Gemini takes the wheel, Meta goes shopping

A big Gemini navigation overhaul and a wave of Meta acquisitions.

Here’s the rundown:

1. Google Maps' biggest update in a decade adds Gemini-powered conversational search and immersive navigation

Google Maps is rolling out two new Gemini-powered features. The first, Ask Maps, lets users search for places conversationally: users can ask complex questions like “Where can I charge my phone without having to wait in a long line for coffee?” and get personalized recommendations drawn from 300M+ places and community reviews. Google also gave Maps its biggest driving update in over a decade with Immersive Navigation, which uses AI-analyzed imagery to give drivers a clearer view of their route, alternate options, and how to find parking and entrances once they reach their destination. Ask Maps is rolling out in the U.S. and India to start, while Immersive Navigation is launching across the U.S., with broader availability to follow.

Why this matters: Google is shifting Maps from a "look it up" utility to a planning and navigation companion. Ask Maps is a lock-in play: instead of bouncing between Search, Yelp, and TikTok, users can keep their full decision flow inside one app. And by launching first in the U.S. and India, Google leverages its massive existing Android and Maps footprint to help these new search behaviors stick. Immersive Navigation builds reliability. It targets navigation pain points that typically break user confidence, like choosing the right lane or finding parking, so people are less likely to second-guess the app.

2. Meta acquires agents-only social network Moltbook

Meta acquired Moltbook, a social network for the 1.4 million AI agents that run on OpenClaw to interact with each other. The deal, expected to close mid-March, brings Moltbook founders Matt Schlicht and Ben Parr into Meta Superintelligence Labs. According to Axios, Meta’s Vishal Shah said that Moltbook allows agents to verify their identity and connect with one another on their human's behalf.

Why this matters: Meta is picking up infrastructure to identify agents and link them to humans. This is relevant because as agents proliferate, it’s increasingly important to know whose behalf they’re acting on. Moltbook’s answer was identity infrastructure: a registry where agents are verified and tied back to human owners. Meta owns this infrastructure now. The timing is notable, too, with OpenAI hiring the creator of OpenClaw, the tool Moltbook was built on, last month. In an agent’s world, security and accountability might start to determine how far autonomy will go.

3. Meta signs $27B AI infrastructure deal with Nebius

Meta signed a five-year AI infrastructure agreement with Nebius Group, a Nasdaq-listed AI cloud company, with a total contract value of up to $27 billion. Under the deal, Nebius will provide Meta with $12 billion of compute capacity starting in 2027. The capacity will run on Nvidia’s next-generation Vera Rubin platform, one of its first large-scale deployments. Meta has also committed to purchase up to $15 billion in additional compute capacity across Nebius clusters over the same period.

Why this matters: Meta secures compute, and Nebius gets the revenue to build more of it. By locking in early access to Nvidia’s Vera Rubin platform, Meta gets ahead of what will likely be intense competition for limited next-gen supply. The additional $15 billion is framed as “available” capacity, effectively an option on future demand, letting Meta secure upside without fully committing to the spend upfront. For Nebius, multi-year commitments at this scale justify building clusters faster and cement its credibility as an AI cloud player.

In other news

1. Nvidia forecasts $1T in chip sales through 2027, unveils new products at annual developer conference (Bloomberg)
2. Meta reportedly eyes cutting at least 20% of its workforce as AI costs climb (Reuters)
3. Amazon holds engineering review after outages tied to AI-assisted code changes (Financial Times)
4. Ramp acquires guest travel company Juno to expand travel capabilities for non-employees (PR Newswire)
5. Tesla’s AI chip manufacturing project Terefab set to launch next week (Reuters)

Trending: OpenAI's private equity distribution channel

OpenAI is in talks with private equity firms including TPG, Bain, Advent, and Brookfield to form a joint venture, Reuters reports. The PE firms would collectively commit around $4 billion for equity stakes and board seats in the venture, in a deal valued at roughly $10 billion pre-money. Similarly, Anthropic is reportedly in talks with Blackstone, Permira, and Hellman & Friedman to form a joint venture that would sell Claude technology to the firms’ portfolio companies.

The takeaway: enterprise AI is also a channel game. One approach is to partner with private equity firms that sit close to software budgets and operating priorities across portfolio companies. For AI platforms, this creates a more repeatable rollout path and speeds up adoption across many companies at once. What’s in it for PE firms? They can get earlier access to enterprise AI tools, like in the case of the OpenAI deal, and influence over how these tools get deployed across their portfolios. That can help portfolio companies adopt AI more consistently, and keep up as AI changes the nature of work and customer expectations.

Bonus: The AI boom is hitting home

AI is affecting more than just jobs. Renters are feeling it.

The Wall Street Journal reports that after a yearslong slump, San Francisco’s housing market is bouncing back as people move there to capitalize on the AI gold rush. Rents are up 14% year over year, the fastest growth in the US, according to Apartment List. Single-family home prices are up 23% year-over-year while condos are up 12%, according to Compass. Meanwhile, housing supply is tightening, with 35% fewer homes on the market than there were a year earlier, according to City Real Estate.

How might this housing trend impact employees’ pay packages and zip code choices? Worth watching.

3 new AI tools to try

1. Claude’s Import Memory, syncs context from ChatGPT, Gemini, Grok in under a minute – try here (Anthropic)
2. TADA, text-to-speech model – try here (Hume AI)
3. Crafting, cloud sandboxes for AI coding agents – apply here (Crafting)

Recommended reading

1. When using AI leads to “brain fry” (Harvard Business Review)
2. The state of consumer AI. Part 1: usage (Apoorv Agrawal, Partner, Altimeter)
3. The state of consumer AI. Part 2: engagement and retention (Apoorv Agrawal, Partner, Altimeter)
4. Adaptive loops and memory in transformers: Thik harder or know more? (Markus Frey, Behzad Shomali,, Ali Hamza Bashir, David Berghaus, Joachim Koehler, Mehdi Ali, Lamarr Institute, Fraunhofer IAIS, University of Bonn)

See you next week!