Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Happy Tuesday, friends! Holden is on a flight today, so I am flying solo. Please anticipate, and forgive, an errant comma or two.
It’s going to be a busy day. Not only are the markets working hard to recover some of their losses posted yesterday, but we’re finally starting the technology inning of earnings season. That means Tesla, SAP, Iridium today, IBM and ServiceNow tomorrow, and Alphabet the day after.
Tesla, always a fun company to watch quarter to quarter, has a bit more pressure on it shoulders than normal. We’ll tear its numbers apart tomorrow morning. Until then, to work! — Alex
📈 Trending Up: 3M, after earnings … chatbots … my head, spinning … the Keeling Curve … Circle … history as a flat circle … gaming … European trade centrality? …
The government’s teeth? Google is still facing a possible Chrome divestment despite Trump in the White House ….
📉 Trending Down: Good financial choices … freight … hoodwinking consumers … China-Vatican relations … China-US relations …
Domestic Indian ecommerce companies? The USG is considering strong-arming India into allowing greater US-corporate penetration of its ecomm market
📰 Headline of the Day: “Dow Headed for Worst April Since 1932 as Investors Send ‘No Confidence’ Signal”
Remember the college kid who invented a way to cheat on BigTech technical interviews, and then showed the world how he did it? Columbia crumpled like used tissue to Amazon’s ire, but Roy Lee raised more than $5 million and turned his idea into a real company. Jason and I interviewed him during the collegiate scandal, which you can watch here. I figured his rulebreaking ways and technical chops crossed with a proven ability to get attention would lead to something like this!
The ‘Sell America’ Trade
Yesterday, as stocks dove yet again I made a note to write up a segment discussing just how far the worth of technology companies have fallen since Trump took office. In essence, the argument was going to be Trump’s tech backers have some ‘splaining to do.
But that would be too narrow a point to make. I think the tech selloff, while material, is merely component to the ‘Sell America’ trade. You can see global capital rotating out of the United States in a number of ways:
Equity outflows from the United States crossed with inflows to European and Asian shares
The dollar in retreat as the euro, Canadian dollar, and Japanese yen rally
All of which add up to a “‘no confidence’ trade” in the United States
You don’t have to search hard to find tweets predicting the end of the USD as the global reserve currency. And with trade deals slow to come to fruition — thanks in part to confusing, conflicting demands from a disorganized White House — there’s little respite on the horizon apart from the chance that POTUS decides to throw in the white flag yet again. (‘Reciprocal’ tariffs were merely paused, recall, meaning that we could actually see things get worse before they get better.)
So, what’s the optimistic take? That power centers apart from the Executive Branch will make enough noise to force to White House to capitulate, and soon. Apple got help; Nvidia did not. Tariffs on China went up even more as other tariffs were put on hold; there is a threshold of pain — grievance? — at which point even POTUS may listen to the market. So, here’s hoping that we’ve already paid in full on the pain bit, and are merely waiting for someone to convince Trump over golf that, yes, trade really is good and his multi-decade view that trade is zero-sum is wrong.
After selling off 3.5% yesterday, cloud stocks are down 30.5% from their 52 week highs. Wee.
Why X + xAI might actually work out
News that Elon Musk may be considering yet another fundraise for his combined social media and AI foundation model company was puzzling. Surely $12 billion raised in less than a year was enough capital to get to at least the midway point in 2025?
Whatever we think about AI model capex, investors may yet put more capital into the Musk AI project. The logic — per Antonio Gracias — is simple enough:
AI models that lack proprietary data will become commodified, leading them to rapidly lose value
AI models that benefit from proprietary data will not become commodified, leading to them holding or gaining value
Everything else held equal, it’s a fair argument. So long as the data in question is not utter shit, it should help.
But is social data from X enough to make xAI into a real contender? While its Grok models have reached the upper-ends of AI model leaderboards at times, I can’t recall hearing from a startup that actually uses — prefers? — Grok via its API in their product or service over something from OpenAI or Anthropic. (I am still looking for someone who uses Grok so that I can learn why; if that’s you, hit reply!)
Perhaps in time xAI will learn to lever X (Twitter) data more effectively, and Grok will improve to best its better-known rivals. We’ll see.
But in the meantime, there’s a good reason to think that xAI has a good shot at at least mass-consumer adoption. Monetization and profitability at xAI will still be hard to come by, but something brilliant that Grok has made mainstream is chatting to AI inside of your timeline.
From a real-time search of X data, I found several tweets tagging @grok inside of the last few seconds when I ran a query:
@grok confirm this news? How many tariff deals we have so far?
@grok what percent of the United States citizens earn over 60 k a year?
Folks love having a fact-checker inside their social! Is it ironic that the people who hated trad fact checking are now the purveyors of synthetic fact checking. Artificial fact checking? Sure, but regardless of what you want to call it, Perplexity is also getting in on the @-tagging fun over on X. If X remains the room where it happens, then commanding local mindshare matters. And Grok is doing a good job there.
Fun digression: The company that commands the most-used AI chatbot on social platforms will, for a great many people, occupy the seat of the Arbiter of Truth. Therefore, AI models are our future oracles. And you don’t own one!
If you can make money offering up endless free AI inference to non-paying users isn’t clear, but xAI is now worse there than OpenAI, apart from the latter having converted tens of millions of people to its paid service. I doubt that X’s paid services are quite so large.
AI’s command interface is either going to be voice, video, chatbots, or social media responses. Or all of the above. We sure do live in interesting times, and that means we’re learning a lot very quickly. Hard to be too sad about that.