Use AI or you hate America
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Friday! Fallout from the Executive Branch pressuring media groups to censor their output has continued to accumulate since we last spoke. Here’s a rundown of what current and former latenight hosts think, heck, even the WSJ Editorial Board took a stand. Spine is good, but I don’t expect a swell of rebellion amongst the party in charge of all three branches at the moment. Current attacks on academic, media, and organizational independence will get worse before they get better.
Today let’s talk about browsers, Palantir, and how much Nvidia and OpenAI are holding up the American economy. To work! — Alex
📈 Trending Up: Paying Hungary off … chances of an October rate cut … impacts from politicized central banks … Google Cloud …
📉 Trending Down: National immunity … Russian gas in Europe … interest rates in Brazil … oil prices …
Things That Matter
Google’s Gemini Gambit: After mostly shaking off being found a monopolist in the general search category, Google is pressing ahead with plans to bundle its AI tool (Gemini) into its browser (Chrome). Given Chrome’s browsing market share lead, inclusion of Gemini thereof could spike the ChatGPT rival’s usage into the stratosphere.
The list of features coming to Chrome via Gemini are what you would expect, including:
A button to ask Gemini questions about your current page (or current tabs)
A way to ask Gemini to find pages from your earlier browsing sessions
New, “deeper integration between Gemini in Chrome and your favorite Google apps, like Calendar, YouTube and Maps, so you can schedule meetings, see location details and more without leaving the page you’re on.”
Agentic capabilities in a few months’ time.
Most importantly, users will be able to use the search bar in Chrome as a chatbot box for Google Search’s AI Mode. Google was not clear whether or not this feature would be default-on when launched in a few weeks, but it’s directionally anticompetitive.
Some of that appears useful. Much of it has privacy implications that will need to be sorted out. But now that Google won’t be required to divest Chrome, and because the company is likely less scared about material punishment for using its platforms to crush competition, here we are.
Microsoft rolled out AI tools to its Edge browser earlier this year, which we missed. Why? Because I doubt that many readers of this newsletter use Edge. (If you do, hit me back and tell me why; maybe I need to give it another shake?)
I doubt that Google will turn on ‘AI Mode’ by default for all Chrome users for a long time—at least until it’s sure that sending fewer people to its traditional search tool won’t collapse its revenue. But that’s more of a when question than an if question, right?
Palantir in the UK: Louis Mosley, Palantir’s UK and European lead, recently announced that his company “signed a strategic partnership with the MoD worth up to £750 million.” That’s just over $1 billion, in case you were wondering.
It’s not the first time that Palantir has landed a big contract with the UK government; in 2023, Palantir secured a £330 million ($445 million) agreement to help handle NHS data. Per PoliticsHome, that deal led to Palantir “being used across the NHS, Ministry of Defence (MoD) and Metropolitan Police to untangle and make sense of large data sets.”
So, consider the new Palantir deal is more continuation of a theme than a new melody.
Of course, announcing the Palantir megabuy around the same time that POTUS is hobnobbing with the hereditary king of the United Kingdom doesn’t hurt relations; Palantir CEO Palmer Luckey is a notable Trump supporter.
For Palantir, the expansion of its business across the pond is great news. Palantir’s growth rate in Q2 2025 was 48%. In the United States, during the same quarter, its “US Commercial” revenue grew 93%, while its “US Government” revenue grew 53%. The implication? That Palantir’s data analytics business is growing much faster at home than abroad; the UK just helped equalize that equation.
xAI’s summer: More reporting on the foundation model company dropped via the Times since we last touched on its operations. The NYTimes digs through several of the company’s more public missteps over the summer, including the embarrassing South African genocide fiasco; that came after an employee made changes to Grok after Elon Musk had complained to “more than 100 employees in a group chat that Grok was too woke,” the Times writes.
The ‘MechaHitler’ moment came after two of its leaders tried to “make Grok’s responses edgier, aiming to draw more attention so the chatbot would go viral on X,” per the Times.
There’s an interesting collision between moving very quickly in AI, winning, and creating products that go off the rails. Google has run into related problems, it’s worth noting. And in a somewhat related fashion, OpenAI has managed to step on many of its own rakes.
The question before xAI is whether its reorgs, hirings, and firings are generating a product suite better than what its competitors offer and one that can quickly monetize so that it can continue to raise sufficient capital to continue its compute buildout.
Nvidia and OpenAI are holding up the world
Along with the Palantir deal, the UK is set to deepen its relationship with US companies thanks to a planned deal by Nvidia (American) to invest $500 million into Wayve (a British self-driving company). The Wayve deal is part of a new £2 billion commit from the American chip giant to the country, for context.
Jason and I interviewed Alex Kenadall, the co-founder and CEO of Wayve last December for This Week in Startups. Full interview here in all formats.
Nvidia is a massively profitable company, with operating cash flow of $15.4 billion in its most recent quarter. It can afford to do a lot, but it’s also making some pretty big moves:
A £2 billion investment into the UK (September 2025)
A $5 billion investment into Intel (September 2025)
A $6.3 billion deal with CoreWeave to absorb any potential excess compute capacity (September 2025)
That’s a lot of spend for any company, even the richest in the world. But it’s great stuff for the global economy. Which brings us to OpenAI.
The American tech shop added a few hundred billion dollars to Oracle’s market cap this quarter after agreeing to buy a bunch of compute from it over time. OpenAI’s rapid growth has also helped fuel Microsoft’s burgeoning cloud business. And CoreWeave’s.
If we were less specific and said, instead, that demand for AI compute from both AI model companies and the platforms that run them on behalf of customers is helping prop up the global market, we’d be closer to truth. Google Cloud has a host of startups on its servers, including (amongst a laundry list of recent wins), leading unicorns like Lovable. Simiarly, Anthropic’s quick growth is helping power Amazon’s cloud growth, as the two companies are twin legumes in a single wrapper.
We’ve riffed on this topic before, but a new datapoint makes it feel slightly more worrying than before: Capex from big tech companies has reached levels (50% to 70%) that match what we saw from AT&T back in 2000 (source report). Given how much overcapacity we saw in fiber back then, you have to wonder.
All the investment is fine so long as the demand it is built to support shows up as expected. Thus far, that has always happened, with major cloud companies spending heavily all quarter and then telling their investors during earnings calls that they remain compute-constrained. But as investments pile up, as future compute commits continue to scale, the amount of risk that AI usage has to make good grows.
In short, the pressure is on the AI application layer to continue creating new demand for AI compute. Or, more simply, the better AI applications become the less risk there is for a compute overcapacity crisis that could bring down the stock market (which is heavily concentrated in companies building or supporting the AI boom).
Startups are therefore a key component in defending our current national wealth; if they can’t use AI to build lots of stuff that customers want, we could see both AI model companies and their rack-based butlers both sharply contract.
Given that Nvidia is becoming increasingly embroiled in our current experiments in state capitalism, and much of OpenAI’s queries run on Nvidia chips (across clouds), we’re not too far off from this:
