Happy Birthday to Cautious Optimism
I am here to sing my own praises
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Today, last year, this little blog was renamed from Alex Writes to Cautious Optimism. Its goal was simple: to “sit between trenchant big tech and affirmative early-stage tech coverage,” offering the “benefit of the doubt for smaller companies, expectations of excellence from larger firms, and a bullish vibe on new technology.”
We mostly accomplished our aims, albeit with coverage leaning more bullish than neutral. The balance skewed positive thanks to rapidly improving AI technologies, a massive global data center buildout, and a new cohort of rapidly scaling startups. We’re being led by our nose a little, and I think that’s fine.
Again, this newsletter is not called Begrudging Pessimism!
How did the publication perform? The newsletter’s audience scaled from very small to yet modest on the back of 117% growth in the last year, though its free readership is still in four-figure land. Perhaps more importantly, CO went from making no money to a baby five-figure run rate in its first year.
All told I am happy with the results, even if I’m hardly about to quit the day job and write here full-time.
Speaking of which, I crossed the one-year mark at This Week in Startups yesterday. Between a blizzard of shows, interviews with dozens and dozens of startups, a mountain of newsletters under the TWiST banner, and the nearing-completion TWiST500, I am pretty proud of my work there. I still have some interviewing-related burr-smoothing to do — but I think that with more work and study, I can tidy up my leaks, to use a poker phrase.
A big thanks to Jason and the TWiST crew for scooping me up and letting me have more fun than most people get to have at work.
Most importantly for these pages, thank you.
CO’s subscriber growth has accelerated in recent months, which is encouraging. And even though we paywall less frequently than we should — best practices, etc — a good number of you all have put your credit card up. It means the world that I get to write for you five times each week, along with Holden’s weekly Sunday-night economic calendars. The money makes it even sweeter, of course.
Enough auto-back-patting. To work!
📈 Trending Up: Crypto enforcement? … new venture funds ... even more venture funds … drama in Trump-land … Bilibili … Arabic-language AI models … free society … Chapter 11 … Every … ChatGPT usage … Grok-WNBA … stablecoin legislation … Warby Parker stock …
📉 Trending Down: Apple-Microsoft relations … cybersecurity … paranoia … Ready to Learn … iOS … human intelligence … Luminar …
The Good News: Waymo has delivered 10 million paid rides, the company announced recently. That tally has doubled in the last five months. Presuming $15 per ride, and 5 million rides in the last 5 months, that’s $15 million per month in Waymo revenue. Sure, a $180 million annual run rate is probably a fraction of what Waymo needs to break even but, hey, it’s still an impressive ramp.
Meanwhile, over at Tesla we’re getting more notes on the rollout of its own, rival robotaxi service.
IPO Watch
Hinge Health, a medtech company focused on joint and muscle pain care, is expected to price its IPO after the bell today, and begin trading tomorrow morning. You can read the S-1 filing here. Hinge is targeting a $28 to $32 per-share IPO price range.
Backers include: 11.2 Capital, Atomico, Insight Partners, Bessemer Venture Partners, Coatue, and Tiger Global Management.
Voyager Technologies is a space tech company that—thanks to a series of acquisitions—offers everything from signals intelligence help to space airlocks to a big chunk of the Starlab project, which features inputs from Airbus, Mitsubishi, MDA Space, and Palantir, per the SEC document.
Financially, Starlab is a nine-figure, slow-growing and unprofitable company. But, with a hand in so many different elements of space technology, and lots of NASA contracts, it’s a very interesting potential wager on the future space economy for those so inclined.
Backers include smaller checks from several investors, but few names that we recognize.
Past that we’re waiting on Klarna to get off the bench, Chime to set a pricing date, and other S-1s to materialize.
Behold, a new standard
The new flex in AI is to build a standard. Anthropic started the trend with its Model Context Protocol, a “new standard for connecting AI assistants to the systems where data lives, including content repositories, business tools, and development environments.”
Model Context Protocol, or MCP for short, was joined by A2A by Google this April. A2A, or Agent2Agent, was launched with support from dozens of technology firms and is designed to empower “developers to build agents capable of connecting with any other agent built using the protocol and offers users the flexibility to combine agents from various providers,” while also “complement[ing] Anthropic's Model Context Protocol.”
