Tech had a weird weekend

Meta announced plans to sell excess AI compute to third parties, surprising investors and causing shares of neocloud providers CoreWeave and Nebius to drop 19% and 23% respectively. The move comes amid reports that CEO Mark Zuckerberg was underwhelmed with AI agent progress, though the head of Meta's AI team denied any performance issues and promised improvements to the Muse Spark model.

Tech had a weird weekend Tech spent the long weekend arguing on X, as it has for time immemorial. If you were offline being a human, let me catch you up Welcome to . Cautious Optimism https://www.cautiousoptimism.news/ , a newsletter on tech, business, and power. Modestly upbeat Monday. Welcome back. Today, we’re talking Meta selling compute, the distrust of AI labs, GDP and the Vice President, and the latest from the tech-right. To work — Alex 📈 Trending Up OPEC production https://apnews.com/article/opec-increase-oil-production-iran-hormuz-bae40a1146cea569ddfdfc39d4867441 … geopolitics through the lens of soccer https://www.cnbc.com/2026/07/05/trump-fifa-balogun-world-cup-red-card-suspension.html … ElevenLabs, now past $500M ARR https://x.com/mati/status/2073065496270766461 … self-driving https://www.reuters.com/business/autos-transportation/tesla-rolls-out-robotaxi-service-miami-2026-07-03/ … layoffs at Microsoft https://www.theverge.com/news/961528/microsoft-layoffs-july-2026-sales-xbox … Anthropic buying yet more compute https://investors.terawulf.com/news-events/press-releases/detail/142/terawulf-announces-anthropic-lease-at-justified-data-campus-and-sale-of-majority-interest-in-abernathy-joint-venture-to-fluidstack … open source AI https://x.com/eliebakouch/status/2074011171661701466 … bubble concerns https://www.notus.org/economy/treasury-internal-report-warning-dangers-ai-bubble 📉 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 American doctorates https://www.nytimes.com/2026/07/06/us/research-universities-fewer-phds-science.html … USA-Italy relations https://truthsocial.com/@realDonaldTrump/116869343915433662 … analyst vibes regarding Datadog https://www.investing.com/news/analyst-ratings/bernstein-downgrades-datadog-stock-rating-on-q3-earnings-caution-93CH-4775657 … not using the hot dog man meme https://www.bloomberg.com/news/newsletters/2026-07-05/netflix-viewers-are-abandoning-shows-after-one-season … cybersecurity https://www.bleepingcomputer.com/news/security/jadepuffer-ransomware-used-ai-agent-to-automate-entire-attack/ From the tech water cooler Tech spent the long weekend arguing on X, as it has for time immemorial. If you were offline being a human, let me catch you up: Meta selling compute: Conquest or capitulation? News that Meta intends to sell some of its compute https://www.bloomberg.com/news/articles/2026-07-01/meta-is-building-a-cloud-business-to-sell-excess-ai-compute?taid=6a450cb068410d0001628b09&utm campaign=trueanthem&utm content=business&utm medium=social&utm source=twitter&embedded-checkout=true to third-parties took the tech market by surprise. Why would Meta, which was securing third-party compute deals as recently as two weeks ago https://x.com/alex/status/2067740225464386028 , decide to rent out some of its recently-purchased GPUs? Investors in the neoclouds that Meta had contracted with sold off their shares late last week, and CoreWeave and Nebius shed 19% and 23%, respectively. Reporting that Mark Zuckerberg was underwhelmed with agentic progress https://www.reuters.com/business/zuckerberg-says-ai-agent-development-going-slower-than-expected-2026-07-02/ led some to presume that Meta’s recently-formed AI team was not performing as well as expected, leading the company to rent out some capacity it had previously reserved for internal use. That’s hogwash, per the head of the new AI team, Alexandr Wang https://x.com/alexandr wang/status/2072848108342677597 , who promised that an update to the still-unreleased Muse Spark model would bring “big improvements in coding and agentic capabilities to be more competitive with other leading models”. Soon. What’s going on? SpaceX is what’s going on. The space-datacenter-AI hybrid lavished on building data centers to train and serve frontier models, only to fail to crack the AI mainstream and end up with capacity to spare. By monetizing those idle GPUs, SpaceX added tens of billions of dollars https://www.cautiousoptimism.news/spacexs-ipo-is-a-bet-that-anthropic-stays-compute-constrained/ in quick possibly temporary revenue to its income statement, which should allow it to continue scaling capex across its various arms. Meta, which has been criticized https://www.reuters.com/business/meta-lifts-capital-expenditure-forecast-doubling-down-ai-push-2026-04-29/ by some investors for spending like a hyperscaler without a hyperscaler’s business model, likely wanted a slice of the same investor goodwill that SpaceX has been enjoying. - Morgan Stanley views Meta’s move https://x.com/zen tropy/status/2072502976367718563 as a way to monetize some first-party compute to drive near-term profitability. That’s reasonable. SpaceXAI’s contracts have 90-day opt-outs built in, so