And: Long live the drone boats. #
Wednesday. I’m helping host an investor panel on ** TWiST** this afternoon, so today’s missive will be slightly shorter than usual.
This morning, we’re talking drone boats, AI cost control in the context of historical cloud expenses, AI joy and market selection, what to look for in Oracle‘s earnings today, and two-tier AI. To work! — Alex
📈 Trending Up #
SaaS hiring…the Iran War…or not? …public corruption(really!) …Siri, incredibly…the value of power generation(really!) …European defense startups? …AI compute buildout…
Inflation: May saw prices increase yet again, with the U.S. government reporting that the consumer price index (CPI) rose 0.5% last month, bringing inflation to 4.2% (up from 3.8% in April) compared to last year. On a core basis (excluding food and energy), CPI rose 0.2% in the month, bringing the adjusted inflation rate for consumers to 2.9% (up from 2.8% in April).
- Headline numbers were in line with expectations, though core CPI gains were slightly lower than the market predicted. Still, inflation is rising, and that’s bad news for tech stocks hoping for a rate cut. - Stocks are down.
Drone bots: We bemoaned the state of American shipbuilding the other day, especially now that Chinese shipyards are churning out warships far faster than they used to, while the United States falls behind on submarine construction.
Still, we gave a shout-out to a few drone-boat companies (both above- and underwater) that are working to ameliorate our national deficit: Saronic, Blue Water Autonomy, and Vatn Systems.
It seems our compliments were well-timed. The government just said that the pilots rescued in the Middle East after their helicopter was shot down were saved by an unmanned surface vessel (USV) made by Saronic. Nice.
- I was not aware that Saronic’s boats were already in active service; consider me caught up. Anduril is also in the drone game, naturally, and has abunch of underwater drones.
The fact that drone boats are now operating in several theaters of war is an indication of where the future of warfare is heading. Both ground- and water-based combat is increasingly unmanned and autonomous today, and governments are spending to stay on top. And it’s not just the United States and China. The European drone game is strong; they have a war on their doorstep, after all.
Startups working on AI cost control: When AWS was new, price cuts were de rigeur. **Microsoft **had to compete on price when bringing Azure out of preview. But over time, the costs of using cloud storage and compute rose; suddenly, this brilliant idea of using pooled, shared compute from a centralized host started looking more expensive than cheap. Entire businesses
were builtto help AWS and Azure customers save money. The AI era is trending along similar lines, but within a fraction of the time. Service providers are following the money. CloudZero, previously focused on cloud optimization, now pitches itself as “the AI ROI company.” It still does cloud stuff, but clearly, the largest pain point for enterprise companies today is their AI spending.
Tools are getting better, too. Larridin, which bills itself as “the measurement and optimization layer for enterprise AI,” just launched a product that tracks AI expenditure, providing an “attribution layer that connects [AI] spending to the teams, agents, workflows, and decisions that drive it.” I don’t mean to endorse Larridin (though Homebrew backed it, shout-out
Hunter). I merely want to point out that we’re seeing manystartups tackle this problem with increasingly sophisticated tools.
- The cloud got expensive; spend got smarter; the cloud has never been bigger.
- AI is getting expensive; spend will get smarter; AI will continue to grow.
[📉](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)
The yen…Indian GDP growth…deflation in China…Nancy Mace…China-Taiwan relations…TFR in India…EU-Apple relations…
AI joys: As a daily Notion user, I have watched the service expand its remit during the AI era. Mostly, I use Notion to handle a dense collection of show and interview notes, schedules, assets, and other operational details that make the podcast tick. I’m but a small part of the overall team, but it lives and runs on Notion, so I am now a power user of the service.
Enter complaints on X about Notion’s inclusion of AI. Is the company losing its focus? But what I found magical is that, in the comments, a former investor and a founder had utterly different takes:
Is Notion making a mistake by not wasting time in adopting and integrating AI? I can tell you that its transcription tools are a godsend. Does that make me a Notion AI WAU? Maybe.
Startups that had popular products before AI are still trying to figure out how to meet the moment. Should you redesign your entire company, à la Intercom‘s metamorphosis into Fin? Should you add AI to your existing products without changing
howthey work?
No matter what you choose, some users will be upset. A slice of your user base (existing and potential) will want more AI; others will hate you for including it at all. But if the future is moving in one direction, which way do you lean?
Would you trade a little gross retention today for, potentially, much more net-dollar retention tomorrow?
Market sentiment: Cloud giant Oracle reports quarterly results later today. The market expects it to post a solid improvement in revenue (~20%), but we should be paying attention to the numbers inside that headline figure.
Investors anticipate Oracle Cloud Infrastructure to grow revenue by around 90%, up from 84% in the previous quarter. How much backlog Oracle’s compute business reports will matter, too; if the company’s long-term AI compute revenue commitments look weak, the signal could wash the market red. Tech companies are currently valued quite richly, remember.
- More on Oracle’s results tomorrow, but if you are around after the bell, it’s an earnings call worth tuning into.
Two-tier AI is gonna be weird #
Yesterday, **Anthropic **dropped a version of its Mythos model called Fable 5 that comes lobotimized in the cybersecurity and biology niches, and a very high price tag. Fable 5 costs twice what Opus 4.8 lists for.
And, the market is eating it up. Despite all the clamor about stratospheric AI costs, Claude Fable 5 is seeing soaring demand over on OpenRouter
(though the USG might be driving much of the recorded usage).
The issue is not that Fable 5 is expensive — no one *has *to use it, after all — but that it codifies a two-tier AI market. People in tech are worried about this new age. Before, AI labs made new models and rushed them to market, and their value was derived according to their model’s performance. Companies even rushed security work to meet incentives.
No longer. After keeping Mythos under wraps and only allowing a small group of governments and companies to access it early to identify and patch cybersecurity vulnerabilities, Anthropic chose not to make the product generally available.
Mythos is still restricted, and even the market-friendly Fable model will fall back to Opus 4.8 if you try to use it for something it considers spicy. Anthropic is one of the two best companies in the world for building AI models. It’s also the most willing to apply ethics to its release schedule rather than follow market norms.
The result is a cautious AI lab has added a velvet rope to the AI club. You must be *this *important to access the VIP Fable Room, and most people (companies included) are not.
Why is it dangerous for AI to have inner and outer circles of access? Incumbency protection. If Anthropic opens up Mythos only to companies with the most potential bugs, then smaller companies without sufficient clout to secure an invite will be less secure and relegated to second-rate AI.
I’m not saying Fable 5 is not a capable model. Its benchmarks are insane, and it’s rated by Artificial Analysis as the world’s smartest AI. Still, no one today wants second-rate AI if their competitors are getting something better.
As AI gets better, security concerns will force more labs to follow in Anthropic’s footsteps. Would the CCP allow DeepSeek to release a Mythos equivalent before using it internally? Probably not, if ever, I suppose.
It appears the era of open AI releases is behind us. Here on out, us plebs will be fighting over the silver medal. Alas.