Daily AI news, Tuesday, 20 October 2026

Good afternoon. Here's today's briefing. OpenAI is hurrying to raise money before Anthropic's IPO, Amazon's own staff admit AI has pushed the company into negative free cash flow, and the US government is removing DeepSeek from its computers. SpaceX reports results tonight.

LEAD: OpenAI races to raise about $30 billion before Anthropic goes public

Background: OpenAI was last valued at $852 billion in March, after a record $122 billion round. Anthropic is expected to list in mid-November. Some of its investors have argued for a valuation nearer $1.5 trillion than the $2 trillion it wanted, after third-quarter results came in below expectations. What happened: Reuters reported an exclusive on Monday night: OpenAI is trying to lock in investor commitments for a round of about $30 billion before Anthropic's debut. The round could value it at more than $1.4 trillion, and SoftBank, Abu Dhabi's MGX and other existing backers are in talks to join. OpenAI met long-only funds and sovereign wealth funds last week while Anthropic was holding its own investor day. It's telling investors that annualised revenue has reached $50 billion since GPT-6 Astra launched, and that it expects $70 billion by year-end. Reuters' sources say OpenAI wants the money in before Anthropic's IPO filings invite direct comparisons. The two companies count revenue differently, and OpenAI has accused Anthropic of inflating its figures, which Anthropic denies. Reuters also reports that other companies are pushing back their listings until after Anthropic's IPO. Why it matters: The two biggest AI labs are now chasing the same pool of investors at the same time, in a nervous market. Whoever sets the price first shapes how the other is valued. The cash also gives OpenAI a cushion if its own planned 2027 listing slips. Sources: Reuters via US News · Reuters via Investing.com

2. Inside Amazon: AI has made it "free cash flow negative", and some Nvidia racks are waiting in warehouses for power

Background: Amazon plans to spend about $220 billion this year, mostly on AI data centres. Earlier this month, the FT reported that it's in talks to move about $8 billion of Nvidia chips into a separate vehicle and lease them back, which would take them off its balance sheet. What happened: Business Insider obtained a recording of a September AWS staff meeting. In it, AWS finance chief John Felton joked to infrastructure head Prasad Kalyanaraman, "You've taken us free cash flow negative as a company, so congratulations." Kalyanaraman called the AI business "hugely free cash flow negative" and said he expects the company's free cash flow to stay negative for "the next couple of years or so". He also said Amazon still expects demand to outstrip its capacity in 2027. Internal emails from August projected about $5 billion of Nvidia GB300 server racks arriving that month, then roughly $3 billion a month "for the foreseeable future". Under a new process, Amazon now takes ownership of those racks earlier, so it pays sooner. Some sites had room for the racks but not enough power to switch them on, so equipment was held in warehouses nearby. Amazon calls that staging "best business practice" and disputes that it lacks a plan for installing them. Why it matters: Even the most experienced data-centre builder is paying for chips months before the power arrives to earn anything from them. Like Microsoft's exit from the Essex site yesterday, it shows that electricity, not chips, is now setting the pace of the AI build-out. Sources: Business Insider · Reuters on the FT leaseback report

3. Anthropic is in talks for a second $45 billion deal with Nscale, days after Microsoft pulled out of the Essex site

Background: In August, Anthropic signed a deal with Nscale, the Nvidia-backed AI cloud company, for 460 megawatts of capacity at its Monarch campus in West Virginia, worth up to $45 billion. Yesterday's briefing covered Microsoft's exit from Nscale's Loughton site in Essex because of grid delays. What happened: Bloomberg reports, and Reuters confirms through a source, that Anthropic is negotiating a second deal of about the same size for another 460 megawatts in a second building at Monarch, using Nvidia's next-generation Vera Rubin chips. Anthropic reportedly also has first refusal on a third building. Power at the site will come from on-site gas generators, and the first 2 gigawatts are due online by the first half of 2028. No deal has been signed. Why it matters: It's a big lift for Nscale just before its New York IPO, and it shows where AI capacity is heading: places where companies can generate their own power, not ones waiting on the national grid. Sources: Reuters via Investing.com · DCD