it can profit until it wants to revoke capacity. - So, Wang can continue to train new models while Meta rents out spare capacity. And later, if the upcoming Muse models are aces, the company may pull those contracted FLOPs back home as needed. While CO is generally bullish on AI, we hold that excessive cleverness usually fails https://x.com/edzitron/status/2073969652498067584 , and the industry is at least one foot in that domain when it comes to compute deals these days. Hyperscalers and neoclouds’ Q2 results will help us better understand the state of play. AI labs: Will data concerns scare off enterprise revenue? Where will the AI industry see value accruing? Companies serving the global data center buildout are doing so well, the market has rotated https://www.reuters.com/business/ai-investors-may-pivot-hyperscalers-chipmakers-morgan-stanley-says-2026-07-06/ away from hyperscalers to some degree while boosting their suppliers. Down the stack, in other words. Where does that leave AI labs? At, or near the top of the stack. That’s worrying some folks. The concerns run as follows: - Open-source AI will undercut frontier labs’ pricing power. This is not a new concern, but it’s grown prominent recently thanks to a steady drumbeat of commentary from tech leaders https://x.com/arthurmensch/status/2073157738276749354 that major AI labs will see what their customers are doing with their models, and replicate that work to capture their markets. - Palantir’s Alex Karp made that argument recently https://www.youtube.com/watch?v=0A3sGymV6kY . He says speaking from atop his own book his customer base doesn’t want to pay through the nose for access to Fable and GPT-5.6 loosely in exchange for little value while handing over their IP. - These are not idle concerns, at least in practice. Alibaba is working to ban staff from using Anthropic’s Claude models over “ security fears https://www.morningstar.com/news/dow-jones/202607032148/alibaba-set-to-ban-staff-from-using-claude-over-security-fears ,” for example. Naturally, the folks shouting the loudest are the very people who want to sell you the tools to train open models for your company’s peculiar needs. But there’s something to the fear that using frontier models is a risk. Mercor’s Brendan Foody here’s my interview https://podcasts.apple.com/us/podcast/twist-500-interviews-with-cortical-labs-turing-and/id315114957?i=1000720278246 argued recently https://x.com/BrendanFoody/status/2073912646592778513 that “authorizing a foundation model to train on your company’s data is like giving a competitor access to your Slack workspace and GitHub repository.” Here’s some more commentary https://x.com/quxiaoyin/status/2073540332353208325 , for flavor. OpenAI and Anthropic are not yet public, so it’s hard to tell how they are faring. Perhaps the labs will leak new revenue numbers to assuage such concerns, but the pendulum has swung from “ leading AI models have inifinite worth ” to “ AI models are not worth much .” We’ve been here before. Back in early 2024 https://www.cautiousoptimism.news/why-that-ai-model-you-love-is-the-fastest-depreciating-asset-in-human-history/ , Atreides Management’s Gavin Baker pointed out that without unique data, no AI model would retain value. And if large companies pull their data back from the frontier labs… - Here’s an idea: Maybe the frontier labs become pure-play cybersecurity companies GDP: A good economic metric? Excerpts from VPOTUS JD Vance’s recent book https://www.harpercollins.com/products/communion-j-d-vance bounced around X over the weekend, including his notes on GDP https://x.com/PhilWMagness/status/2073728969992454641 , which he decries. Now, it’s worth recognizing that GDP is a flawed statistic, in that it has known limitations, and thus should be part of a balanced economic data diet. But declaring economics “fake” as a result of those shortcomings is a bit much. The GDP truther-ism matters. The Veep said in a recent interview https://x.com/robbysoave/status/2072751925305466908 that he wants to move from Milton Friedman backwards: “Hamiltonian tradition is gonna be what we see in the American right and will dominate American conservative economic thinking for the future which is not laissez-faire, it’s actually much more about building the kind of tools, building the kind of infrastructure that allow human beings to flourish … and if you turn economic development into a sort of idol, then you end up sacrificing a lot of the things that matter most.” - Cross that view with his jokes in the same interview about how POTUS received little pushback for the proposal of partially nationalizing AI labs, and you have a recipe for a future GOP that is protectionist in the worst possible ways. - Such a GOP will also be regulation-forward, it’s worth noting. Investor Chamath Palihapitiya recently held a fundraiser for Vance, so perhaps he’ll hold his nose and allow the high priests of GDP investors? run the show? The tech-right and its anti-democratic views: The public march of the technology to the right in the run-up to the last presidential election was a regular subject on CO a