The "Kill Chrome Olympics" Has Started

And: Grok FTW?

Welcome to Cautious Optimism, a newsletter on tech, business, and power.

Thursday! Hellos and good wishes from the rainy Northeast. This morning we’re talking the latest chapter of the browser wars, Grok 4, the power of patronage, Groq’s new raise, and more.

Before all that, we’ve discussed fintech’s possible comeback in 2025 and the potential for a fintech summer of sorts. Well, Revolut is considering another share sale at a +$20 billion valuation, which backs up our read of the tea leaves vibes. To work! — Alex

• 📈 Trending Up: Friendly competition … European teeth … TSMC revenue … more regressive tax cuts? … AI content deals … you sure about that? … Varda, and the space drugs race … Taiwanese military drills … mixed employment data … hmmm … simple grading …

  • Friends in high places, after Anduril appears to have snagged an all-but-guaranteed multi-billion dollar Federal contract …

  • State ownership of yet another industrial input? After steel, that is.

• 📉 Trending Down: Electricity affordability … reading news with a straight face (context, context) … the inverse-price of capitulation … Florida’s labor market … consumer spending? … Russian fighting strength … rule of law …

Grok FTW?

As the world awaits OpenAI’s open-source model and GPT-5, xAI dropped Grok 4 yesterday and it appears to be a very strong model. While OpenAI, Anthropic, DeepSeek and Google all ranked higher on LMArena’s AI model leaderboard before the release of Grok 4, we could be set for a new leading company in the AI race.

For now, at least.

The good news for folks like myself who want faster, better, and cheaper AI models over time is that Grok 4’s metrics indicate that there’s still room for AI improvements with current technology. Concerns over an ‘AI wall’ or a ‘training data wall’ have dogged genAI models since I can recall. But the models keep getting better.

To wit, here’s the progress that xAI managed with Grok 4, per Artificial Analysis:

In raw benchmark terms, Grok 4 looks strong. Here’s a TechCrunch screenshot from the model’s debut livestream:

Impressive, yeah?

I know that about half of the people reading this are cheering Musk on, and the other half are slightly irked that his AI company appears to have a winning model on its hands. Just keep in mind that competition sharpens swords, and we’re seeing xAI push every other foundation model company to nerd harder. That’s good for everyone, no matter who owns a large chunk of equity in the company currently pushing the curve further out.

And Grok 4 may not sit atop the charts for long. OpenAI’s open-source model is reportedly coming soon — and GPT-5 is set to drop sometime this summer. Given how hot it’s been in my neck of the woods, that must be soon.

And of course, there’s Google, which is in the kitchen working like an artificially-enhanced short-order cook who is one mistake away from being fired on a night when everyone else is out sick. (That Google’s AI models are sometimes cheaper to run is a fact to keep an eye on; the power of vertical integration?)

So, while the market digests Grok 4, Grok 4 Heavy, and an upcoming coding model from xAI, just keep in mind that the law in AI-land is that whomever last grabbed the crown is next in line to defend it.

Wait, what about prices?

Right, right. You can now pay up to $300 per month to xAI for expanded Grok 3 and Grok 4 usage, and “exclusive preview” access to Grok 4 Heavy. That’s a lot of coin for a consumer-facing subscription (we’re not discussing API access here).

It was just a few days ago that we took a look at the growing list of expensive AI subscriptions. You can pay Anthropic $100 per month, OpenAI, Perplexity, and Cursor $200, and Google AI a fat $250. xAI going with $300 is new terrain, as far as I can tell.

Wait, I thought AI was getting cheaper over time, not more expensive? It depends on how close to the bleeding edge you need to be. If you can content yourself with models a generation behind, yeah, AI is pretty darn affordable today. Recall that just last month OpenAI cut the price of its o3 model by 80%. That’s pretty insane, in a good way.

But as models continue to improve, and people in the market building atop, and leaning more heavily on AI models demand the latest, providers are exploring the pricing curve to find the revenue-maximizing point. I’m mostly a 4o guy on a day-to-day basis, so I am content to wait a bit.

xAI needs the money. The company is reportedly burning $1 billion per month, a huge sum even for an AI model concern, already one of the most expensive businesses to build in history. I suppose major oil projects are also rather costly, though Exxon’s Q1 capex was $5.9 billion. Google’s capex expenses in the first quarter were more than $17 billion.

• Amazon is considering putting even more money into Anthropic, the FT reports. Given how quickly the Claude-maker is growing, I don’t think it’s too speculative a wager.

Musk’s AI group has enough cash for now. It recently raised $10 billion more worth of equity and debt, meaning that it can keep working into 2026 before it may need to return to the capital well for yet another draw. If it can access more private cash will depend, in part, on how many folks are willing to pay $300 each month for the latest Grok has on offer. We’ll see. If these numbers change, we’ll know something is afoot:

Ready for an AI browser?

Yesterday Perplexity dropped Comet, its browser that features deep integration with the AI search company’s marquee product, and a neat interface pane that is a combo query engine for your browsing activities, and an agentic system to do stuff for you. Like enter a natural language request for a Google Maps query, with Perplexity handling all the messy typing and clicking for you.

I compared Comet to Dia, The Browser Company’s own competing browser. It features a similar right-pane, but without the agentic capabilities.

Perplexity wants its browser to take off for the same reason that Google wants Chrome to stay on top of its market: Search volume. If Perplexity can secure a chunk of the AI search user market, it may be able to greatly expand its query load. That means more data, more revenue, you get the picture.

Mix in the enterprise-tuned browser from Island and there are three startups with real browser competitors in the market today. And that’s not counting what OpenAI is cooking up. The latest from OpenAI is news that its browser is real, coming soon, and should ape much of what Comet can do today. But with OpenAI’s insane market share, its slightly later entrant to the “Kill Chrome Olympics” could steal more share than what its smaller rivals can manage with a headstart.

Google is in a jam. On one hand, it controls the browser market, the search market, and has big AI market share. But if it wanted to build a truly new browser, it would necessarily have to sacrifice some of its current incomes; you can’t swap users from the trad Google search interface to an AI query page without taking a bath on AdWords income.

This is the innovators dilemma: Google could cede long-term browser market share in order to protect its near-term search incomes. That could cost it its shirt over a long enough time horizon, but AI doesn’t pay for share buybacks. Today, at least.

The other Groq

We spent a little time recently pointing out that inference provider Groq’s expansion into Europe spoke well both of its own market position and AI adoption in Europe. Well, we nailed the timing: Groq is now pipped to be prepping a $300 million to $500 million raise at a $6 billion valuation, up from the $2.8 billion price tag it was valued at last August.

• Hot damn.