- Welcome to *. Cautious Optimism, a newsletter on tech, business, and power. Modestly upbeat
Thursday. Checking yesterday’s analytics, it appears that the Holiday Week is truly upon us! That in mind, we’re going to move with great focus today. (If you are reading this from outside the United States, we’re heading into the July 4th celebration, which is a big deal domestically. Lots of people take time off.) * CO *is off tomorrow.
On the docket? The latest payroll data, OpenAI’s idea of giving the government shares in AI labs (quick reaction), the FED push, and a few thoughts on the Nvidia-neolab deal that has people worried. To work! — Alex
📈 Trending Up #
American industry supporting European rearmament…SpaceX handhelds? …rugging your voters…building your own chips, Anthropic edition… EV sales atTesla,Rivian…neocloud fundraising…
Global venture capital investment: Crunchbase News has the final tallies, but would you believe that massive venture investment into the most valuable private companies in history is pushing numbers up?
Bribing the government: Worried that the government isn’t treating you right? Perhaps you should cut them in on the action! Worried that the public views the AI race as a gold rush that they are not invited to? Perhaps you should cut them in on the action!
That appears to be the logic behind OpenAI’s nascent pitch to give the government 5% of its shares, provided that other AI labs buy into the scheme. Per the FT’s reporting, the collected equity would be placed in a sovereign wealth fund of sorts. The dollars in question are staggering. For the sake of large, round numbers, presume that OpenAI and Anthropic are worth $1 trillion apiece, while Amazon, Alphabet, Microsoft, and Meta are worth a combined $11.4 trillion. So call it $12 trillion for the sake of modesty. A 5% stake in that total would be worth $600 billion, a sum large enough to actually put a bit of weight on the asset side of the national indebtedness ratio (speaking loosely).
This ain’t gonna happen. Not only are investors in AI labs not likely interested in diluting themselves to gift equity to the same entity that taxes them, but if we want all the foundation model labs of note to join hands in the project, you can’t let the hyperscalers off the hook; Meta, Amazon, Microsoft, and Alphabet each build foundation models, to varying degrees of success. Not only are they worth more, but allowing them to skate past the wealth fund would provide an unfair advantage to incumbents! And I doubt they are willing to undo so much share repurchasing just because Sam wants them to. (Now Uncle Sam, on the other hand?)
Why is the idea of donating AI equity to the public bubbling now? I think it’s because the AI labs want to sell internationally, and giving their regulators’ bosses financial upside on allowing them to serve non-domestic customers is a potentially good way to ensure their success. Enter Sam Altman, Mr. 5%, in the FT yesterday:
I would like to propose a simple framework: a US-led international forum that establishes accepted standards, provides expert and impartial analysis of capabilities and risks, and makes the technology available to nations and companies that participate and follow the rules. This forum might include government representatives, independent technical experts and others. It could also serve as a governance mechanism over the labs, and guard against the commercial pressures that can lead to unsafe racing. […]
But if global safety standards are not established, AI restrictions, possibly via actions taken by individual countries, will be the only way forward.
The United States leading a union of democratic nations building AI for non-authroitarian purposes sounds pretty good to me; get the EU involved, and a large portion of global GDP could operate under a shared AI ruleset that would ensure barriers between participating nation’s don’t get too high. And that’s good for OpenAI and Anthropic, who are currently fending off not only low-cost, open-weight models from China but also the sovereign AI push amongst certain wealthy nations to create their own domestic models. (And limitations from their own government.)
- If state capitalism is the required antidote to the current AI regulation crisis, shame on us.
[📉](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)
Free-ish trade…social media access for teens…EU climate focus in the AI era(context) …Starlink’s market stranglehold? …clean government…working at Lucid…
Job growth: After a preview via ADP, the US economy added a mere 57,000 jobs in June, under an expected +115,000 result (129,000 jobs were added in May, revised down from +172,000). Unemployment, however, ticked lower to 4.2% from 4.3% as the labor force participation rate fell by 0.3% in the month.
- Stocks are up! Why, given the soft numbers? Fears that a stable labor market and hot inflation would lead to interest rate hikes cooled after the labor market deteriorated.
Good news! **Not becoming an FDE: **Following major AI lab + PE deals (OpenAI here, Anthropic here), other companies are finding the idea of building teams to deploy engineers at customer companies to help them roll out AI irresistible. To wit:
- “ AWS invests $1 billion to embed AI forward deployed engineers with customers“ - “ Microsoft Frontier Company: AI engineering that amplifies and protects your intelligence“
If you want to AI-pill your company, the world’s major AI labs will send a team to help. Call it the Palantir-ization of American technology companies. Some will be arms-length JVs, others more internal. But the gist of greasing the path towards tokenmaxxing persists regardless. The need for paywalls? Way back in the day, I helped found a startup that wanted to create a way to pay content owners (websites with words) for their work. The idea was simple: Pay a few dollars per month into an account held by my company, and we’d divvy up your funds to the websites you visited, provided they included a little bit of code from our end. Some websites signed up, and some users did too, but the model failed.
And with good reason: The technology wasn’t there yet. Users didn’t want to pay a flat rate for benefits that were vague and mostly feel-good (support the people you read!). Website owners, in contrast, wanted more control of how much they were paid. And we had to charge customers more than they were willing to chip in because of credit card fees. Hell, the entire pooling method was nothing more than a way to make payment costs a low enough percentage of user revenue to make the exchange reasonable. Since that early failure in my career, other companies have tried related approaches to the issue. They also failed. (When in doubt, don’t rebuild a graveyard?)
