Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Happy Wednesday, meine lieben freunde, it’s Fed rate decision day. Investors have priced in a roughly 96% chance that JPow doesn’t cut today. If you are hungry for rate cuts, there’s currently a 30% chance of a cut in June. Expect some mean tweets if the Fed meets market expectations.
Shares of Uber are off this morning after the ride-hailing giant announced earnings (raw numbers here) that missed expectations. Don’t worry about Uber, however. The company is growing, profitable, and cash-rich. Fiverr also dropped earnings this morning (data), with a more positive market response than what Dara could summon. Shopify reports tomorrow. Let’s get to work! — Alex
📈 Trending Up: The lack of an added zero … violence between India and Pakistan … oversight … taking a swing at it … no shit … AI localization … AMD … AI-led job replacement … hopes for a stablecoin bill … Periodic … farce …
Wins for the tech-right, after the CFPB changed its mind on treating BNPL firms like credit card companies and the CFTC stopped protesting a legal ruling that allowed political betting.
📉 Trending Down: Apple’s App Store tax … no, really … ecommerce lending? … time until quantum computing takes off? … rule of law … Pete Hegseth … trade tensions with China? … maybe not … Sinofuturism …
Chart of the Day, via Apollo’s Torsten Slok:
Dispatches from the AI frontier
Google dropped an updated version of is Gemini 2.5 Pro model this week, bringing “even stronger coding capabilities,” including “meaningful improvements for front-end and UI development, alongside improvements in fundamental coding tasks such as transforming and editing code, and creating sophisticated agentic workflows.”
Very cool.
While there’s some debate about standards testing, the new model is currently atop the LMArena leaderboard. Points to Google for once again pushing the standard further.
It’s no small thing to earn a top-of-the-board ranking. Companies that build genAI models and applications atop them are racing one another to earn as much of a growing market as possible. A recent Amazon study of nearly 4,000 IT “decision makers” from around the world found that they are more focused on deploying genAI inside their companies than any other product category:
I suspect that no small amount of that planned spend will be used to accelerate development tasks, and automate work that would have fallen under G&A budgets. So, I would look at those two line-items on corporate earnings reports for the next few quarters. Do we see headcount start to drop overall? Do we see those two cost categories begin to dip? If so, when?
The other nibble to consider when we wrap our minds around the AI goldrush is just how many tokens Microsoft is processing today. And how many it will by year end. The global AI market is going to require hundreds of trillions of processed tokens this year. In fact, you could argue that the figure could reach a quadrillion in 2025 with just a little bit of growth forecasting in your numbers.
We’re turning the AI against the code
The other day I caught wind of a startup called Lightrun. It wants to scan and fix code, autonomously. And it recently raised a huge Series B. With a website banner that reads welcome “to a new era of software that fixes itself,” you get why investors are bullish.
If we’re going to use AI to write more code, and doing so means that the average quantum of written code needs more polishing, then AI tools that can handle the back half of the work will make it more palatable to use AI for the former.
In essence, Lightrun is helping grow the market for Cursor, Lovable, and their peers. (Replit! Bolt.new!)
There are other names in the mix that I am nearly ready to declare a startup cluster. Endor Labs, which I believe we’ve mentioned here before, wants to use AI to find security issues in both human and machine-generated code.
And to close the action, Ox Security just raised a $60 million Series B to, like Endor Labs, lever modern computing techniques to squash security issues in code. Or as one of its founders said: “OX is pioneering agentic code review, powered by AI and enhanced with critical thinking modules that mimic the judgment of top security engineers.”
Something that I’ve noticed when it comes to AI usage is that qualms are quickly quashed by quantifiable improvement in AI itself. There were worries that AI models were neat but lacked use cases. No one says that anymore, because AI is being used across the enterprise and consumer markets. Similarly, there was a wave of complaint from knowledgeable developers that using AI coding tools would generate a scalding soup of serious security situations. Well, now we’re using AI to fix AI-generated code.
The lesson here is simple: If AI can attack a workflow and AI can also plug the gaps generated by the original use of AI, don’t bet against the new tech.
The M&A size premium is still falling
Reading the PitchBook Q1 2025 Global M&A report this morning, the following pair of charts made me sit up:
This is North American and European data only, mind, but we’re seeing two things happen at once:
The EV/revenue multiple (a more inclusive way to calculate a price/sales ratio) that small M&A deals command is rising while:
The EV/revenue that megadeals command is falling.
As a result, the ∆ between exit multiples for big deals and their micro-siblings, is narrowing. From a 5.6x differential in 2022 to just 2.3x today. And even that last figure is down from 2.8x last year.
Takeaway? The new exit rule that bigger is always better is still true, but losing some of its edge. Will the trend gift us with a wave of smaller tech IPOs? Doubtful. But there’s more reason to get smaller M&A transactions done for equity holders because they are now getting a slightly better price in both raw, and comparative terms than they once were.
The expensive acquihire is still far from its old luster, but the trends are positive, at least.
eToro IPO updates
Closing out this morning — we’re going to have a huge show today on This Week in Startups — eToro’s newly updated IPO filing indicates that it expects to sell its shares at a price of $46 to $50 apiece. That gives it a midpoint valuation of around $4.5 billion, per RenCap.
I wonder about that price. The company’s revenue growth per the filing is single-digit, and while the company is profitable, it looks expensive. I mean, get the money if you can, but if eToro does list at its expected price I’d hazard the IPO market is more welcoming than we expected.