In English, MCP helps your AI model ingest your data, while A2A helps AI agents talk amongst themselves.
Enter Microsoft, with its new NLWeb effort. NLWeb stands for Natural Language Web, and is designed to be “the fastest and easiest way to effectively turn your website into an AI app, allowing users to query the site's contents by directly using natural language.”
Microsoft notes in its release that each “NLWeb instance is also a Model Context Protocol (MCP) server, allowing websites to make their content discoverable and accessible to agents and other participants in the MCP ecosystem if they choose.” Pretty cool.
We now have a way for AI to access information, talk amongst itself, and better sort out the Web’s information. Sounds like a recipe for a smarter crop of self-directed AI technology, yeah? Perhaps agentic will wind up being more industry fact than 2025 buzzword.
Google’s AI deluge
It was an insane week of AI announcements. TechCrunch has a good rundown of everything that Google announced at its I/O event, but here’s the material I find the most important:
An expensive, powerful AI plan: Called Google AI Ultra, you can now pay Mountain View $250 per month for access to its newest and best models, along with “Google’s Veo 3 video generator, the company’s new Flow video editing app” and a future Gemini product called “Pro Deep Think.”
Veo 3, Imagen 4: Google’s video (Veo) and image (Imagen) models both got a refresh. We’re in the these are good and getting incrementally better stage of AI work here, but it’s still notable to see Google update both of its visual models at once. Expect more, regular improvements in this area in the coming quarters and years from, well, all quarters.
Gemini usage: We linked to OpenAI’s announcement that ChatGPT grew like a weed above, a news item that was a bit of a frontrun around Google’s own AI usage milestones. To wit, Google announced that its “Gemini AI app now has more than 400 million monthly active users,” TechCrunch writes.
Stitch, Jules: The newly announced Stitch is, per the company, “a new experiment from Google Labs that allows you to turn simple prompts and image inputs into complex UI designs and frontend code in minutes,” which certainly sounds familiar. Jules, meanwhile, is an agentic coding tool that was announced a while back and is now in public beta.
And, finally, Project Mariner — an agentic service that ingests natural-language prompts from the user to “tackle tasks simultaneously in browsers running on virtual machines — and Project Astra — a search tool that has agentic capabilities built-in.
That’s the super brief overview of the new material from Google that I think is the most important.
It’s an insane release slate. Sure, not all the products above are freely available, but Google has planted a super serious flag in the ground. Not only that Google is at the very point of the AI arrow, but that startups doing well in the market today should look over their shoulder. You can come up with startup names for most of the above entries, yeah?
And, finally, Perplexity
The Information has a brilliant report out on Perplexity’s financials. Read it in full here.
For our purposes, let’s extract at a few datapoints regarding the AI search engine’s 2024 financial results (non-exhaustive):
Gross revenue: ~$60M
Contra-revenue discounts and refunds: $27M
Net revenue: $34M
“Web services for R&D, including AWS:” $48M
OpenAI/Anthropic spend: $8M
Staff payroll: $19M
Net income: -$68M
Is it odd that I am shocked that Perplexity didn’t lose even more money? I would have expected its AI API spend to come in higher, frankly. The fact that humans cost more than double its AI costs is notable; how long until the latter is larger than the former, in line-item terms?
Worth noting: I reached out to the company for TWiST, as I had the above information docketed for the show today. The company was pretty negative on the clarity of the numbers, and their accounting. So, take them to be general and directional instead of concrete and absolute.
But the losses don’t matter. Not really. The Information writes that the startup “ended last year with about $850 million in the bank,” so all it needs to do is grow.
We can presume that Perplexity had a cracking start to 2025, it’s worth noting. How do we know that? The company is raising $500 million at a $14 billion valuation, reporting indicates. You don’t get to that valuation off the company’s 2024 revenue result unless you had a torrid Q1. Or investors have lost their minds, but, hey, it’s 2025 and everyone has.
We have to stop. More tomorrow. And, again, thank you for being here. — Alex