4. The US orders federal agencies to rip out DeepSeek, including apps built on it

Background: DeepSeek is the Chinese lab whose cheap open-weight models compete with OpenAI and Meta. A law passed last December already blocked it from intelligence agencies' national security systems, and a US report in September accused DeepSeek and five other Chinese labs of distilling American models, meaning training their own on outputs taken from US systems. What happened: The Department of Homeland Security published an order on Tuesday telling all federal civilian agencies to remove DeepSeek products from their IT systems and stop buying them. It also covers third-party products and services built with DeepSeek technology, although agencies can ask for exceptions. DHS cites concern that government users' data could be reached by Chinese authorities under Chinese law. China's foreign ministry said it opposes "politicizing technology and trade issues" and will protect Chinese companies. Why it matters: The "built using DeepSeek technology" wording reaches software companies that quietly run DeepSeek models underneath their own products, including some coding and agent tools. Any vendor selling to the US government now has to be able to show exactly which models it uses. Sources: TechCrunch · Federal Register · BBC

5. The engineer accused of bugging OpenAI is denied bail as prosecutors point to Saudi links

Background: Last week's briefing mentioned the WSJ report that an OpenAI engineer was accused of hiding a recorder in a confidential conference room. What happened: On Monday, a federal judge in San Francisco ordered Abdulrahman Alfozan, a 35-year-old Saudi national, held until trial, citing a "significant" risk that he wouldn't appear. Prosecutors say he plugged an encrypted audio recorder into the back of a TV in an OpenAI executive meeting room around 28 September. OpenAI found it on 30 September, and executives had met government officials in that room while it was there. They say he was in frequent contact with a member of the Saudi royal family, and that a Saudi military official paid for his wife and son to fly out one way the day after his arrest. He was arrested on 9 October, three days before a planned one-way flight to Riyadh, and OpenAI has fired him. He faces up to five years in prison if convicted. The Saudi government didn't respond to the BBC. Why it matters: Frontier labs are now treated as national-security targets. Expect tougher physical security and closer vetting of staff, adding to the pressure from this autumn's talent-theft fights. Sources: BBC · CNBC · Bloomberg

6. Big rounds: an Nvidia challenger more than doubles its value, and a team AI assistant becomes a unicorn

• Tenstorrent: The AI chip company led by veteran chip designer Jim Keller raised $1.5 billion at a $6.2 billion valuation, up from $2.6 billion in late 2024. Trace Capital and Innovation X led, with Fidelity, T. Rowe Price, ARK, Morgan Stanley and Tether joining. TechCrunch notes that Intel and Qualcomm reportedly showed takeover interest at up to $5 billion earlier this year. The money will go into mass-producing its Galaxy servers, which run open-weight models. TechCrunch · Reuters via CNA
• Town: Founded by former Plaid CTO Jean-Denis Greze, Town raised $108 million at a $1 billion valuation, co-led by Forerunner and Menlo, just four months after launching. Each employee gets an AI assistant that connects to email, calendars and work systems, and the assistants can now pass work to each other. TechCrunch

Why it matters: Investors are still writing big cheques for AI, even in a shaky market, as long as the company is selling inference hardware or tools that already have paying users.

7. Competitor product moves

• Greptile: The AI code-review start-up raised $50 million led by Menlo Ventures at a $750 million valuation, with Benchmark and YC joining. It says it now reviews and tests more than 1% of all pull requests opened on GitHub each day, for teams including Nvidia and Coinbase. It pitches itself as a single, independent check that sits on top of whichever coding agents a team uses. That puts it in direct competition with built-in reviewers like Cursor's. Greptile
• OpenAI: ChatGPT Slides now supports real-time collaboration for Pro, Business and Enterprise users, so teams can edit a deck together, leave comments and have ChatGPT apply them. On day 15 of its 28-day challenge, the Codex team reset usage limits instead of shipping a feature. X summary · X summary

Why it matters: Now that agents write most of the code, the fight is moving to checking that code. Independent reviewers like Greptile benefit when teams use several agents at once, which recent usage data suggests is now common.

Key hires

No notable new hires at coding or AI-agent companies since the last briefing.

Coming up

SpaceX reports third-quarter results after the US market closes today, with a webcast at 9:30pm UK time. Analysts expect revenue of about $12.9 billion. Tesla is expected to report on Wednesday. SpaceX IR