The most interesting release in this batch is a product launch, and it’s a security one. Rubrik has built Code Guardian, a service that runs Anthropic’s Claude Mythos 5 against customer source code. The model never touches the live repository. It works on a clone of Rubrik’s immutable, air-gapped backup copy, looks for vulnerability chains that cross files, services, authentication patterns and cloud boundaries, checks whether each chain can really be exploited, and sends the confirmed critical ones to Jira or GitHub with file-level fix guidance.

Wonderful closed a $550 million Series C at a $5 billion valuation on September 2, led by Insight Partners with Salesforce joining as a new investor. The round size is the headline. The number that decides whether the price holds is 650, the employee count, measured against annual recurring revenue tracking toward roughly $100 million by year end. That is about $154,000 of ARR per head. If the company reaches the 900 employees it told investors in March it would hit by December, the ratio drops toward $111,000.

The Information reported Wednesday that Nvidia has agreed to acquire Hugging Face for $12.9 billion, citing a person with knowledge of the agreement. Reuters carried it within the hour. Neither company commented, and no terms were disclosed: no cash-stock mix, no close timing, no retention package, no governance commitments around the platform itself. Everything that determines whether this works is still unreported. The price is the disclosure that matters. On Saturday, Business Insider reported that Hugging Face had retained a bank to gauge buyer interest at a valuation above $13 billion.

Apple’s reported move toward Chinese memory is turning into a much bigger story for the sector than the first headlines suggested, and I think the market is worrying about the wrong part of it. The reports say Apple is testing CXMT for DRAM and YMTC for NAND, aimed at iPhones and other devices sold specifically inside China, with Washington possibly deciding whether to allow it after a Trump-Xi meeting in September.

Walk any dense block in Manhattan and you will pass under one. Green pipe, plywood deck, a contractor’s sign zip-tied to the netting, a permit number nobody has read in years. The structure is called a sidewalk shed, and the official story is that it is temporary. It protects pedestrians while the facade above gets fixed. The unofficial story is that for a large number of buildings, the shed is not a stage in the repair.

Two funding rounds closed the same day at opposite edges of the AI stack. Enigma, which runs large-scale experiments exploring how humans interact with robots, emerged from stealth with a $70 million seed led by Index Ventures and Ribbit. Multiverse Computing, based in Spain and focused on shrinking large language models to reduce energy and compute costs, raised a $570 million Series C at a $1.7 billion valuation. One is betting that the interface is the unsolved problem.

Moonshot AI released the model weights for Kimi K3 under a custom license this week, landing in the middle of an unusually concentrated policy argument about whether the United States should restrict access to Chinese open-weight models at all. The release did not resolve the argument, but it changed what the argument is about. Four positions in twenty-four hours The debate produced four distinct public positions in a single day.

Reporting on Meta’s Hyperion data center in Louisiana describes a secretive negotiation process that produced tax breaks and proceeded with limited public input. The specifics are local. The pattern is not, and it is worth separating the two, because the mechanics described in Louisiana are the standard operating procedure for large data center siting across the United States. How the process is designed to work Large facility negotiations typically begin under a code name, with the identity of the company withheld from local officials and the public through non-disclosure agreements signed by economic development staff.

Two mechanisms control how search engines treat a page, they solve different problems, and confusing them is one of the most consistently damaging mistakes in technical SEO. The distinction moved from a specialist concern to a mainstream news story this week when Anthropic’s Claude conversation pages appeared in Google and Bing results despite a robots file instructing crawlers not to access them — the pages reportedly lacked a noindex directive.

Infinity.inc, an early-stage AI infrastructure research company, has closed a $15 million seed round at a $100 million post-money valuation. Touring Capital led the participation, joined by Principal Venture Partners, executives from major chip companies, researchers from OpenAI and Anthropic, and a group of angel investors. The company’s core product is Ignition, an autonomous AI agent that writes, tests, and optimizes the low-level compute kernels that determine how efficiently a chip runs AI models.