Crypto Rises, Agents Drop, Factories Hum
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
After some high drama, crypto bills are advancing in Congress. Over in the House, the Clarity Act (crypto regulatory clarity, including more permissive standards for decentralized projects) passed, the GENIUS Act (stablecoin rules that previously passed the Senate), and the Anti-CBDC Surveillance Act (ban on the United States having a Central Bank Digital Currency, or state-run stablecoin) all passed.
Partially in response to recent regulatory wins, the value of major digital assets have risen. Now that you can add a little crypto via managed ETFs to your extant investing accounts, I presume that you have a little exposure as a hedge. Enjoy your gains responsibly!
Before we get to work, this is a bad idea. Let’s go! — Alex
📈 Trending Up: The Chinese consumer … Chevron-Hess … EU sanctions on Russia … Anthropic’s commercial efforts … canal politics … exit bans … cash for babies … media cowardice … the surveillance state …
📉 Trending Down: Working at Amazon … inflation in Japan … national industrial planning …
Remember yesterday when we discussed adding PE to 401ks, the risks and potential benefits of the idea? Splitscreen that with the exit of University of California capital from the PE game, after discovering that it was paying a lot of fees to hedge fund folks for less-than-market performance.
Building stuff is big business
During the SaaS era of venture capital, hardware was out of vogue. The simple reason for the focus by founders and their backers on software monetized by recurring subscription payments was that it was asset-light, high gross-margin, and understood. You could boil startup risk down to a few Excel entries, and plow money into GTM machines that you had seen before. Then wait and collect carry later on (in theory).
Yes, it was boring, but it also built a lot of big companies.
Things have changed. Now, startups building hardware (and the software required to run it) are among the hottest companies in the market, period. SpaceX has long been a coveted equity to hold, but Anduril is also viewed as a ticket to profit.
Don’t take my word for it. Secondary marketplace EquityZen reported that in the second quarter, SpaceX and Anduril were two of the top three most popular companies on its platform, while “defense tech companies Mach Industries, Helsing and Anduril [saw] indications of interest surged, over 80% from May to June” on the platform.
Expect to see more of the same. The drone industry is enjoying a renaissance thanks to Russia’s invasion of Ukraine, which is driving a rethink of how war is fought in the modern era. As large ships are falling prey to new weapons that might render them all but obsolete, so too are traditional land-based weapons systems facing new foes that could shake up what battlefield dominance looks like.
Enter startups like Vatn Systems (underwater drones), Saronic (floating drone ships), Skydio (dual-use drones), Perseus Defense (counter-drone weapons), and the like.
But if the United States and Europe each want to foster new technology companies that can meet their defense needs, they probably need to ensure that the supply chain that feeds those new concerns is also secure. It’s no good to build a drone industry here in the States with an eye on countering China if all the parts that go into those drones is sourced from the same nation we’re bent on containing.
It’s not an idle concern. I interviewed the CEO of Skydio earlier this year for TWiST, who told Jason and I that his company had set up its manufacturing efforts in the United States, but that moving away from a supply network that included China was tough. Before his company got hit by sanction by the government of China, it worked to de-risk from the nation, and managed to get all the way there apart from a single component.
Put simply, if we’re going to have secure drone tech here in the States for personal, commercial, and defense uses, we’re going to need to build more stuff here at home; outsourcing our industrial base to a geopolitical rival probably isn’t tenable with Taiwan still on the table — and supporting democracies is not a place where we should bend.
So, what to do? Build. That’s why the recent Hadrian funding round matters. The startup just closed around a quarter billion dollars to scale up its manufacturing footprint here in the United States, including new facilities in Arizona to complement its factories in California.
Hadrian uses robotics to squeeze efficiency out of the high-precision parts game. Its model appears to be working well enough for its backers to push another nine-figures into its coffers, and good. We want more high-end manufacturing here in the States, just as we do not want to reorient our domestic economy to low-end manufacturing. Value-add? Good. No value-add? Bad.
I grew up in a household with lathes, milling machines, and a cutting torch. I have made more metal shavings than I want to admit to. My CNC experience is all but nil, but having done metalwork by hand let me say that automation in this area sounds like a godsend.
And I do not say that merely to foment more F1 boondoggles from industry.
The United States is not alone in building new defense tech champions. Helsing over in Europe is working along similar lines. And the work to support them with domestic part supply is perhaps accelerating. Yesterday, some of the folks behind the ‘California Forever’ new city project announced the ‘Solano Foundry,’ a huge industrial park that will be part of the larger city effort, which is based in Solano County.
Here’s California Forever’s Jan Sramek:
Silicon Valley made its name in hardware. HP, Intel, Lockheed, and other pioneers didn’t split their R&D and factory floors. Engineers and tradesmen worked side by side. In 1965, Lockheed alone employed 28,000 people in Sunnyvale. 28,000 people!
But we offshored to China, and broke that model. And now it’s Shenzhen and Guangzhou that lead drones and robotics. It’s time to invest back home. But R&D teams can’t waste days traveling from a design lab in SV to factories scattered everywhere. We need the American Shenzhen.
We also can’t win by building factories off of random freeway exits in the middle of nowhere. The best people don’t want to work there. To actually re-industrialize, we need industrial ecosystems, where R&D, production, and training are located within amazing places to live.
Enter the Solano Foundry, which wants to lever 2,100 acers and proximity to talent (Silicon Valley), the army (Travis Air Force Base), and other industrial bones to get America building more critical stuff. The variety of industries that the Foundry wants to attract is diverse:
I know that I do not share politics with many of the current defensetech barons. But if we’re going to bolster democracy around the world, we’re going to need the drones to back up our views. Let’s build.
OpenAI pushes ahead with agents
Its either the year of agents, the start of the decade of agents, or yet another trip ‘round the sun during which AI promises more value than it can deliver. If OpenAI has anything to say about the matter, however, it’s more 1 and 2 than three, and perhaps even more 1 than 2.
The company shook up the AI industry yesterday by releasing ChatGPT Agent. It’s a blend of ChatGPT (ask AI stuff), Operator (AI that can use your computer to do stuff), and Deep Research (AI researching stuff) into a single product. The combo is threaded together with a “unified agentic system” that “carries out [your prompted] tasks using its own virtual computer, fluidly shifting between reasoning and action to handle complex workflows from start to finish, all based on your instructions.”
OpenAI is releasing the service to its paid plans, so if you are on the $20 per month tier, you should get access in time. My account has yet to grant me a key, but the folks over at Every are — perhaps predictably — enthused by the release. That means 20 million folks or so are going to be able to use ChatGPT agent starting just about now.
Because the new service can interact with spreadsheets and the like, it could have real-world impacts especially on white-collar work. And challenge Microsoft’s own efforts to imbue its Office suite with AI, but let’s save that for later.
Early reviews indicate that the new service has more chops than preceding agentic offerings, but also that it can be slow today and is still imperfect. I’m withholding full judgement until I get hands-on time, but the short gap from the release of Grok 4 to the Kimi-K2 drop to now ChatGPT Agent from OpenAI is indication of how quickly AI is still advancing. Encouraging!