Welcome to Cautious Optimism, a newsletter on tech, business and power.
Thursday! Trade relations between the United States and China thawed a little, meaning that the rare earth crisis and possibly even-higher tarrif rates are behind us. For now. The TikTok question is not yet fully settled, but folks in tech are watching what happens with beans and chips between the two nations.
China stopped buying US soybeans this year, which has greatly harmed a key POTUS electoral constituency. At the same time, the United States has continued to limit China’s access to high-end AI chips. Trump said before talks with Xi that Nvidia’s new Blackwell chip would be part of the conversation. That made tech folks critical of China in geopolitical terms — this publication is part of that group — worried that the United States was going to trade access to cutting edge chips for bean purchases.
Happily, POTUS walked back the Blackwell point, but we could still see a relaxation of chip export rules in exchange for greater agricultural purchases. Watch this space. Now, to work! — Alex
Grammarly-Coda-Superhuman
Earlier this year, Grammarly (AI-powered copyediting) teamed up with Coda (team productivity fabric). Shared investors and related visions brought the two companies together. But I didn’t quite get the deal (my interview with Grammarly’s then newly-appointed CEO Shashir Mehrotra here). A little while later, Grammarly snapped up Superhuman, the well-known email service beloved by tech folks. Again, cool, but I didn’t quite get the pairing.
Yesterday, Grammarly announced that it’s changing its name to Superhuman while also launching an agentic assistant. Jason and I had Mehrotra back on the show to chat about the news, which helped fit the pieces together:
Grammarly did a bunch of work to ensure that its grammar-checking tool worked across computing surfaces; this was required so that it could edit your writing, wherever it occurred.
Coda had a strong agentic toolset called Coda Packs, meaning that as a pair, Coda+Grammarly could run AI agents everywhere Grammarly had already done the world to lay down digital rails.
Finally, Superhuman mattered as a buy, per Mehrotra, because it’s the “surface you actually work in.” (That logic went for Coda, too.)
Add it all up, and you have Grammarly’s distribution rails, Coda’s agentic prowess, and ownership of two core workspaces (docs, email) to boot. Mix in an ur-agent (Superhuman Go) and you have a lightweight AI OS built atop existing operating systems (Windows, MacOS), and even second-layer OS-similars, like Chrome. Very neat, and perhaps a good example of why we should see more startup combinations in the coming quarters.
Ok, we’re still compute-constrained
Yesterday, Microsoft, Alphabet, and Meta reported earnings, telling the market how much they spent on data centers, how much more they intend to spend, and how the current frenetic pace of compute buildout is performing against market demand.
In brief, here’s how earnings went down at the full-company level:
Microsoft: Beat revenue and profit expectations. Calendar Q3 2025 (Q1 fiscal 2026) revenue of $77.67 billion (ahead of an $75.33 billion billion), and EPS in the quarter of $3.72 per share (ahead of an expected $3.67 per share). Shares of Microsoft fell in response.
Alphabet: Beat revenue and profit expectations. Q3 revenue of $102.3 billion (ahead of an expected $99.9 billion), and Q3 EPS of $2.87 (ahead of an expected $2.26 per share). Shares of Alphabet rose sharply in response.
Meta: Beat revenue and profit expectations. Q3 revenue of $51.24 billion (ahead of an expected $49.41 billion), and Q3 EPS of $7.25 (adjusted, ahead of an expected $6.69 per share). Shares of Meta fell sharply in response.
Confused about the share price movements from Microsoft and Meta in light of their earnings beats? Market consensus is that Meta and Microsoft’s plans for greater capital expenditures relating to compute buildouts scared investors. Perhaps.
Earlier this week, CO flagged the risk of compute overshoot; the risk that the hyperscalers were spending too much, and were therefore burning good cash chasing theoretical revenue. To sort out market jitters over capex, let’s hear from the companies themselves, starting with Microsoft.