Welcome to Cautious Optimism, a newsletter on tech, business, and power.
A domestic commercial passenger jet (flight 5342) collided with a military helicopter (a Blackhawk, on a training flight) this morning in the DC area. The collision resulted in a massive explosion (not linked here). Dozens died. The FAA doesn’t have a leader at present after its past administrator Mike Whitaker announced he would resign on January 20th. (DOGE’s Musk has previously said that Whitaker should resign.) Welcome to the meeting point of administration changeover and tragedy.
📈 Trending Up: OpenAI’s checking account … satellite cellphone calls in Europe, the United States … packaged hot air … thin skin … DeepSeek-inspired nationalism … Chinese economic jitters … US GDP … not EU GDP … Swedish support for Ukraine …
Article of the Day: Has Europe’s great hope for AI missed its moment?
Mistral, one of a mere handful of AI foundation model companies in Europe, is collecting doubts regarding its long-term viability. Why? Its revenue is only measured in the tens of millions instead of the billions, and its American rival has zillions of dollars to play with.
Yes, but: Earlier this week, the fact that China’s DeepSeek had managed to accomplish much with less than what US-based AI providers have, was considered a signal that Mistral and friends had a shot.
Someone is wrong; someone is right.
📉 Trending Down: Separation of powers … housing affordability? … competence … spine at Meta … China-India tensions? … freedom of the press and religion in India …
Today in WTAF: Why do they hate trans people so much? Also, is there no penalty for breaking the law?
Here, have another model
After DeepSeek made waves, Alibaba claimed to have bested one of its models. Now an American company — Ai2 — what TechCrunch describes as a “nonprofit AI research institute based in Seattle” has a new model out that it claims has “a consistent edge over DeepSeek V3.” Viva la open source domestic AI?
Earnings: Microsoft, Meta, Tesla edition
Microsoft
The headline: Microsoft Q2 cloud revenue falls short of expectations, sending stock lower
The quarterly results: Revenue of $69.6 billion (+12% YoY), net income of $24.1 billion (+10% YoY), “AI business” now on a $13 billion run rate (+175% YoY), Microsoft Cloud revenue of $40.9 billion (+21% of YoY), Azure growth of 31%.
What about AI and capex? Microsoft spent a lot of time during its earnings call on the subject (transcript here). There’s too much to quote, but here’s a few excerpts worth your time (emphasis added):
Satya Nadella, CEO: “AI scaling laws are continuing to compound across both pre-training and inference-time compute. We ourselves have been seeing significant efficiency gains in both training and inference for years now. On inference, we have typically seen more than 2X price-performance gain for every hardware generation, and more than 10X for every model generation due to software optimizations. And, as AI becomes more efficient and accessible, we will see exponentially more demand.”
Amy Hood, CFO: “We expect quarterly [capex] spend in Q3 and Q4 to remain at similar levels as our Q2 spend. In FY26, we expect to continue investing against strong demand signals including customer contracted backlog we need to deliver against across the entirety of our Microsoft Cloud. However, the growth rate will be lower than FY25 and the mix of spend will begin to shift back to short-lived assets which are more correlated to revenue growth.”
Takeaways: Microsoft’s AI business is real, growing, and represents a huge channel of top-line expansion for the software giant. Investing into hardware to meet AI demand — capex of $22.6 billion in its MRQ — is reasonable based on the demand signals it’s seeing. The small difference between street estimates and Microsoft’s own guidance are small enough to not matter from a long-term perspective.
One more thing: Microsoft is impressed by DeepSeek, and now supports its R1 model. The company also made very public hay over its partnership with OpenAI, saying that it was “thrilled OpenAI has made new large Azure commitments” and that “OpenAI has a lot more coming soon.”
Meta
The headline: Meta earnings top expectations as company forecasts higher costs, AI investments in year ahead
The quarterly results: Revenue of $48.39 billion (+21% YoY), costs +5% YoY, ad impressions +6% YoY, daily active people +5% YoY, net income of $20.8 billion (+49%).
What about AI and capex? $14.84 billion in Q4, and “full year 2025 capital expenditures [in] the range of $60-65 billion.”
