And: New AI models, submarines, and how many companies can rent Elon's GPUs at once? #
Monday. Stocks were down on Friday after a better-than-expected labor print (more on this from NPR), as a result of which tech stocks, including semiconductor stocks, took heavy body blows. Basically, it’s now normal to worry that a strong labor market would lead to interest rate increases. Crypto didn’t want to let stocks have all the fun, though, and so ended the week with a blisteringly negative trend. Companies focused on facilitating crypto activity also fell sharply — Coinbase, for one, lost 7% on Friday.
But it’s a new week and so we can look forward to new beginnings! Stocks are trending higher today, crypto has recovered a little, and early indicators suggest we won’t see the sell-off extending into the new week.
Speaking of which, this morning in the newsletter we’re looking at new AI models, shipbuilding, xAI’s megadeal with Google, and Bending Spoons’ IPO filing. To work! — Alex
📈 Trending Up #
The North Korean economy? …Marvell, joining the S&P 500…state ownership of the (AI) means of (model) production…tensions in the Middle East…Trump crashing out…South KoreantechnologycompaniesheartingNvidia……inflation in IranCorning shares, after an Amazon deal…Alphabet-Intel chip ties
New AI models: Folks are expecting OpenAI to drop GPT-5.6 shortly, and we keep seeing hints that **Anthropic’s **Mythos model is inching toward release.
But I think we should care more about the fruits of OpenAI’s new product direction. The Financial Times reports that OpenAI wants to grow its business revenue to 50% (from 40% currently) of its total top line this year, and is plotting a combined service — a *super *app — in hopes of encouraging the billion-plus ChatGPT users to cough up some actual money.
- Codex, my current AI daily driver, is brilliant. But it’s hardly something my ChatGPT-using parents will be able to figure out. So I am skeptical that OpenAI has managed to solve the problem of helping people spin up software as easily as they use it. Still, I’m also excited to see what they have cooked up.
Sticking to the new model theme, over in the United Kingdom, there’s a £1.1-billion push to build more computing power, and British AI lab **Cosine **is tapping the state and a host of local industrial players to build a coalition for training a new frontier AI model for the island nation.
Sovereign AI is no small effort around the world, though the results have been middling for the most part. France’s Mistral and Germany’s Black Forest Labs are doing well, but the U.K. could definitely use its own horse. With **Lovable **and **Legora **taking flight from European shores as well, AI homegrown on the continent is a real phenomenon that we must pay attention to.
New submarines: While U.S. naval drone companies are the collective apple of my eye (shout-out to Saronic, Blue Water Autonomy, and my darling Vatn Systems based here in Rhode Island), there’s reason to fret that our nation’s inability to build ships has become a national security issue.
Sure, China’s shipbuilding capabilities are legendary (an example of the genre), but it wasn’t until recently that American *submarine *construction fell behind.
A new “sailless” Chinese submarine recently caused a ruckus amongst boat-watchers who’re now busy trying to sort out what the new vessel is for. Perhaps seabed waiting? Silent, quick attacks? Who knows!?
- While I appreciate the American submarine industry trying to juice employment by sponsoring the Connecticut Sun, we need to do more to keep pace. I worry that aside from Vatn’s operations beneath the waves, no startup is ready (at present) to replace our lagging submarine construction industry. Perhaps the American Dominance startup cohort literally needs to think bigger.
[📉](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 cattle…American literacy…sexting in the United Kingdom…Japanese GDP…EU GDP……chances of a rate cut this yearminimum SpaceX share buys
xAI, the AI company: When **SpaceX **announced a major partnership with coding startup Cursor (including an option to buy the smaller company for $60 billion later this year), the larger company brought compute to the table. And when it announced a major deal with
Anthropic, Elon Musk’s conglomerate again brought compute to the table.
Now, most recently, this is what SpaceX proposed in its deal with Google:
On June 5, 2026, we entered into a Cloud Service Agreement with Google LLC (“Google”) with respect to access to compute capacity. The compute capacity provided includes approximately 110,000 NVIDIA GPUs, CPUs, memory, and other related components. Pursuant to the agreement, the customer has agreed to pay us $920 million per month from October 2026 through June 2029, with capacity ramping up through September at a reduced fee. If we fail to deliver access to the committed amount of GPUs by September 30, 2026, then following a one-month grace period, Google may immediately terminate the agreement or accept the number of GPUs provided, with a corresponding pro rata reduction in the monthly fees. After December 31, 2026, the agreement may be terminated by either party upon 90 days’ notice. The customer will retain ownership of, and intellectual property rights in, its content, Al models, and related data.
Thus, SpaceXAI has practically overnight morphed into the cloud buddy of a leading AI lab (Anthropic), a hyperscaler-cum-foundation-model company (Google), and a leader in the agentic coding space (Cursor). That probably won’t leave much spare capacity for SpaceX’s own AI efforts, right?
It’s ironic, but merging **xAI **with SpaceX will likely save the space company’s IPO. **Starlink **is profitable and growing, but it isn’t large or profitable enough to support a valuation north of $1 trillion. The company’s space launch business is also valuable, but it’s too small right now to justify the valuation Musk wants for his company. The ghost of xAI as a neocloud, however, is worth a mint.
A couple of weeks ago, we noted in our first read of the SpaceX IPO filing that the listing was predicated on Anthropic staying short of compute, and therefore willing to shell out $1.25 billion per month for access to SpaceX’s spare capacity.
SpaceX is today more diversified, with both Google and Anthropic ponying up for its unused GPUs. It is therefore a more stable company with a stronger customer revenue mix, and larger, thanks to more compute deals. It’s odd that GPU rentals are the space launch company’s chariot to the public markets, but it’s the AI era after all.
Bending Spoons is going public #
From the headlines, you might think that the business of buying up software companies and turning them around is dead. The Financial Times reported this week that “software buyout deals have collapsed to the lowest levels since the Covid-19 pandemic,” due to market worries that AI will consume traditional software companies in a single gulp.
Subscribe to Cautious Optimism to unlock the rest.
Become a paying subscriber of Cautious Optimism to get access to this post and other subscriber-only content.