Welcome to the newest edition of ID Tech’s AI Update. Here’s the latest big news on the shifting landscape of AI and identity technology:
Andreessen Horowitz Partner Anjney Midha is reportedly setting up a new firm that will provide AI compute and capital – a hybrid model aimed at solving the GPU bottleneck for startups. The Silicon Valley-focused investment firm now lists Midha as a “Venture Partner” rather than a “General Partner”, suggesting that he’s well on his way in starting the effort. Midha previously led a16z’s Oxygen program, a massive private H100 cluster that let portfolio companies trade equity for below-market GPU access.
China-based AgiBot has unveiled a new platform allowing users to upload videos of human movement (dancing, gesturing, etc.) in order to trigger a robot to mimic them. The firm has been working on AgiBot World, a large embodied-AI dataset and ecosystem, and has been developing general robots; the new video platform appears to be the software-based front door to feed its data engine. It’s also reportedly planning a Hong Kong IPO for next year.
Yangtze Memory Technologies Co. (YMTC) is reportedly planning a huge IPO for next year, targeting a RMB 200–300 billion (about $27–$41 billion) valuation. Formerly under Tsinghua Unigroup and now state-linked, YMTC specializes in the kind of NAND memory chips used in smartphones, SSDs, USB drives, and data centers, and has lately been pushing into AI-specific memory technologies such as High-Bandwidth Memory and AI-optimized packaging and controllers.
Anthropic will get access to up to a million Google TPUs under a multi-year deal valued in the “tens of billions”. The deal gives Anthropic over a gigawatt of AI compute capacity starting in 2026. Anthropic has previously named AWS as its “primary cloud provider” and committed to train on its Trainium and Inferentia chips, while also using Nvidia. Now it’s diversifying by adding one of its early backers to the mix.
OpenAI is pushing deeper into agentic AI. It launched its own browser this week, “Atlas”, which features a ChatGPT sidebar that can be toggled while browsing as well as a built-in agent that can autonomously take action on a user’s behalf, such as searching through your Google Docs for a certain image you misplaced (look, I’m just trying to come up with a different example than the stock “booking a hotel” idea). OpenAI also acquired the startup Software Applications Inc., which has been working on a voice-based automated interaction platform for Apple’s macOS operating system.
Synthesia is in talks to raise funding at a valuation of about $4 billion, almost twice its $2.1 billion valuation in January. The startup’s platform gets photorealistic presenter avatars to mouth the words of its customers, aiming for a professional look suitable for things like training and HR videos. Notably, the platform restricts political, sexual, or news-anchoring use cases with its stock avatars in order to avoid deepfake misuse. Google Ventures, which has previously invested, is said to be leading the new round.
London-based Wonder Studios has raised $12 million in a seed funding round led by Atomico. It’s an AI-first creative media studio doing commercials, music videos, and some original media content, and is pitching itself as “Y Combinator for IP” – not just a production shop, but a pipeline to incubate creator-owned franchises and then scale them. Prior backers include LocalGlobe, Blackbird, Adobe Ventures, and angel investors tied to OpenAI and DeepMind.
A deepfaked news broadcast shared on Facebook falsely claimed that Ireland’s Presidential election frontrunner Catherine Connolly had withdrawn, and that her opponent had won by default. The clip featured simulated news presenters reading a fabricated news script, and then cut to video of a deepfaked Connolly announcing her withdrawal. Meta said it removed the video after being alerted to its presence. (The election was today and votes will start being counted on the weekend.)
The chatbot’s take: We got GPT-5’s aerial view on what’s going on with AI-generated media.
–
By Alex Perala
