The wire went out. The announcement did not.
Bloomberg reported on November 6, 2026, that people familiar with the matter said roughly $11 billion had been wired to OpenAI as of Friday, at a valuation of $1.25 trillion. That is an initial slice of a larger round. The raise is not finished. OpenAI declined to comment.
What got scaled back is the round. That first check still went out. In late September, Bloomberg reported that OpenAI had set out to raise $30 billion or more at a $1.4 trillion valuation, and now says those plans were pared back. One person told Bloomberg the rest is expected in the coming weeks and that the round could increase to as much as $20 billion. That figure is a ceiling from a single unnamed source. It is not cash in the account.
A partial close, not a finished raise
Read the banners more narrowly than they are written. Bloomberg’s account is a first close inside an unfinished round, sourced to people familiar with the matter. It is not a company announcement, and it is not a final total.
The check OpenAI is trying to raise fell harder than the valuation it is trying to defend: from $30 billion or more toward a possible $20 billion, while the mark moved from $1.4 trillion to $1.25 trillion. Bloomberg tied the retreat to a selloff in public-market AI names and to a sharp change in sentiment around infrastructure spending and possible roadblocks to data-center construction after the midterms.
$1.25 trillion is still a step up from the last closed round Bloomberg describes, $122 billion raised in March at an $852 billion valuation. On an unfinished deal, valuation is a negotiated weather report, not an audited measure of what the business is worth. Bloomberg’s account includes no OpenAI revenue, loss, or compute figures that would let anyone justify the number from the outside.
For an operator, valuation is not the number that matters. What matters is whether the power and the permit show up on the timeline a vendor already sketched.
Who is said to be in, and who would not confirm it
None of the firms named has confirmed the terms. People familiar with the matter told Bloomberg the first-close investors were Coatue Management and T. Rowe Price, and that SoftBank Group, Sequoia Capital, Founders Fund, and Andreessen Horowitz were also in the round. Those same people said OpenAI is working with investment banks to raise from high-net-worth and retail investors. SoftBank, Founders Fund, and T. Rowe Price declined to comment. Coatue, Andreessen Horowitz, and Sequoia had not responded when Bloomberg published.
The shape of the list still matters. This is no longer only a story about one or two strategic backers writing a check to keep a lab in chips. It is a syndicate trying to pull in public-market institutions and, through banks, wealthy individuals and retail money. And the company that would eventually face public shareholders has delayed its IPO. Bloomberg says OpenAI has been raising after that delay, and points to earlier reporting that a listing would not come until 2027. Private capital is doing the work a prospectus has not yet been asked to do.
The other lab is being repriced in conversation, not in a deal
The same week, Anthropic was being marked down in private conversations, not in a completed offering. Citing its own reporting, Bloomberg’s funding story said potential buyers of Anthropic IPO shares were discussing about $1.2 trillion to $1.3 trillion, down from $1.5 trillion previously. That is a sentiment shift. It is not a priced deal.
What makes the shift harder to shrug off is the spending plan sitting next to it. Bloomberg reported on November 2, 2026, that Anthropic’s amended S-1 showed third-quarter revenue of nearly $17 billion as of September 30, against $1.4 billion a year earlier (a twelve-fold increase, on Bloomberg’s account of those rounded figures), and an intention to spend more than $895 billion in the years ahead. Those numbers are Bloomberg’s reading of the filing, not an independent pass through the SEC document, and the excerpt does not lay out a year-by-year schedule.
Nothing in this record says customers have walked away. Revenue at Anthropic, on that account, has already reached a scale most software companies never see. The forward bill for compute, power, and buildings is larger still. A lower price in an IPO conversation can mean buyers want a fatter margin of safety against that bill.
The permit is becoming the product constraint
The political reporting from the same week is more specific than a vague fear of regulation. NBC, drawing on Associated Press reporting, described a big Democratic win Tuesday night and a rough tape for AI and data-center stocks. It did not call the result a landslide, and no federal ban is described. On Thursday, that reporting said, the Nasdaq was 11 percent below its all-time high, with worries that data-center construction may slow as borrowing gets more expensive. That market read comes from NBC and the AP, not from a separate market-data vendor.
Local votes are already doing what a federal bill has not. Redbud Consulting, via NBC and the AP, counted 47 referendums in 44 municipalities across 12 states. Restrictions passed in 44 of the 45 elections where counting was complete. An AP Voter Poll found about three-quarters of U.S. voters oppose a new data center in their community, with majorities in both red and blue states. Only about a quarter said AI data centers were “crucial” to their congressional vote, below government corruption and the cost of living. Backyard opposition is broad. It was not, on this poll, the issue that brought most people to the polls.
