Generalist, the physical AI startup building robots that can handle multiple tasks in warehouses and factories, just closed a $200 million extension at a $3 billion valuation. That’s a 50% jump from the $2 billion valuation it raised at earlier this year, according to sources familiar with the deal.

The extension comes as investors pour capital into robotics companies that promise general-purpose manipulation rather than single-task automation. Generalist’s pitch is that one robot platform can learn to pack boxes, sort inventory, and move materials without needing custom programming for each job. It’s the same “foundation model for robotics” bet that’s pulled in billions across the sector, but Generalist is moving faster than most on deployment.

The valuation puts it in range with other well-funded robotics plays, though still well below the outliers. What matters more than the number is the velocity. Going from $2 billion to $3 billion in months suggests either very strong traction or a market that’s willing to pay up for scarcity in a category where there aren’t many credible players at scale.

OpenAI’s chip benchmarks look real

OpenAI’s Jalapeño inference chip posted results that actually back up the hype. Tested on SemiAnalysis’ InferenceX benchmark, it delivered more tokens per user and higher throughput per kilowatt than current state-of-the-art chips.

This matters because inference is where the cost is. Training gets the headlines, but inference is the recurring expense that determines whether an AI product can be profitable at scale. If Jalapeño can deliver better performance per watt, that’s a direct hit to operating costs for anyone running models in production.

OpenAI hasn’t shared pricing or availability, but the benchmarks suggest this isn’t vaporware. The company’s been public about wanting to reduce its dependence on Nvidia, and custom silicon built specifically for transformer inference is the obvious move. Google did it with TPUs. Amazon did it with Inferentia. Now OpenAI’s doing it with Jalapeño.

The broader question is whether this kicks off another wave of custom chip projects from the big model labs. If Jalapeño economics are materially better than off-the-shelf GPUs, Anthropic and the others will have to respond.

The M&A market is moving again

Gamma acquired Lica, the Accel-backed design startup, bringing the co-founders onto a new research team. Deal terms weren’t disclosed, but it’s another signal that the AI tooling market is consolidating.

Lica built design tools that used AI to generate layouts and visual assets. Gamma makes AI-powered presentation software. The overlap is obvious, and acqui-hires like this are how the category leaders fill gaps without building from scratch.

Separately, Keenable emerged from stealth with a $26 million seed round from Accel to build a web search index specifically for AI agents. The pitch is that existing search infrastructure wasn’t designed for programmatic access at agent scale, so Keenable is indexing the web in a way that makes it easier for agents to retrieve and act on information.

It’s early, but if AI agents actually take off as a category, the infrastructure to support them, things like specialized search indexes, structured data layers, and tool integration platforms, will be valuable. Keenable is betting that search is the wedge.

Funding moves in India and image generation

Ringg raised a $10 million Series A extension from Peak XV. The Indian startup started with voice AI for phone calls but is now pushing into broader voice interfaces. The extension suggests Peak XV sees traction beyond the initial use case.

Stability AI, which makes Stable Diffusion, raised $76 million, bringing total funding to $232 million. The company’s had a messy few years with leadership turnover and questions about its business model, but it’s still one of the most widely used open-source image generators. The new capital should give it runway to figure out monetization, though the window won’t stay open forever.

Exits at OpenAI continue

OpenAI lost another senior exec. This time it’s a top data center leader, part of what’s now a steady stream of high-profile departures. The company says it “recently reorganized” its infrastructure team to support scale, which is the kind of statement you make when you don’t want to talk about why people are leaving.

It’s worth watching. OpenAI is still the category leader, but talent churn at the executive level usually means something. Either the company’s strategy is shifting in ways that don’t work for long-tenured leaders, or the internal environment has changed enough that people with options are taking them.