OpenAI raises $11bn at $1.25tn, short of $30bn target

OpenAI raises $11bn at a $1.25tn valuation, China reportedly targeted Nvidia's Jensen Huang, and DeepSeek nears OpenAI's Daybreak at hacking.

GENERAL Newsletter
H
Hugenode

OpenAI has raised $11 billion at a $1.25 trillion valuation. The round is smaller than the company wanted, and it is now courting wealthy individual investors. A New York Times report says Chinese intelligence recruited an executive at an Nvidia customer to spy on Jensen Huang, adding to pressure on Nvidia’s China chip sales as Democrats prepare to take over congressional committees. Separately, new tests suggest a free Chinese model is close to OpenAI’s best gated hacking model. These stories bear on what you pay AI vendors, where you can sell or deploy chips, and how you defend your network.

OpenAI raises $11bn at a $1.25 trillion valuation, well short of its original goal

According to Bloomberg, as reported by TechCrunch, OpenAI has raised $11 billion in a round that values it at $1.25 trillion. The round has not closed. A source told Bloomberg it could reach about $20 billion in the coming weeks. The Verge notes that OpenAI had hoped for as much as $30 billion at $1.4 trillion. That was before tech stocks fell and the midterms showed how far public opinion has turned against AI companies and data centers. In March, OpenAI raised $122 billion at an $852 billion valuation.

The Decoder lists the investors as Coatue, T. Rowe Price, SoftBank, Sequoia, Founders Fund, Andreessen Horowitz and a UAE consortium. It also reports that OpenAI is working with investment banks to bring in wealthy individual investors. The Decoder reads that as a sign the AI funding market is either tapped out or too concentrated on OpenAI and Anthropic, and argues that Anthropic’s B2B success with Claude Code is making OpenAI’s position harder. That is the outlet’s interpretation, not a disclosed fact.

Anthropic is under pressure too. It filed its S-1 last week. Bloomberg reports, per TechCrunch, that some potential IPO buyers think Anthropic could be valued as low as $1.1 trillion, against the $2 trillion investors were said to expect about a month ago. OpenAI has delayed its own IPO, and TechCrunch says a listing is expected at some point next year. Sam Altman said earlier this year that OpenAI was holding off until it could make more confident claims about its models’ safety, while adding that it would be “kind of bad for the world if OpenAI waits too long to go public.”

What it means: Nothing changes today for teams building on OpenAI or Anthropic APIs. In our view, though, labs facing tighter capital and lower valuations will feel more pressure to raise revenue per customer. When you lock in long-term commitments, assume prices and terms are more likely to move. For the market backdrop, see our editions on the Nasdaq rout and data center ballot measures.

Nvidia’s Jensen Huang reportedly targeted by Chinese intelligence through a customer’s executive

The New York Times reports, as summarized by The Verge, that China’s Ministry of State Security recruited Alice Huang, an executive at Nvidia customer Megaspeed, to spy on Jensen Huang. The two are not related. The reported aim was to learn about American AI chips and possibly influence Nvidia to push for its most advanced chips to be made available to China. The NYT describes their relationship as close, and reports that President Trump was briefed on it last year but took no action. The recruitment detail is new; last week’s NYT report had already described Huang as the target of a spy plot.

Lawmakers want Huang at a closed-door briefing. The Verge says three House committees asked him to discuss his interactions with Alice Huang; Semafor describes the request as coming from three Republican House members and concerning whether he was the target of a Chinese spy. Nvidia spokesman John Rizzo did not directly say whether Huang would attend. He said the company is confident the government would tell it if any person were a threat, and that it acts on government updates about parties of concern.

Semafor’s broader point is political. Huang’s close relationship with Trump is unlikely to shield Nvidia as Democrats prepare to take over regulatory committees and key Republicans push for stronger export controls. Semafor says a must-pass defense bill in the lame-duck session could include further limits on chip sales to China. It also says Sen. Elizabeth Warren, likely the next Senate Banking chair, has long wanted Huang to testify on Nvidia’s deals with other tech companies and its China sales.

What it means: For most buyers of Nvidia hardware outside China, nothing changes yet. If your business depends on selling or deploying advanced chips in or near China, the risk of stricter rules within months has gone up. That is our assessment of where the politics are heading, not a passed law.

DeepSeek nears OpenAI’s gated Daybreak model at hacking, as exposed Artex tools spread

Vals, a company that measures AI capabilities, tested models on simulated corporate networks, Wired reports. OpenAI’s Daybreak, its most advanced cyber model and available only to trusted partners, completed 81 percent of tasks on average across two tests. The open-weight DeepSeek model completed 73 percent, and Vals says the two scored equally on some tasks. Vals also puts the token cost of an attack at $17 to $46 for GPT-6-Sol, compared with $2 to $3 for DeepSeek. These are Vals’ own findings, as reported by Wired. OpenAI did not immediately respond to Wired.

Wired says the findings suggest closed models no longer have a clear edge in cyber skills, so efforts to limit access may be moot. Anthropic premised its limited releases, starting with Claude Mythos Preview in April, on the idea that gating would only buy time before such capabilities spread; its Project Glasswing gives trusted defenders early access. In April, Anthropic’s Logan Graham told Wired: “We need to prepare now for a world where these capabilities are broadly available in 6, 12, 24 months.” There are caveats. Wired notes that real attackers often make do with older models. Edera’s Alex Zenla thinks closed labs have likely trained stronger unreleased models.

The risk is already showing up in practice. Censys found more than 800 deployments of Artex, a Chinese-made open-source autonomous penetration testing tool released in July, reachable on the web without a password, TechCrunch reports. Censys says not every instance is malicious. Security firms have linked Artex to attacks on the Korean finance industry in September, and Microsoft says hackers used it in the October attack on ConnectWise. Many Artex-linked attacks exploited weak or known passwords and missing multi-factor authentication. Artex says it is going closed-source, but dozens of copies remain on code-sharing sites.

What it means: Assume capable attack automation is cheap and widely available. The most effective defenses here are still basic ones: enforce MFA, remove default and weak credentials, and check what you expose to the internet. Vals CEO Rayan Krishnan also argues that the cost gap gives defenders a strong reason to use open models for their own security work.

What to watch

  • Whether OpenAI’s round reaches about $20 billion when it closes in the coming weeks, and how many wealthy individual investors join.
  • Where Anthropic’s IPO prices relative to the reported $1.1 trillion to $2 trillion range.
  • Whether Jensen Huang attends the closed-door House briefing.
  • Whether China chip export controls make it into the lame-duck defense bill.
  • Whether other testers confirm Vals’ results, and whether OpenAI and Anthropic change their trusted-access policies in response.

Sources