few examples https://www.cautiousoptimism.news/what-silicon-valley-thinks-it-will-get-from-a-second-trump-term/ of the genre https://www.cautiousoptimism.news/techs-political-moment-my-birthday-and-the-internet-turns-against-ai-crawling/ in case https://www.cautiousoptimism.news/i-for-one-am-opposed-to-monarchy/ you want https://www.cautiousoptimism.news/i-thought-we-liked-free-markets/ to go back https://www.cautiousoptimism.news/i-mean-its-one-banana-republic-michael-what-could-it-cost-14-billion/ in time and read along https://www.cautiousoptimism.news/nativism-is-going-to-wreck-the-tech-economy/ when it was happening. Today, with most tech leaders willing to bend the knee to POTUS, it’s merely the water we all swim in. But there are indications that some tech leaders are pushing the envelope even further . Back in August 2025, Elon Musk argued https://x.com/elonmusk/status/2022919799752294493 that “common culture” is critical to national survival, and that “nobody dies to defend a ‘“’multicultural economic zone’ ” I’ve had a draft entitled, “I will die for the multicultural economic zone” that I have yet to finish in the CMS ever since. As you can surmise, I disagree with Musk’s argument that diversity is bad, and the broader American experiment of limited government, free markets, and free speech are insufficient. Musk recently endorsed a film criticized for endorsing extra-judicial violence against immigrants https://www.theguardian.com/film/2026/jun/30/elon-musk-anti-migrant-armie-hammer-film-free-download-citizen-vigilante . And he came out against free speech https://x.com/JeremiahDJohns/status/2073803856484413759 during the July 4th holiday https://x.com/AriCohn/status/2073801257412665776 , only to add one more banger https://x.com/elonmusk/status/2073312715985309698 to his remit: Here’s the context: Many wealthy Americans which Musk is, as he immigrated here despite reportedly breaking the rules to do so https://www.theguardian.com/technology/2024/oct/26/elon-musk-illegal-immigration are angry with the current mayor of New York City. They despise him. The maker-taker discourse is currently popular in tech circles here’s Friedberg https://x.com/theallinpod/status/2068761044139364832 , as it neatly sorts society into people that the tech-right want to support their friends and the people they want to deport those who disagree with them . Musk stating out loud that he thinks universal franchise is a mistake is terrifying, as it implies that the powerful want to decide who gets to vote; i.e., who gets to rule. I’ve long believed that the the founder-first/founder-only viewpoint that many VCs hold today would yield political thinking that leans toward fascism. The viewpoint is that only founders matter, which is a micro-form of the great man theory. If founders are the solution to capitalism, why not also politics? So I think it’s not rare today amongst tech oligarchs to nod along with the Yarvin viewpoint https://x.com/alex/status/2073873577481265570 that democracy is unworkable in its outcomes, and thus a different system should be tried. A CEO, perhaps? Does anyone know of one? Vance’s backer Peter Thiel made a similar argument back in 2009 https://www.cato-unbound.org/2009/04/13/peter-thiel/education-libertarian/ , writing he no longer believed “that freedom and democracy are compatible.” Thiel’s argument is that people can’t be trusted to vote, as they vote for things that are bad: The 1920s were the last decade in American history during which one could be genuinely optimistic about politics. Since 1920, the vast increase in welfare beneficiaries and the extension of the franchise to women — two constituencies that are notoriously tough for libertarians — have rendered the notion of “capitalist democracy” into an oxymoron. Not that Thiel is the first person to make this argument. Murray Rothbard argued back in 1996 https://mises.org/journal-libertarian-studies/origins-welfare-state-america that women were partially to blame for the welfare state. Note how much of today’s tech-right agitation against women is simply recycled thought: Of all the Yankee activists in behalf of statist “reform,” perhaps the most formidable force was the legion of Yankee women, in particular those of middle- or upper-class background, and especially spinsters whose busybody inclinations were not fettered by the responsibilities of home and hearth. It’s simply terrifying to see the world’s richest man agitate for a more limited franchise while shouting that “common culture” is a national requirement, and try to stoke anti-immigrant hatred around the world. I am not alone https://x.com/Noahpinion/status/2073539317054755150 in this fear. I think the upcoming midterm election is going to prove a turning point for tech and politics. If the GOP gets its ass handed to it, as many expect incumbent parties often struggle https://www.npr.org/2010/11/03/131046118/obama-humbled-by-election-shellacking during midterm elections , the democracy-doubters will have fresh reason to doubt their entire process, given that their friends lost power. I doubt they will take it well.