2026-11-07
By Vadym · Generated with AI, curated by me
OpenAI’s new funding round landed well short of its own ambitions on the same day Anthropic’s IPO buyers were reportedly shopping a lower price too — a sign investors are repricing frontier AI labs downward, not just writing bigger checks. Underneath that, Google bought out a stalled data center developer, a Chinese pentesting tool turned up exposed on the open internet by the hundreds, and a 32-megabyte speech model quietly showed how much capability no longer needs a trillion-dollar valuation behind it.
OpenAI has closed $11 billion of a round that could total around $20 billion once finished, valuing the company at $1.25 trillion, per Bloomberg. That’s well below the $30 billion raise at a $1.4 trillion valuation OpenAI was reportedly seeking back in late September. The pullback landed the same day Bloomberg reported that prospective buyers of Anthropic’s IPO shares are weighing a valuation as low as $1.1 trillion, down from roughly $2 trillion expected a month earlier.
Meta introduced Spaces, breakout workspaces inside its Muse AI agent where a user describes an idea in chat and Muse builds a tracker, planner, or project board around it — vacation planning, a gift tracker, a wedding planner. Spaces can connect to outside apps and keep themselves updated in the background without being asked. Zuckerberg posted that he’d built a BBQ game, an event tracker, and a home dashboard with it. The feature lands six weeks after Muse debuted to 10 million US and Canada users as of October.
Physicl raised a $10 million seed round led by Inovia Capital to close the “sim-to-real gap” in robot training. Founder Alex de Vigan, who previously built 3D product-visualization startup nfinite, is launching Studio: 10,000 simulated 3D environments and 100,000 asset models that let teams generate synthetic training data for specific problems, like spotting a cup in a cluttered, shadowed scene. Early customer SpAItial’s CEO, Matthias Niessner, also a robotics professor at TU Munich, said the real bottleneck is that “you have to curate the data by hand, and this is a very long cycle.”
Security firm Censys found more than 800 internet-facing deployments of Artex, a Chinese-made, open-source autonomous pentesting tool, running without a password. Microsoft has described Artex as a system of AI agents that can run an attack “from reconnaissance through data exfiltration with little needed human interaction,” and it’s been linked to real intrusions since its July release. Artex says it will go closed-source going forward, but dozens of existing copies remain on code-sharing sites and usable by anyone.
Google acquired an unfinished data center campus in southeastern Wyoming from Crusoe, the $30.9 billion data center developer that had planned to scale the site to 10 gigawatts with Blackstone-backed partner Tallgrass. Google had reportedly balked at the project’s cost and timeline under Crusoe and had already declined to be its customer there. Terms weren’t disclosed, but Google says it will now invest $17 billion to build the campus out itself; the two companies remain partners elsewhere, with Crusoe still building Google a separate site near Amarillo, Texas.
Zoom out and today is a market quietly recalibrating two things at once: how much frontier AI labs are actually worth, and who’s willing to carry the operational risk of building for them. OpenAI’s raise landing at $1.25 trillion instead of a hoped-for $1.4 trillion, and Anthropic’s IPO book reportedly being shopped as low as $1.1 trillion, both point the same direction — investors are pricing in a rougher two years, not a longer boom. Meanwhile Google chose to just buy out a stalled developer rather than keep waiting on one, and 800-plus exposed AI attack tools show the infrastructure holding all of this up is still catching up to the ambition. The one bright note: a 32-megabyte speech model doing useful work completely offline is a reminder that not everything in AI needs a trillion-dollar valuation to matter.
— Boba
Curated by Vadym