Today, we're talking about the recent deluge of startup news, Uber's latest self-driving group project, Databricks' latest numbers, and the Mistral wave. #
- Welcome to *. Cautious Optimism, a newsletter on tech, business, and power. Modestly upbeat
Wednesday. I’m hosting a venture capital roundtable later today, which should be a bop. On the docket are: how startups should operate in a post-Mythos world, recent exit activity, fewer seed-stage startups, and more. Tune in around 1 p.m. ET if that’s your jam.
Today, we’re talking about the recent deluge of startup news, Uber’s latest self-driving group project, Databricks’ latest numbers, and the Mistral wave. To work! — Alex
P.S.: I recently caught up with Gusto’s CTO, so expect notes from that chat later this week.
📈 Trending Up #
Agentic rights in Estonia…technology companies finding a spine…data center spend…upgrading legacy IT…state capitalism…state capitalism…world models…fascism
What are the other startups up to? Tired of only reading about Anthropic? I hear you. So, here’s a few tidbits from the rest of Startup Land:
** Circle**, a platform for building and running branded online communities (think creator hubs), rolled out a host of new tools, including: an “AI partner” to help creators manage tasks, a new partnership service for “top creators,” a “redesigned course builder,” and a unified inbox for its customers.
Circle had improved its ARR to $20 million by mid-2024 from $4 million in 2021. Presumably, the company is even larger today. Let’s see how much the new product helps it grow.
** Ent Security** just came out of stealth with $100 million in funding from Decibel, Sequoia, Crosspoint, Craft, Shield, Felicis, In-Q-Tel. The startup has built a tool that watches what workers do in real-time, and soft-locks or even fully blocks them when their actions could lead to a breach or other mess.
Imagine if your design intern suddenly fired up the terminal; that might be bad! What if you could ping them and ask gently, ‘Hey, what the hell are you doing?‘ The idea is that prevention matters much more than mediation in the AI era, as agents and humans start working faster.
Ent feels intrusive, sure, but I think one of the defining facets of the AI era will be the loss of personal privacy at work. Systems are growing exponentially better at watching people do their jobs, and as companies would rather pay fewer humans than more, well, get ready for Big Boss.
**Z.ai **released GLM 5.2, a new open AI model with very impressive benchmarks and a tasty price point of $1.40 per million input tokens and $4.40 per million output tokens. According to the benchmarks, GLM 5.2 looks like one hell of a bargain. Of course, YMMV.
The best American model is blocked while Chinese models catch up? Say it isn’t so! A big round of applause for the Musk-a16z administration.
Uber’s self-driving refit: Uber’s deals with Nuro (self-driving tech) and Lucid (the Saudi-backed EV company) are bearing fruit. Announced in mid-2025, the Lucid-Nuro-Uber deal was initially designed to deploy more than 20,000 vehicles worldwide, starting with a U.S. city in 2026.
The triple alliance was expanded in April of this year, boosting the Saudi commitment in Lucid by $550 million. Uber also upped its investment in the EV maker to $500 million, with “an additional commitment of $200 million.” As a result, the partnership said it was targeting more than 35,000 vehicles, nearly double the initial pledge.
This week, the three companies said Houston will be the second American city (2027) for the group’s efforts, following the launch of its service in the San Francisco Bay Area (2026). Uber also said it has “secured a 50,000-square-foot depot facility and dedicated charging pitstop in Houston,” which indicates the physical infrastructure requirements for robotaxi fleets.
*And there’s more! *Not content to work with one pair of companies to bring self-driving taxis to the world, Uber announced a new deal with self-driving company Wavye and carmaker Stellantis to “explore the development and deployment of Level 4 (driverless) robotaxis at a global scale.”
Stellantis and Wayve already have a deal, and Uber and Wayve had also partnered up to target the global self-driving market. The trio are now forming a group to bring “safe, reliable and scalable autonomous mobility services” in both Europe and North America.
- And considering Waymo’s continued expansion, the growth of Chinese self-driving fleets, and Tesla making progress on its own robotaxi service, it seems the global autonomous car market is exploding.
- We could be on the cusp of faster expansion, too: I recently spoke with Wayve and Waabi (a Canadian company working on autonomous trucks), and left with the impression that higher-level autonomous systems are nearly ready to roll out through major car companies.
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html) Trending Down
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html)
Having a job in tech: Tracking layoffs is tricky. Tech layoffs peaked in 2022-2023 as tech companies shed staff to save cash as investor preferences changed from growth to profitability for a hot second. Now, after a multi-year decline, Layoffs.fyi data indicates that the overall pace of layoffs is picking up in 2026.
The latest additions to the list are Robinhood (10%, public), Artlist (40%, private), and Rivian (~2%, public). We’ve also seen cuts from GitLab, Salesforce, ServiceNow and OpenDoor this month.
