption … H20 availability in China … FDI … US-India relations …
One of my favorite games to watch these days is the ‘One Up AI Tussle.’ One company, say Moonshot AI, releases a great new model (the Kimi K-02 family). Next, another company rocks up days or weeks later (Alibaba) with new, possibly even better tech (the latest Qwen models).
The same game is played with AI revenue results. After it was reported by the crack team over at Bloomberg that Anthropic’s valuation might reach $170 billion — and the $150 billion we anticipated— thanks to its its revenue reaching $5 billion ARR after achieving ARR of $4 billion earlier this month, I expected the news would lead to OpenAI getting a new revenue number out into the market to counter Anthropic’s press cycle. And? $12 billion, it turns out.
Ambiq soars
Low-power AI chip company Ambiq — our notes on its finances here — soared from its mid-range IPO price of $24 per share. After opening at $38.43, shares of Ambiq closed at $38.53 for a first-day gain of 60.5%. Its shares have drifted another 5.7% higher in pre-market trading.
Recall that Ambiq added 600,000 shares to its IPO, meaning that it raised more capital than expected in its debut. The company’s pivot away from selling in China is now well-capitalized.
Figma to debut near the price Adobe wanted to pay
When Figma initially indicated a $25 to $28 per-share IPO price range, we said it was light and would likely be raised. Figma then raised its range to $30 to $32 per share. After that, CO said that we expected “Figma to price a dollar or two above range perhaps getting close to the psychologically-important $20 billion mark.”
Yesterday evening Figma priced at $33 per share, giving it a valuation of “close to $20 billion, fully diluted,” per Renaissance Capital. Well done, Figma.
After Ambiq priced more conservatively and traded with great enthusiasm, there’s a good chance we’re about to see Figma go bonkers when it starts trading later today. Here’s hoping. Not only because it will be fun to watch folks whine about how IPOs are priced, but because it could give other unicorns a spine.
Microsoft and Meta wow the street
Shares of Microsoft are up 8% in pre-market trading, pushing the value of the Redmond-based software giant to $4 trillion, making Nvidia now merely the first of two companies to breach the valuation threshold.
Microsoft reported a top-and-bottom beat in the second calendar quarter — the fourth of its fiscal 2025 — with revenue of $76.4 billion, ahead of an anticipated $73.8 billion. CNBC reports that Microsoft’s earnings per share of $3.65 bested expectations of $3.37 worth of per-share profit.
But more impressive than Microsoft’s 18% revenue growth, 23% operating income growth, and 24% net income expansion it reported were its cloud numbers. Revenue from Microsoft’s ‘Intelligent Cloud’ arm grew 26%, far in advance of total revenue growth, while “Azure and other cloud services [turned in] revenue growth of 39%.”
Azure grew 34% in the year-ago quarter, and 33% in the sequentially-preceding period. Acceleration!
Microsoft expects Azure to grow 37% in the current quarter. And the company expects to spend $30 billion on capital expenditures in the first quarter, despite reporting strong improvements in compute efficiency for AI inference. Why? Demand. Microsoft CFO Amy Hood told investors that the $30 billion outlay is predicated on “$368 billion of contracted backlog” at the larger Microsoft Cloud group.
AI played a big role, too. Hood said that Microsoft’s “investments, particularly in short-lived assets like servers, GPUs, CPUs, network and storage, is just really correlated to the backlog we see and the curve of demand.” Hood also said that she had expected to be in better supply-demand shape by June. Now? Hopefully December. That’s AI-bullish.
Microsoft also reported that Azure now generates $75 billion worth of “annual revenue,” that its Foundry Agent Service now has 14,000 customers, Copilot apps now have more than 100 million MAUs, and GitHub Copilot now has 20 million users.
Shares of Meta are up just under 12% in pre-market trading after the social giant reported its own Q2 earnings. Why? It crushed expectations, reporting revenue of $47.5 billion and earnings per share of $7.14, ahead of expectations of revenues of $44.83 billion and earnings per share of $5.89.
Meta’s 22% revenue growth, 38% operating income growth, and 36% net income growth were stellar, especially given $4.5 billion worth of operating loss generated by Meta’s Reality Labs business.
So, what has investors so excited? Not merely that Meta’s Q3 revenue guide of $47.5 to $50.5 billion is solid against an expectation of $48.85 billion, but because the company’s AI strategy is attractive to its backers. The core details are as follows:
Meta is spending heavily on AI, both in talent (headcount) terms, and how much it invests in hardware. The company’s $66 billion to $72 billion worth of 2025 capex — up $30 billion year-on-year — will be surpassed next year. Or as Meta CFO Susan Li put it, her company “currently expect[s] another year of similarly significant capex dollar growth in 2026.”
To what end? Better AI models. Mark Zuckerberg said that Meta is making “good progress towards Llama 4.1 and 4.2 -- and in parallel, [is] also working on [its] next generation of models that will push the frontier in the next year or so.”
What’s that good for? Zuck outlined five areas: “[I]mproved advertising, more engaging experiences, business messaging, Meta AI, and AI devices.”
Notably, Meta said that it currently has no plans to rent out its data center capacity for third-party inference needs. Investors believe its massive investments into AI capability and capacity will both help Meta’s core business (the world of social apps and supportive advertising) and build the future.
Before earnings, Meta’s CEO dropped a letter indicating that his company is starting to see signs of AI being able to improve itself. He reiterated the point during earnings:
Over the last few months we have begun to see glimpses of our AI systems improving themselves. The improvement is slow for now, but undeniable. Developing superintelligence -- which we define as AI that surpasses human intelligence in every way -- we think is now in sight.
Those results were good for a few hundred billion worth of market cap. It’s good to be a tech giant these days!
What’s that article worth?
While Cloudflare’s Pay Per Crawl technology remains in private beta, and startups like TollBit and Human Native work to build data marketplaces to rent information to AI companies, we’re getting a better view into what the largest one-off transactions between media companies and AI shops are worth.
The previously announced deal between the New York Times and Amazon is worth $20 million to $25 million per year, according to new reporting from the Wall Street Journal. In the first quarter, the Times generated revenues of $635.9 million. While the first quarter of its year is historically the paper’s weakest in dollar terms, the figure was up 7% compared to year-ago results.
A quarter of $25 million is $6.25 million. So, for its smallest quarter, Amazon licensing revenue was worth about 1% of top line. The full-year number expressed in percentage terms will be lower.
As traffic from search continues to fall, deals like the Amazon-Grey Lady agreement will become all the more important to drive media revenue growth.
Even at 1%, isn’t that a bit of a letdown? That’s all the Times’ material is worth? No. That’s what it’s worth to one company, likely licensed more for training than for inference. Mix in similar deals in the future with other AI companies, and you can see the Times generated 5% of its total revenue from such deals.
But the real picture is actually rosier than even that. Licensing deals for training purposes are one way of two that major content holders can lever their data for AI lucre. Charging other firms for inference pings — scraping by AI companies to answer a particular user query — could generate a lot more.
That’s why Cloudflare’s Pay Per Crawl could wind up being accretive to what the data marketplace startups are building, instead of suffocating:
Licensing deals: Bloc revenue generated by reaching a particular agreement between data and AI companies.
Scraping income: Revenue generated by content holders charging inference scrapes on a per-pull basis.
I expect both to matter to the Times, in time. Still, Amazon paying the Times tens of millions of dollars per year today doesn’t make me sad. The opposite, in fact