Who’s raising, IPO updates, and what company is buying Grok?

Happy Friyay!

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

Friday! Welcome to the end of the week. Up top Shein has confidentially filed to go public in Hong Kong. While we would have preferred a US-listing given that you and I are both better versed at reading domestic filing documents, I’ll take the IPO either way.

• 📈 Trending Up: Spinouts … corruption … the price of bitcoin, up 8% in the last week … the agentic economy? … AI censorship … layoffs … Datafy … Coinbase shares, as the value of cryptocurrencies rise … airline stocks, after Delta earnings … silly comparisons …

  • This is fun: The New Stack has a piece up on Intel’s most recent ex-CEO’s new AI alignment benchmark. Called Flourishing AI, it intends to probe AI alignment not from a “harm prevention” viewpoint, but instead by asking how well they participate in the “active promotion of human welfare.”

• 📉 Trending Down: Math homework … investors getting work done … Chinese real estate prices … Tether dominance? …

  • This is less fun: A new study makes the argument that for complicated developers tasks that require extensive context (large open-source projects, for example), use of AI can slow developers down instead of speed them up. As AI improves and can take more code into its personal context, I expect this issue to recede. But for now, a good reminder that no new tool is panacea.

We’re counting down (again) to a tariff meltdown

After experiments in full-chested mercantilism with the United States’ recent trade deal with Vietnam, the White House is gearing up to unleash a wave of tariffs that POTUS claims he really means this time.

• For reference: Goldman Sachs found earlier this month that businesses are eating more tariff costs thus far than expected; if more of the weight is shifted to consumers, the result could be inflationary.

The latest deluge of new tariffs include a 50% tax on copper, a 35% rate applied against Canada (our largest single trading partner), a huge 50% rate on Brazil, and a blizzard of other anticipated import taxes on other nations.

While the TACO trade is all the rage — the idea that Trump Always Chickens Out, and thus new tariffs set against a future deadline won’t actually materialize — POTUS claims that this time he’s not budging on his August 1st pledged starting date. We’ll see.

The chart I can’t get out of my head is this little beauty from the Minneapolis Fed:

The underlying article is from late June, so the data is now a bit out of date, but we’re heading towards the tariff line going up (again), and therefore a likely continued rise in the effective tariff rate.

Cross that with businesses eventually passing along more price rises to consumers, and we could be staring down a hot H2 2025 inflationary environment.

• Priors alert: I’m a free trade guy. Not a complete absolutist per se, but darn close. And since I have actually read Smith and his economic theory fellow travelers, I really don’t put much stock in mercantilist economic theory. If you do, write in and tell me why the tariff experiment we’re running is going to work out in our favor.

Who’s raising money today?

• Bilt: Bilt became famous in startup circles when it was reported that its ‘pay your rent on your credit card and get points for it’ business model had led to lots of financially savvy folks farming points from their rent while not leaving balances at rest to generate interest (our coverage). Wells Fargo, its erstwhile partner, wound up reportedly eating a lot of losses. Bilt is back with a $250 million round, a new, larger $10.8 billion valuation. Its Wells Fargo deal is kaput, as you can imagine.

• Foundation EGI: Spun out of MIT (and this paper), Foundation EGI wants to build foundation models for the engineering market. The startup “combines purpose-built large language models with physics-based context and engineering best practices to tackle the complex nature of engineering,” SiliconAngle reports. It now has $23 million fresh dollars to bring against its vision.

• Agora: Agora is a startup with its own stablecoin. It’s also white-labeling its technology so that other companies can have their own stablecoin. The twist is that as the branded stables are clones of the original Agora stable, they can all interact and play nice with one another sans issue. Agora just added $50 million to its accounts, but faces competition from other startups like Stably that want to help companies of all stripes bring their own stablecoins to market.

IPO Updates

While we we await an updated S-1 from Figma, here’s the rundown on 2025 IPO performance to-date:

• CoreWeave: $40 IPO price, worth $138.29 per share at the end of regular trading yesterday +246%

• eToro: $52 IPO price, $59.37, +14%

• Circle: $27 IPO price, $202.90, +651%

• Chime: $27 IPO price, $30.81, +14%

All in the green, with two outsize post-listing winners, and two others that are doing just fine. The water is warm, y’all, don’t let Zoink hog the pool!

Ok, but what about the MechaHitler in the room

Yesterday, we took a quick look at benchmarks for xAI’s new Grok 4 model. The gist was that the model looks pretty strong, but with lots of new models expected from rival provides in the coming weeks and months, its time atop certain technical leaderboards might not be long for the world.

Reader Corey Lewis wrote in, asking (in summary):

Why would any company choose Grok 4 over models from other companies given that Grok 3 recently went off the rails. Are we ignoring the 'MechaElephant’ in the room?

It’s a very good — and politely worded — rejoinder to what we focused on yesterday, and did not. The good news is that there’s fresh reporting to take into account, so we can single stone two birds here.

Since our note yesterday, many Grok 4 users have found that when asked certain questions, the new AI model appears to spend more time than we might have anticipated checking Elon Musk’s views on the prompted query, and reading his tweets. Here’s Simon Willison, TechCrunch, and one or two others.

This is not as simple a case as Grok 4 being programmed to parrot Musk’s views in all things, but the model does — in certain circumstances — appear to repine and lean on the CEO of its parent company for direction. A lack of a system card for Grok 4 makes it hard to chase down what is going on, precisely, but I think that that’s somewhat the point.

Grok 4 is pretty darn good at a lot of stuff. But if its predecessor is at risk of being detuned to the point when it declares itself MechaHitler (at which point it was taken offline for a time), and its successor at times reliant on the views of a single, politically-minded individual, you have to wonder which startup is going to look at the offered API hook and think yes, this is better than what Anthropic and OpenAI have on offer.

The riff there is somewhat specious; model-switching programs exist that allow companies to quickly swap between models as needed, so it’s not like model lock-in is terminal today. Still, Anthropic and OpenAI make oodles of money by offering API access to their models from developers directly, meaning that how well the actual technology performs is material. That sounds tautological, but it’s 2025 and both vibes and tribalism are thriving, so it’s worth stating out loud.

What’s somewhat funny is that while Grok 4 may be perhaps too sympathetic to the views of a single person, xAI’s rivals are being threatened over their alleged, sinister, and deeply-seated hatred of the current POTUS. Yes, Missouri is kicking up dust because its AG thinks that AI models are being deliberately mean to Trump.

• Would a ban on state-level AI regulation have precluded this particular fishing expedition? I don’t think so?

It appears that when it comes to keeping politics out of AI, the industry is going to have recurring problems. Especially if single actors with their hands on the levers are keen to tune their AIs to lean in one particular direction. Which, it’s worth noting, is not a charge often leveled at Microsoft, Google, OpenAI and Meta. Irony at its finest!

More next week, you lovely humans. — Alex