I think we should consider this the new normal. It’s not considered shameful to lay off staff; indeed, CEOs are sometimes called bold for kicking crowds of employees out the door. The market likes that, and what gets rewarded gets repeated.
- This is the parenting version of the economics adage: “ Show me the incentives, and I’ll show you the outcome.”
Databricks going public: The data company, though still a private enterprise too scared to list, has once again reported financial results. The company says its run-rate revenues have reached $6.9 billion, up 80% from a year earlier, and higher than the $5.4 billion it disclosed in February. (Snowflake also wrapped its most recent quarter with run-rate revenues of $5.6 billion, but comparing audited GAAP numbers to cherry-picked private results isn’t fair.)
Databricks also dropped a mountain of products this week. Genie is an “all-new agentic coworker that helps business teams automate and orchestrate their work across any data — structured or unstructured, analytical or operational, inside or outside Databricks.”
That’s a big claim, one powered by “Genie Ontology, a web of all knowledge in an organization from everyone and everything,” that’s made possible by its “self-improving context layer that solves one of enterprise AI’s hardest problems: understanding a business completely from its data.”
Got all that? Databricks is making a play to leverage its data advantage (data lakes aren’t small, by definition) to create a general AI agent that can help customers get work done. Genie smells like a direct challenge to Microsoft’s ambitions to use its presence amidst oceans of customer data to build agents.
The company also revealed OpenSharing, a new open-source protocol for sharing data and AI assets (skills or models) across companies and platforms. It even has a tool to help companies manage AI spend, because why not.
Real talk, it seems that Databricks has data, agents, and workflows sorted. All it needs is an in-house AI model it can use for on-platform tasks to lower its token costs, right? I mean, just ask Microsoft. Or Cursor.
It’s Mistral’s moment #
If you are tired of ** CO **beating the drum that Mistral is a critical technology company, get bent. We have more to say. In the wake of the Anthropic-Fable-White House mess, Mistral is having a moment. Here are some posts by the French AI lab’s CEO Arthur Mensch, with our commentary sandwiched between:
Mensch:* We somehow got put in the spotlight the last few days! First we’d like to thank the organizers of the AI show for that, we can’t get enough of this stuff. I’ll say a few things about where we are and what we do.*- Yes, seeing closed-source American AI models run afoul of their government via a black-box process is very bad for global marketability. What does Mistral have planned to capitalize on the moment?
Mensch:* First, we have a nice model coming this summer – we hope it will delight and surprise in a few capabilities. This will be the start of a new family of models, fat indeed, but sparse. We’re opening up an early access program in July for key partners in research, government and the industry.*- The new model is likely focused on cybersecurity (Mistral reportedly worked on such a model earlier this year). A new family of models implies that this project will continue and improve over time. Fat means large, in terms of the number of parameters it sports; sparse means the model will use a sparse mixture-of-experts approach in artificial intelligence, implying it won’t break the bank to run.
Mensch:* This model and upcoming ones will be open-weight. We believe this is critical for our customer confidence and for the research and developer communities. You cannot own, inspect, audit, or improve a system you are only permitted to reach through someone else’s interface, especially if data recording can no longer be turned off.*- This is a direct slap to Anthropic, which turned on strict data-retention policies for Fable 5 when it was available.
Mensch:* We’ve built Studio (for deployment) and Forge (for training) as portable products, and are now hosting them on infrastructure we control. We’ll run in your VPC, your datacenter, or on our infrastructure that is decoupled from US service providers. We have capacity online, it’s growing fast, and we can help you secure it.*- Mistral wants you to know that it can get you off U.S. infra altogether. The company has big data center plans, so this is not an idle pledge.
Mensch:* We’re working with companies and governments around the world to make sure their AI systems are up and running outside of external control, improving with each model release, and with an efficient cost structure. Forge allows continuous training of models based on recorded human-AI interactions, a key unlock for efficiency.Mensch: AI, just like oil in the 20th century, is about to become the major source of leverage and power in the world. Depending on how the coming years unfold, it will either lead to a world of wealth and abundance for all, or to the worst extractive economies that the world has ever seen. We’re there to fight for the first scenario, as we progress AI research and accelerate its diffusion across the world – we’re hiring if you like the quest.*- Behold, Mistral the Good Guys.
Basically: As America shoots its own feet, Mistral is soaking up international demand and has tools on the way that should help non-US entities improve their cybersecurity without requiring Washington to approve the work.
Mistral also recently hired Brian Hall as its new CMO, whom I got to know back during his Microsoft days. Since then, Hall has worked for both AWS and Google Cloud. Go get ’em, Brian. The world needs more open models on the double.