Sarah Friar spent Friday explaining that customer demand didn't change on election night. The local ballots that did change say the argument has moved to a level OpenAI can't win with a press release.
What Sarah Friar said about OpenAI data centers
Friar, OpenAI's chief financial officer, spoke to CNBC's Sara Eisen at the Citi TMT Leadership Summit in Pebble Beach on Friday. Her answer on the midterms was blunt. "When I think about the midterms, I need to keep thinking about how do we keep investing to meet that customer demand."
She added that the work starts with relationships with the hyperscalers and "will continue to mean that we raise capital." The line that got clipped for headlines was shorter. "The demand for AI is real."
None of that surprises anyone who has watched a CFO talk through a raise. The timing is what makes it worth reading, since she said it the morning after voters in ten states had spent the week deciding on data centers.
The midterm elections that went the other way
Ballotpedia's analysis, plus results CNBC gathered, counted 41 local data center ballot measures across 10 states. Voters backed restrictions, or rejected pro-development measures, in all but one.
Local is the word that matters. These weren't national referendums on AI, and they weren't aimed at OpenAI. They were county and city questions about zoning, water, noise and who pays for the substation. CNBC reported that opposition built over the summer around environmental impact, higher utility bills and job losses.
A single measure going the other way is the detail that keeps this from being a clean story about rejection. It also says the outcomes turned on local conditions rather than a party line.
Not a verdict on AI.
What the measures had in common was a narrow kind of specificity. A county doesn't vote on artificial intelligence. It votes on a rezoning, a water contract or a tax abatement for one address, which is why a national argument produces local results.
Why AI data center policy is now a county issue
A data center needs land, a power connection and a planning approval. The last one is granted by people who live near the site, and that's the shift no document from Washington can reverse.
For a household, this shows up as a line on the utility bill, a new substation at the edge of town, and a planning meeting that runs past ten at night. For a lab, it shows up as a schedule risk that capital can't remove.
OpenAI seemed to know which way the wind was blowing. On Thursday it published a policy blueprint and a set of commitments it said would apply to company-owned and company-operated data centers, including paying for data center power. A promise to cover the electricity is a direct answer to the complaint behind most of those ballot fights. It also landed one day before the interview, which suggests the company expected the question.
The AI capex math behind the answer
Friar's argument is that the spending isn't discretionary. Training runs and inference both happen inside those buildings. If demand keeps climbing, the choice stops being whether to build at all. It's whether to build somewhere the utility will pick up the phone. What happens if no county says yes? Friar didn't address it, and no single company can answer it alone.
She also said the election results didn't dramatically change anything for OpenAI, and that the company would keep working to build trust in host communities. That's the polite version of a plan: keep building, hire locally, and let the relationships do the rest.
Markets had an opinion that week. The Nasdaq had its steepest drop in 19 months on Thursday, led by chipmakers and the neocloud companies tied to the AI buildout, then partly recovered on Friday. That's sentiment moving, not a project stopping.
The OpenAI round, the Anthropic listing and the AI capex question
Money is the other half of the answer. OpenAI closed an initial $11 billion portion of a funding round reported at a $1.25 trillion valuation. It could raise up to $9 billion more in the coming weeks, according to CNBC and Bloomberg. That's a first close, not a finished round.
The company also filed its prospectus with the SEC confidentially in June. Friar said OpenAI would continue to look into a 2027 debut. Look into is doing work there.
For a sense of what this buildout costs elsewhere, look at Anthropic. It's aiming to list on the Nasdaq in mid-November. Its prospectus shows revenue of $33.2 billion for the first nine months of the year, up roughly 1,200%. Committed spend sits at about $895.4 billion, substantially all of it related to compute and infrastructure. Those are Anthropic's figures from Anthropic's filing, not a yardstick for OpenAI's books. Commitment at that scale is why compute turned into a board-level line item instead of an IT budget.
What to watch when the planning boards meet
The next test isn't in Washington. It's a county commission agenda, a utility interconnection queue, and a ballot measure someone is already drafting for the spring. Those agendas are public, and the queue usually is too.
Watch whether OpenAI's pledge to pay for power survives contact with a local rate case, and whether the companies behind it make the same offer. Watch, too, whether the 2027 talk turns into a filing date. Both are checkable, which is more than you can say for most statements about demand.
