DeepSeek just gave away a coding agent

What happened: DeepSeek published dsh-superagent on GitHub under an MIT license, a permissive open-source license that lets others use and change the code.

• Released October 31, 2026, and covered by The Next Web on November 1. DeepSeek calls it a research prototype, not an official product.
• Watches one GitHub repository per installation and can also listen to Feishu, the workplace chat app from ByteDance. Bloomberg reports it targets Anthropic's Claude Code.
• Needs Node.js 22.19 or later and a dedicated GitHub account. Auto-merge is off by default, and DeepSeek suggests a separate OS account because sessions run with full access.

Why it matters: A free, open coding agent puts pressure on anyone charging for one. If you sell developer tools, compete on workflow and trust, and check the security notes before your team runs it.

What happened: The Information reported that OpenAI plans to buy Google's tensor processing units, chips Google builds for AI work, on top of its Nvidia supply.

• Reported October 27, 2026, citing one source. OpenAI declined to comment to Reuters, and Google did not respond.
• Price, chip count, TPU generation and delivery schedule are undisclosed, so the size of the order is unknown.
• OpenAI began renting TPUs in June 2025, is hiring engineers to optimize its models for them, and is preparing its Broadcom-built chip, Jalapeño, for deployment by the end of 2026.

Why it matters: Your compute costs depend on who has capacity and who locks it up first. Expect more supplier options and long-term buying deals, and keep more than one chip path in your plans.

What happened: Fractile, which makes inference chips built on static random access memory instead of GPUs, appears to have raised about $774 million without announcing the round.

• Companies House filings for September and October 2026 total about £577.4 million across three share classes, mostly Series C-3 at the highest price.
• It raised a $220 million Series B about five months earlier and has an initial deal to sell Anthropic about $250 million of chips.
• Deployment in data centers is expected in 2027. An earlier Sifted figure put the total at £540 million.

Why it matters: Investors are funding alternatives to Nvidia for inference, which is the work of running models for users. If you plan specialized inference, watch whether the 2027 dates hold, since big buyers lock in capacity before chips ship.