By 2035, data centers will be using 20% of what the US grid has available, on Revathi Advaithi's figures, and the grid is already behind.
The argument all week has been about whether AI is being built too fast. Her answer is that the electrical infrastructure underneath it has not changed in two or three decades and takes ten to fifteen years to put in the ground.
"I will tell you, we're behind on infrastructure in every way."
Advaithi runs Flex, the manufacturer splitting itself into two public companies, and she will be the chief executive of the spin-off — an AI infrastructure business building power, compute and cooling from the data center back to the utility — when it starts trading in early 2027.
The full segment is covered here so you can skip it.
Here are the 6 takeaways that matter.
👤 Guest: Revathi Advaithi, CEO of Flex and the incoming CEO of Axiom, the power and cloud infrastructure business Flex is spinning off in early 2027
🎙️ Hosts: Becky Quick and Andrew Ross Sorkin, who anchor CNBC's Squawk Box
🧩 Other segments: Wes Moore, the Governor of Maryland
📰 Published: 15 September 2026 on the Squawk Pod feed
🟢 Spotify | 🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 12 min
Key Takeaways
The grid, not the chips, is what Advaithi says the country is short of
Electrical infrastructure has not changed in two or three decades, and a build takes ten to fifteen years
Data centers are on course to take 20% of available grid supply by 2035
The spin-off is booked out for years and capacity constrained in every way
She has guided to 70% growth this year and 80% next
She understands the data center backlash and thinks the answer is an education campaign, not a pause
People are watching their utility bills go up without knowing what a data center does for them
Asked if AI will upend humanity, the woman building its power supply said no, not at all
Nothing in the order book is slowing: the constraints are labor, permitting and materials
A $4.4 billion acquisition of another power company is being funded with debt and equity
1. Axiom, Named That Morning
Becky Quick opened on the split Flex announced earlier this year, and on the news that the spin-off finally has a name.
Flex is separating into two publicly traded companies, and the new one is called Axiom. Advaithi, who is currently CEO of Flex, will run it, and it begins in early 2027.
Quick noted that Advaithi had sat down with the show in May to explain the split, and that the ground has moved fast since. Advaithi's answer to that was one line: "It's an interesting day to talk about."
Her description of what the company does is deliberately end-to-end. "It's an AI infrastructure company building power, compute and cooling all the way from data centers to utilities."
The naming took less time than Quick expected. Asked how long it took and whether she was part of the selection, Advaithi said not that long, and that they looked at a whole bunch of names.
Quick's own point was that Axiom is a surprising name to still be free — a popular word with a strong connotation, and one already used by an Axiom Space and an Axiom legal. Advaithi's answer was that neither is a public company.
On the cost of finding it, Advaithi said she did not know, and that it was not a whole lot.
2. Behind on Power Everywhere
Advaithi's framing of the opportunity is a shortfall rather than a boom, and she was explicit that she means the United States and means power in particular.
"I will tell you, we're behind on infrastructure in every way."
The gap she names is not only the data center's needs but the grid behind it, which she said is the middle of what Axiom is about.
"And so if you think about the next ten years, by 2035, data centers are going to be using 20% of what's available from the grid, which is a significant amount of power."
Efficiency, not just capacity, is how she describes the mandate. The focus is deploying power infrastructure in the most efficient way possible, so that the grid at the end of ten years is an efficient one.
Her reason for finding the job appealing is that the technology has stood still. Electrical infrastructure in the US and across the world has not changed in the last two or three decades, and the same technology used decades ago is still being deployed.
3. Booked Out and Growing 70%
The growth numbers came out in the same answer, and Quick stopped her to establish which company they belong to.
"So I'm excited about new technology, more efficient technology. We've talked about 70% growth this year, 80% growth next year."
Quick asked whether that was Flex or the spin-off. Advaithi said it is the spin-off.
"We're booked out for the next few years in terms of how far we can see. We're capacity constrained in every way because we know compute infrastructure is not enough."
The planning horizon she works to is three to five years, and within it she said she can see no demand slowdown.
4. The Data Center Backlash
Quick put the political turn to her — that opposition to data centers is building fast and has become a talking point.
Advaithi's first move was to concede the premise. "So I understand the pushback from the standpoint that the grid is not ready for what we're trying to build-out."
"People are seeing their utility bills go up." They do not fully understand the impact on the communities the data centers are being built in.
Her prescription is an education process, repeated three times in one answer — explaining what data centers actually do, how they affect a community, and how to get ahead so that people's utility bills are not the thing that moves.