Analysts wanted to know about the ROI of open-source AI investment. Meta argued that similar to its work on compute, building open-source AI models will help with standardization, and acceleration of the state of the art.
Meta said that “Llama 4 is making great progress” while “Llama 4 Mini is done with pretraining, and [Meta’s] reasoning models and larger model are looking good, too. The goal, per Meta CEO Mark Zuckerberg, with Llama 3 was to “make open source competitive with closed models,” while its aim with Llama 4 is to “lead.”
Notably the company appears undaunted regarding its capex spend when comparing Zuck’s plans that were shared pre-DeepSeek. In other words, akin to Microsoft, it’s full-steam ahead on the existing AI strategy.
Takeaways: Meta’s limited cost growth and more quickly-growing revenue helped make the social media giant even more profitable. So much so that it appears Meta can keep losing billions per quarter on its VR and AR strategy and investing in AI. Not a bad place to be.
Tesla
The headline: Tesla stock rises after company pledges return to growth after Q4 results disappoint
The results: Revenues of $25.7 billion (+2% YoY), automotive revenue of $19.8 billion (-8% YoY), GAAP gross margin of 16.3% (-138 bips YoY), net income of $2.3 billion (-71% YoY), free cash flow of $2.0 billion (-2% YoY)
What about AI and capex?
Musk spoke at length about self-driving, AI, robotics, and the future of Tesla during his company’s earnings call. Tesla’s investor presentation noted that the company “[i]ncreased AI training compute by over 400% in 2024,” and Musk added during his post-earnings chat that the electric vehicle giant was more than doubling down on autonomy.
So much so that Musk has all but primed the market to anticipate quick expansion of “unsupervised [full self-driving]” this year, saying that he is “confident that we'll release unsupervised FSD California this year as well” as in Austin, and that “the only thing holding us back is an excess of caution.”
Takeaways: Tesla’s growth as a car company has halted, but investors seem pretty content that self-driving and the company’s humanoid robot will become big businesses (Musk: “And my prediction, long term, is that Optimus will be overwhelmingly the value of the company.”) Whether you agree with the market on Tesla’s short-term trajectory or not, at least the company isn’t thinking small.
What’s the aggregate vibe? Some folks felt that DeepSeek’s AI progress undercut the need or financial efficacy of AI investment. Major tech companies seem unbothered. The CO view is that conservatism during platform shifts is a recipe for obsolescence.
Expect a lot of this
When the DOGE crew decided to rip through government spending, I started waiting for it to find things that looked silly without context, decide to can them, and then take a premature victory lap.
Enter ‘DEI scholarships for Burma.’ You can’t imagine a more DOGE-friendly target. Helping people with an eye towards diversity in a different country? Quelle horreur.
The effort, apparently called the Lincoln Scholarship Program, was put together to help “develop a more stable, pluralistic and prosperous Myanmar [Burma] by equipping young professionals and academics with technical and leadership skills.”
Tom Malinowski, a former congressperson and Assistant Secretary of State was a bit confused, asking:
What could be "woke" or "DEI" about these scholarships? The only clue is that USAID says they're for students "of diverse backgrounds" (oh no, diversity!). This is essential for Burma, where the military has exploited ethnic and religious divisions to stay in power.
He went on to note that Secretary of State Marco Rubio argued in favor of strong engagement with Myanmar to “preserve Burma’s fragile democratic transition after last week’s military coup against the democratically-elected, civilian-led government of Burma, also known as Myanmar” back in 2021.
You could argue that we don’t have the money as a nation for such work. That’s fair, even if I disagree. But if you are advocating for cutting taxes while arguing that the United States’ work abroad to help build a stable, prosperous world for us to transact with is too expensive, get bent.
Hyperscalers gonna hyperscale until the event horizon, as more usage demands more stable architectures. Your DOGE ending, along with the former Congressman’s commentary deserve their own piece and highlight. If that program is about cutting fat/waste, it deserves a critical companion to all cuts and that doesn’t take much anymore with LLMs and soliciting the program creators/managers for commentary or to defend their work. That’s a desperately needed conversation about the lens American neo-austerity is using to say “we did great work here” with.