Let’s eat our vegetables
Rounds, earnings, satellites, and AI
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Happy Tuesday! Spotify and PayPal reported earnings — See: Trending Down — with Snap to come after the bell. Tomorrow brings Microsoft, Meta, and Robinhood numbers. More here from our weekly economic calendar.
In startup-land, Palo Alto Networks snapping up Protect AI made the podcast. More on that transaction here, notes on Sentra’s $50 million Series B here, and a chat about Lightrun’s $70 million Series B here. It’s a busy week for interesting rounds!
And up in Canada, not only did Mark Carney (Liberal Party) win, but Pierre Poilievre (Conservative Party) actually lost his seat. After heading for a win only to get pantsed by Trump, Canada’s great political realignment came to naught. Sorry, Tobi. — Alex
📈 Trending Up: Self-focusing glasses? … cheap robot arms … the price of fly kicks … your ability to hire Carly Page … final sales to private domestic purchasers … tariff impacts … tariff detente? … Alternative Media …
Spine, thanks to the EFF standing up for former government employee Christopher Krebs and his employer were unfairly attacked by the Trump administration.
📉 Trending Down: Satya-Sam relations … worker freedom … straight grift … domestic GDP … growth at PayPal … Spotify shares on Q2 growth concerns … China’s regional hegemony? … the economy, Apollo warns …
Spine, after the Trump administration folded like a cheap tablecloth regarding select automotive tariffs. Again, if they are so great, then why cave?
Nibbles
Amazon has two items out this morning. First, its Starlink competitor is finally making it to orbit. With its first 27 space-internet satellites now above the planet, Amazon is officially competing with SpaceX. For consumers, this is great news; that there are two satellite Internet providers in the United States, and one apiece rising in the EU and China, implies that we’re heading for a true era of uninterrupted digital connectivity.
And, Amazon will show consumers how much tariffs are hurting their wallets, Punchbowl News reports in what I consider to be a prime example of how online media economics in the journalism space can become a little funny:
The Take It Down Act passes: Back in March, CO reported that while the bill had good intentions — protecting people from non-consensual intimate imagery — the way the law was written left it open to abuse. You can read the EFF’s argument regarding how it could be misused here, for more.
The bill passed the House 409-2 after passing the Senate and now heads to the White House.
Let’s see if the concerns are made warranted immediately, or if it takes more time.
Creeping authoritarianism: If our worries about the Trump admin going after private companies for whom they hire, universities for what they teach, media outlets for what they say were not enough, how about one more? Polling is, apparently, also fake now and the people who execute them are worthy of being investigated for (??) fraud.
And then, AI
There’s enough AI news every day to fill the entire length of this newsletter. I try to avoid it. Tomorrow, we’re going to look at a number of startups that I am adding to the TWiST500. Today, however, let’s eat our vegetables.
Self driving is making real progress: Past Waymo announcing 250,000 paid trips weekly and Tesla preparing some sort of launch of its own robotaxi business, it’s worth remembering that there are other companies in the mix. Pony.AI, for example. It’s in the news for coming up with a self-driving hardware stack featuring a 70% lower bill of materials than its prior iterations, and a partnership with Tencent, including a slug of the Internet giant’s “robust cloud computing, big data and AI infrastructure.”
Alibaba’s new model looks slick: I am still digesting the massive Qwen 3.0 release from Alibaba, including a host of models from the flagship Qwen3-235B-A22B (235 billion parameters) to Qwen3-30B-A3B (a smaller, mixture-of-experts) model. Along with strong metrics-based performance, Alibaba is sharing weights for the models, which is very generous of the company. Let’s see how the models rank on LMArena.
Grok 3.5 is coming: Recall that xAI, the merged entity of X (Twitter) and xAI (The Contra-OpenAI Revenge Project) is worth north of $100 billion. That means it has huge shoes to fill, both in terms of creating competitive models and yanking cash flow from them. Groke 3.5 will make its way to “SuperGrok subscribers” in an “early beat release” next week That’s soon. Per Musk, the new Grok model “accurately answer technical questions about rocket engines or electrochemistry” and can reason “from first principles “ to come up with “answers that simply don’t exist on the Internet.”
Take that, Reddit!
Also, other AI companies are working along similar lines.
OpenAI’s rising search fortunes: TechCrunch writes that “OpenAI claims its search product is growing” on the back of users making “more than a billion web searches in ChatGPT last week.” That’s actually a material number of queries. Enough to actually indicate that Google — at last — has a real rival that can do volume.
For OpenAI, the volume is both indicative of product-market fit, and chance at a huge revenue stream.
On that note, OpenAI is adding shopping features to its search product. That sound in the distance is a cash register being rung repeatedly at lightspeed. For fun, one billion queries per week is about 143 million per day. Which is just under 6 million per hour. Or just under 100,000 per minute.
Closing today, the massive search stream that OpenAI now commands is creating an entire industry: GEO, or generative engine optimization. The AI search version of SEO, in other words.
I interviewed the CEO of AthenaHQ, a recent YC export that is building GEO tools for websites and similar to track their appearance in AI search, and work to earn even more — and more prominent — inclusion. It’s not alone. Profound, another startup, is working on the same problem.
Seeing micro startup-clusters form is always good fun, and the AI boom has created several. Another fun little nexus of startup activity is tech companies working to sort out the copyright holder-AI model company training data issue. Human Native, Trainspot, Protege AI, and TollBit are all working to own that particular space.