Morgan Stanley, as reported by NBC and the AP, said decisive Democratic results carry greater expectations for restrictions that could split the country into places open to new projects and places without them. It called Texas’s election of Democrat Gina Hinojosa as governor the most important result for data-center development. She defeated Republican Gov. Greg Abbott, who had called Texas “the epicenter of AI development” and then tried to clamp down late in the campaign. Hinojosa has said she would block a permit that was not a good deal for Texans, or where residents had not had a say. The bank also said Ohio’s election of Democrat Dr. Amy Acton creates significant risk of project delays. She has pledged to block projects that do not meet certain standards.
A two-speed map, if it hardens, runs through states the industry had been counting on. A pause is not free for the town that wanted the payments, any more than an unwanted substation is free for the people who live beside it.
OpenAI’s public comment that week was about siting, not the round. The AP and NBC reported that the company published a framework on Thursday on commitments to data-center communities and asked policymakers for clear rules on how the industry builds.
Lawmakers should establish rules to advance responsible investment, protect residents, and ensure that the benefits of AI infrastructure are broadly shared.
That line, reported from the company post, is the closest OpenAI came to speaking in its own voice that week. It is aimed at neighbors and legislatures, not at limited partners.
If you buy the software, you still inherit the power plant
Most companies will never write a check into a $1.25 trillion round. They will sign a contract for AI automation, for AI agents that draft follow-ups and update a CRM, for a copilot stapled to an AI for sales process. The mistake is treating those products as detached from the financing and the permits underneath them.
Labs at this scale are pre-selling a future grid. Training and serving the models behind AI for business still depends on chips, megawatts, and buildings that now have to survive a local vote and, in some states, a new governor’s standard for what counts as a good deal. If a 2027 plan assumes always-available inference at last spring’s price, the risk to underwrite is slippage: a delayed project in Texas or Ohio, a higher cost of capital after a selloff, a private round that closes smaller than the one drawn up in September.
The tools do not stop working on Monday. The abundant-compute assumption in a lot of enterprise roadmaps does deserve a second pass. The money is still enormous. The permission to spend it on a substation next to someone’s house is the part that just got harder. The labs are not running out of believers. They are running into neighbors, and the neighbors have started winning the votes that decide where the next building goes.
Questions worth asking before you rewrite the plan
-
Did OpenAI close an $11 billion round at $1.25 trillion?
Not quite. Bloomberg reported that an initial slice of roughly $11 billion had been wired as of Friday, at a $1.25 trillion valuation, inside a larger round that has not closed. OpenAI declined to comment. The figure comes from people familiar with the matter, and the firms named as investors have not confirmed the terms.
-
Will the full raise reach $20 billion?
Unknown. One person told Bloomberg the round could increase to as much as $20 billion in the coming weeks. That is an upper estimate, not a result, and it still sits below the late-September target of $30 billion or more.
-
Is $1.25 trillion a step down?
It is a step down from the $1.4 trillion valuation OpenAI was reportedly seeking in September, and a step up from the $852 billion valuation Bloomberg assigns to the March round. Without OpenAI revenue or loss figures, neither mark can be checked against the business from this record alone.
-
Does a lower price in Anthropic’s IPO conversations mean customers left?
Not on these numbers. Bloomberg said potential buyers were discussing about $1.2 trillion to $1.3 trillion, down from $1.5 trillion. Separately, its account of Anthropic’s amended S-1 put September-quarter revenue near $17 billion, against $1.4 billion a year earlier, and forward spending above $895 billion. That reads as buyers wanting a cushion against the bill, not as a collapse in use.
-
What should you reopen in a 2027 plan if a region slows or frontier capacity gets expensive?
Three things. Ask vendors what happens to price and capacity if Texas, Ohio, or another region slips. Ask what share of the workflow can run on a smaller model if the frontier tier gets costly or queued. And do not treat a trillion-dollar headline as a delivery date for the infrastructure your roadmap assumes.
-
Did the midterms outlaw new data centers?
No. What the reporting shows is expectation and local action: restrictions passing in 44 of 45 counted referendums, campaign pledges from the governors-elect in Texas and Ohio, and a Morgan Stanley warning of a split map. That is not a named federal statute.
-
Did voters treat data centers as the election’s main issue?
The AP Voter Poll says about three-quarters oppose a new data center in their own community, and only about a quarter called AI data centers crucial to their congressional vote. Local hostility and national turnout are not the same measurement.