In the AI era, a rush of startups emerged to mediate contracts between websites and AI labs that wanted to use their data to train AI models; however, the main drag on websites and their material today is more inference-mediated than training-derived. Put another way, the AI labs have already scraped the world for their training data. Today’s common AI usage, however, fires off a barrage of search queries, which go out and collect information to present it to the user without the website in question getting paid for creating and publishing information, and serving bots that show up begging for free material.
Enter x402. We’ve mentioned the project on ** CO **before, but in simple terms, x402 builds on an old, unloved piece of Internet architecture (‘HTTP 402 Payment Required’). With the required bones already built into “
existing HTTP requests, with no additional communication required,” x402 is a neat way to charge humans and agents for accessing websites, provided the webmaster (now that’s an old-timey phrase) in question agrees to take part.
x402 is a neat blend of new technology (stablecoins, agents) and existing infra that could literally revolutionize how content is valued and paid for online. What it needs is a groundswell of support. Enter Cloudflare, which has been a loud and consistent voice against unrequited, for-profit use of the Internet by AI labs and their products:
Today, we are announcing the Cloudflare Monetization Gateway, an engine that will give Cloudflare customers the ability to charge for any asset protected by Cloudflare: web pages, datasets, APIs, or MCP tools.
It will provide a single control plane to manage payment policies and access controls across your applications, while also protecting your origin from high payment volumes by handling payment verification and enforcement at the edge. At launch, payments will settle in stablecoins over
[, the open protocol]x402[with a coalition of more than 25 industry leaders via the]we are building[.]x402 Foundation Inject it directly into my veins. Naturally, given the stablecoin element, Circle is excited. Coinbase, too, for obvious reasons.
- I don’t likepaywalling this newsletter and its constituent blog. Perhaps if there’s real revenue in becoming x402-native, we could dispense with the wall? Perhaps in the future. In the meantime, your money is welcome. (Really.)
About the latest AI funding brouhaha #
News that Nvidia has a novel approach to supporting neolabs that rent access to its GPUs drew concerned noises from technology-watchers. Nvidia, the American chip giant and the world’s most valuable company, will rent back unsold GPU capacity from neoclouds that they fail to sell. Or as The Information reports:
[Nvidia is] promising to financially backstop young cloud providesr that rent out its graphics processing units in exhange for a share of their revenues. The backstop would come in the form of promising to rent back unused GPUs if the company can’t find AI developers to rent them [according to two neoclouds and several industry executives].
My favorite AI skeptic Ed Zitron had the following read of the situation: “Yeah I’m sorry the only reason that NVIDIA is renting back its own capacity from neoclouds is that the demand doesn’t exist to sell it.”
Despite my overall bullishness on AI, some of the AI deals we’ve seen in recent quarters have been concerning. OpenAI’s promised 17.5% annual return to PE partners in its FDE program (see above), for example. SoftBank using OpenAI equity to borrow more money to invest in AI. That sort of thing. Less worrying, in my view, are compute-and-investment deals, under which hyperscalers invest in AI labs that then promise to rent a lot of their cloud footprint — those are win-win agreements in which AI labs get what they need (near-term cash) and hyperscalers what they want (long-term commitments).
- Locking in customers makes data center buildouts more economically palatable to shareholders and, increasingly, folks purchasing corporate paper intended to finance the new compute clusters.
But if not all AI dealmaking is suspect, is this particular set of transactions worrying? There may be some truth to Ed’s point about capacity — though everyone (founders, etc.) I have spoken to recently is compute-constrained, rather than compute-rich — but I think we’re again seeing a mutually beneficial transaction form. Here’s the negative take to start:
- Nvidia wants to sell chips; neoclouds want to buy chips; both expect that demand for AI compute will grow.
- However, Nvidia wants to sell more chips than the market can bear, so it’s juicing its customers (neoclouds) ability to borrow to buy its GPUs, using backstops to keep the chip sales hot while it rakes in enormous profits and ensures the AI merry-go-round rotates for another revolution or two.
And here’s how I view the situation:
- Nvidia wants to sell chips, neoclouds want to buy chips; both expect that demand for AI compute will grow.
- Neoclouds that want to buy Nvidia gear are facing higher costs of capital, making them less able to purchase GPUs. Nvidia is incredibly cash-rich and can therefore afford to backstop their purchases. More importantly, Nvidia isa foundation model lab with an eye on building open-source AI models and taking them to market with partners like Palantir. - Therefore, its backstopping of neocloud demand is more a way for it to purchase an option on future compute that it can bring to bear for its own research and AI model training if needed than a way to unfairly bolster its near-term numbers
- The revenue cut it earns for ensuring neocloud demand is nice, too.
In fewer words, if Nvidia didn’t have a real first-party use case for marginal AI compute, I’d more agree with critics that this is yet another fiasco in the making; round-tripping is a dirty phrase for a reason.
We are seeing risk accumulate in the AI compute market as parts suppliers and data center operators stretch to build capacity at breakneck speeds. Why? Because demand has thus far proved insatiable, Google is trimming customer access to some of its models to free up chips for other work. And during the Q1 2026 earnings cycle, reporting companies had nothing to say about spare capacity; again the refrain was a lack of space CPU and GPU cycles.
We’re heading into the Q2 2026 earnings run. If the hyperscalers, neoclouds, and major AI-providing and using companies report that they are seeing the demand curve for AI products soften, and thus their future capex pledges too rich, Ed’s argument will carry much more weight. If we’re seeing a data center capacity glut form, then Nvidia may really be forcing the AI market forward to its possible future financial detriment! If we are instead treated to yet another round of ‘oh god, we have no fucking spare compute,’ then I think our more cautious read (at this juncture) will be the more weighty.
Anyway, it’s not like we *weren’t *going to spazz about earnings. Now we just have extra reason to be more excited than ever about the coming numerical deluge. Onward!