The counterweight she puts against it is the cost of not building. The country is in the middle of innovation like it has never seen, and the idea of being left behind for want of infrastructure is, in her word, equally scary.
She tied the two together as one dependency: healthcare, innovation and everything else depend on the data centers and the compute infrastructure, and it is all linked.
5. She Is Not Scared of AI
Andrew Ross Sorkin turned the interview to the week's other story — the AI companies asking for a slowdown — and asked her directly.
"Are you scared that AI is going to upend humanity?" — Andrew Ross Sorkin. Her answer was immediate: "No, not at all."
Asked why the people building the models are saying it, she read it as the normal arrival of a regulatory conversation. Every significant change in technology brings a discussion about how to deploy it safely and what the framework should be, and she said this one is like none other, so the conversation is the right one to have.
What she weighs it against is healthcare. "Think about how significantly this can change diseases that happen around this world. I believe those are game changing for humanity."
On the specific calls from Elon Musk, Dario Amodei and Sam Altman: "I think they're asking all the right questions, which is, hey, let's not move too fast if we don't have the safety features around it." Her instruction back to them is to work together, deploy the safety features required, and push on the regulatory framework.
Sorkin pressed with the antitrust problem, and it is his framing rather than hers. He said that if Coca-Cola and PepsiCo got in a room to discuss the secret ingredients in their soda that would be considered collusion, and that Dario Amodei, Elon Musk and Sam Altman technically cannot get in a room together and talk about these things — so whether they should be allowed to is part of the conversation.
"But we don't have to have an all or nothing approach, right?" Move it forward with the players who can get together and work on the right framework; it may not be everything that is needed, but it is the right direction. She said she does not believe it is black and white, and called a self-imposed framework and self-imposed safety requirements a first step.
Sorkin's own observation about who is worried is the sharpest line in the exchange. "It's just so interesting that the hardware guys, the data center guys are like, we're all in. Stop worrying about this. And the people who are making the software like this is a problem." He said there is something very weird about that to him.
Advaithi's answer routes around the question rather than answering it. "It takes ten, 15 years to build-out energy infrastructure, right? That's why we are so far behind." Whether the pace of AI deployment slows or not, the energy infrastructure is behind either way — and her worry is that another decade of waiting leaves the country without a grid ready to carry it.
6. Labor and Permits, Not Money
Sorkin asked whether anything is slowing the project path — the data center pushback, higher interest rates, anything.
"the only thing slowing us down is availability of labor and kind of contracting and permitting processes" — those, she said, are the big things slowing deployment.
Material availability is the other one. She pointed to the shortages being reported and put them down to the demand and supply equation.
Outside those, no demand slowdown, in compute or in power — the two most significant parts of the deployment, and both booked out for the next few years.
Financing is not a constraint either, and she gave a live example. Advaithi said "we just announced a $4.4 billion acquisition of another power company. We're going to fund that through debt and equity. We have no concerns in terms of our ability to fund that."
Her read on why the money is there is that the returns are understood. Looking at balance sheets and the availability of money, this is a place people understand has the right benefits and the right returns over the long term.
Asked whether politics is why she and Jensen Huang have been speaking out, she said absolutely. Slowing the infrastructure deployment has long-term consequences, and "you don't want to wake up in five years and realize this is what held America behind, right?"
Her example of how new this is to the public is the transformer. People suddenly woke up and now know what one is, and have worked out that these things take a while to deploy.
Bonus Insights
Quick's aside on corporate naming was that this is now a recurring ritual — she mentioned doing it with Art and with Versant, and was surprised a word as loaded as Axiom was still unclaimed.
Advaithi never contested the utility bill complaint, which is the substance of the backlash. She treated it as a fact to be managed by explanation and by getting ahead of it, rather than a misunderstanding.
The only thing she said she is worried about is the calendar. Every answer about risk came back to the ten-to-fifteen-year lead time on power, not to demand, financing or technology.
Advaithi's bottom line is that the binding constraint on AI is electrical, not computational: the grid has not changed in decades, data centers will be taking a fifth of its available supply by 2035, and the things slowing her order book are labor, permits and materials rather than any loss of demand.
Products, Companies & Tools Mentioned
Flex (The manufacturer splitting into two public companies; Advaithi is its CEO)
Axiom (The spin-off she will run from early 2027 — power, compute and cooling from the data center to the utility, guided to 70% growth this year and 80% next, and booked out for years)
Watch the full episode:
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: