Monday, 19 October 2026

Tenstorrent, the AI chip company run by Jim Keller, has closed a $1.5B Series E at a $6.2B post-money valuation [1]. Trace Capital and Innovation X led the round, with ARK Invest, Fidelity, Morgan Stanley, Sandisk, T. Rowe Price, Tether Investments and XTX Markets among the other investors [1]. The valuation is more than twice the $2.6B the company was valued at when it closed a Series D of nearly $700M in late 2024, according to [2].

The company says it will use the money to deploy its Galaxy servers at scale, keep serving open-weight models for text and video generation, and develop the AI and CPU designs it licenses to chip makers [1]. It sells compute from a single card or workstation up to racks and superclusters, with an open-source software stack, and licenses its RISC-V CPU and AI cores to Samsung and LG [1]. Customers and partners include Virtu Financial, Equinix, ai& and Smallest.ai, according to [2].

Some details are not public. A spokesperson declined to say whether the round included secondary shares or how much Tenstorrent has raised in total, according to [2]. The spokesperson also declined to comment on reports this summer that Intel and Qualcomm had shown takeover interest [2]. Tenstorrent moved its headquarters from Toronto to Santa Clara in 2023; Toronto is still its largest office, with 200 of its 1,000 employees [2].

"We built one architecture and sell it at every size," Keller said [1].

Anthropic is creating the Anthropic Public Benefit Fund and, subject to stockholder approval, will commit 5% of its shares to it [1]. The board may add another 2% for every 50% rise in the company's value above a set baseline [1]. The company announced the plan on Sunday [2].

The fund is set up as a wholly owned LLC, not a 501(c)(3), so it can switch between forms of giving [1]. Anthropic gives examples: large donations to traditional nonprofits, or direct payments to specific groups of people [1]. It points to AI's effects on the labour market as one area, noting it has already committed $200 million to studying and addressing them, and says bolder approaches may be needed [1]. None of the equity in the fund will be used for lobbying or political giving, in the US or elsewhere, the company says [1].

Anthropic links the fund to earlier pledges. Its co-founders pledged to donate 80% of their equity, and about half of employees take part in an equity donation match [1]. Taken together with the new fund, the company says roughly 20% of its shares are now dedicated to the public good [1]. It contrasts the escalating structure with the escalating compensation awards tech companies commonly give executives [1].

Key details are not public yet. More on the fund's structure and governance, including who decides what it pays for and the valuation milestones that may add to its stake, will be shared "in the coming months," Anthropic says [1][2]. TechCrunch notes the announcement comes as CEO Dario Amodei has described negative public sentiment toward AI as "fundamentally a crisis of trust" [2].

Security firm NorForge says it found more than 20 publicly exposed panels of Cairn, an open-source agent engine, that it identified as likely malicious [1]. One hacker used it to break into more than 100 websites, and Reuters reports the victims included John Deere, SharkNinja and Cummins, according to [2].

The largest cluster, which NorForge calls AIA0001, wrote its instructions in Chinese and ran over 1100 attack projects across 14 panels [1]. Its targets were mostly in the US, Japan and Australia, and the aim appeared to be persistent access to e-commerce and payment systems [1]. The agent worker was named claudecode_deepseek-v4.1-flash, which NorForge says could mean a Claude Code harness running a DeepSeek model [1]. NorForge estimates an average task costs around $5, with long projects possibly over $100, and says its numbers carry considerable uncertainty [1].

NorForge puts the average success rate around 10% [1]. Successful AIA0001 projects took a little under 23 hours from start to goal [1]. The panels were unprotected and exposed hundreds of secrets from targets; 11% of those were verified valid [1]. NorForge says it is working with Reuters to notify victims and trying to get the panels taken down [1].

A John Deere spokesperson told Reuters there was no indication sensitive information was accessed, according to [2]. SharkNinja said a standalone product-support site was affected and the threat is contained, and Cummins said it is investigating a site run by a third-party vendor [2]. Cairn's creator says the platform "was never a 'hacking tool'" [2].

Microsoft has ended its involvement in Nscale's planned £2bn AI data centre in Loughton, Essex, according to the BBC [1]. Microsoft had said it would fund a supercomputer at the site and declined to comment [1]. The data centre was due to get its power supply at the end of this year, but Nscale has been told that will not happen until the 2030s, a delay first reported by the Telegraph [1].

The site was planned at 50 megawatts with room to grow to 90 megawatts, according to [2]. The supercomputer was to hold more than 23,000 Nvidia GPUs, part of the $30bn Microsoft said it would invest in the UK, according to [3]. Nscale says it remains committed to Loughton and is looking at other power sources and other UK sites [1]. It adds that the grid issue has no impact on its other contracts, and the two companies still work together in Norway, Portugal and the US [1].

The timing matters for Nscale. Its prospectus for a New York listing, filed last month, disclosed agreements with Microsoft worth $44bn running until 2033, according to [3]. The same filing showed a $1bn loss in the first six months of the year on revenue of $140.6m, according to [3].

Grid delays are common. A Roadnight Taylor survey found half of UK data centre developers expect a grid connection in one to two years but often wait three to eight, according to [2]. The site was expected to create 250 permanent jobs [1].

Integration platform Boomi has acquired Elementary Data, a data observability company, and will build its tools into the Boomi Enterprise Platform [1]. Terms were not disclosed [1]. It is Boomi's second acquisition this year, after Lunar.dev [1].

Elementary began as an open-source project in 2022 [2]. Its founders say more than 2,000 data teams rely on it, that it is downloaded over 1.1 million times a month, and that a community of around 4,000 people builds it [2]. Its products include data tests and monitors, column-level lineage, a data catalog, an MCP server that exposes context to AI agents, and agents of its own that triage issues and write fixes [1][2]. Elementary's founders say Boomi serves more than 30,000 customers [2].

The pitch is that agents now act on data with no person checking it first. "AI is only as trustworthy as the data behind it," Boomi CEO Steve Lucas said [1]. Elementary's founders write that an agent acting on stale or wrong data "takes a bad action, at scale, before anyone notices" [2].

For existing users, the company says nothing changes for now [2]. The open-source project continues, Elementary still runs as a standalone product, no migration is planned, and existing agreement and subscription terms stay unchanged "at this time" [2]. It promises to give notice in advance of any future changes [2].