Toby Rice and Varun Sivaram offer rival deals to a public souring on data centers

Pew published a number on September 22 that the AI build-out cannot argue with. Half of Americans now say data centers are mostly bad for home energy costs, up from 38% in January, and 60% would be uncomfortable with one going up nearby. The House had already voted 417 to 3 on September 16 to make the biggest data centers pay for their own grid upgrades. In the same few days, three people with money riding on AI went on camera and each offered the public a different deal to take the heat off. Toby Rice of EQT wants permits. Varun Sivaram of Emerald AI wants a trade with regulators. Vinod Khosla wants a tax. They do not agree on who owes whom.

The gas producer says the bill came before the servers

Rice sat down on Bloomberg on September 22 with natural gas prices down more than 20% on the year, which makes him an awkward villain in a story about expensive power. His argument is that the anger is real but aimed at the wrong target.

Your electricity prices were going up before Iran. They were going up before people even started talking about AI. This is actually a really bullish demand for natural gas.

Toby Rice, September 22, 2026, source

He puts the blame on 174 gigawatts of coal and nuclear capacity the country retired and on states that block pipelines, with New York stopping gas from reaching New England as his example. Some of that holds up. Rates were rising before the AI boom. But the man selling the gas also has the most to gain from permit reform, and his poll number is looser than Pew's. He told the anchors 80% of the public is unfavorable toward data centers. Pew's 60% uncomfortable figure is bad enough without rounding up.

Americans need to understand your energy bills being high is completely unacceptable. It's completely self-inflicted. We are a superpower when it comes to producing energy. Our kryptonite though is getting infrastructure built.

Toby Rice, September 22, 2026, source

Look at what his deal asks of the voter. It asks them to wait for Congress to pass permit reform and then for pipelines to get built. That does nothing for this winter's bill in Boston.

Sivaram's deal trades flexibility for a faster hookup

Sivaram spoke at an Axios event on September 21, five days after he launched the AI Energy Management Alliance with Google, Nvidia and Anthropic, plus utilities including AES, Constellation, National Grid and NRG. Asked what had changed since Davos, he named one thing.

If I had to pick one thing, it's the political backlash against data centers, AI data centers.

Varun Sivaram, September 21, 2026, source

His fix is software that lets a data center cut its draw when the grid is stressed. In return, regulators connect it sooner. He puts the prize at 100 gigawatts of room on the existing grid, against a US target of 70 gigawatts of AI load by 2028 of which he thinks only 30 will actually get connected.

The grand bargain is regulators, governors, elected officials in America and elsewhere, they can offer faster and larger power access to the grid for data centers because if they're flexible, you don't need to build out so much new infrastructure. And in return, data centers that are good citizens are going to help to protect customer rates and keep grids reliable.

Varun Sivaram, September 21, 2026, source

Of the three pitches, this is the only one with a date on it. Emerald and Nvidia are building a 100 megawatt facility that Sivaram said will open in the fourth quarter, after an earlier target of the first half of 2026 slipped. Until it responds to a real heat wave, the 100 gigawatt figure is a model output.

The off-grid route is the one he fears

The Axios host pushed him on Amazon, which is building a gas plant with solar behind the meter instead of waiting for a utility. Sivaram would not criticize Amazon by name, but he was blunt about the model.

It's bad, frankly, for the AI companies because it's expensive to run your own private utility. It's more expensive tokens of artificial intelligence. It's bad for the public power sector. You suddenly raise the prices of all the inputs, the turbines, and the switchgear that the public power system needs.

Varun Sivaram, September 21, 2026, source

I think this is the strongest point anyone made this week, and it cuts against Rice too. A campus that builds its own gas plant keeps its load off the grid, but it still bids for the same turbines and transformers as the local utility. The ratepayer pays either way, just through a different line item.

Khosla would pay the people and let the jobs go

Khosla went on The Information's TITV on September 21 with a different theory of the backlash. The bills matter less to him than the fear about work. He expects AI to do 80% of the work in 80% of jobs and argues that shielding the jobs would mean losing the race to China.

The much more important premise I have is we have to protect people not the jobs they have because there's so much fear about AI.

Vinod Khosla, September 21, 2026, source

His funding mechanism is to tax capital gains at the same rate as wages and close breaks like stepped-up basis. He put the take at roughly $300 billion a year in one answer and about $400 billion in the next, against the roughly $250 billion a year the US paid in unemployment benefits at the COVID peak. He wants a 2028 presidential candidate to run on it. Then he conceded the thing the industry rarely says out loud.

But the other thing I think before we get to your safety risks is we haven't shown people why AI is beneficial.

Vinod Khosla, September 21, 2026, source

The productivity payoff has not arrived yet

That concession is where the week's money voices split. Joe Lonsdale told Squawk Box on September 22 that AI will lift productivity growth enough to fix the deficit: "If you get productivity up to 3, 4, 5%, which you 100% should be able to do with these things, if not significantly higher, that solves the deficit problem."

A day later, Jenny Johnson, who runs Franklin Templeton and its roughly $1.8 trillion in assets, gave the number as it stands today. Productivity growth, she said, "is at 2.5%. And I don't think any of that is AI. Nobody's figured out how to get AI really yet." She credited cloud computing and older digital tools, and reached for the long view: "Electricity took 30 years to get into manufacturing."

That gap is the whole backlash in two quotes. The costs show up now, on a utility bill with a date on it. The benefits sit in a forecast. Johnson also flagged the Magnificent Seven moving from capital-light to capital-intensive businesses with spending that has not yet hit the P&L. That is the same build-out the public is being asked to host.

Keep the power fight out of the safety fight

Sivaram closed with a warning to his own side. The case for slowing frontier AI and the case against a data center down the road are merging in public, and he wants them apart.

First of all, it would be a terrible idea to try and pace the frontier by stopping building data centers. That's not the correct lever to pull, right?

Varun Sivaram, September 21, 2026, source

He is right that they are different arguments. He is probably wrong that the public will keep them apart. A local zoning vote is the one lever a nervous voter can actually pull, whatever the cause of the nervousness.

Which of these deals could win back a town that already said no?

Only the one that shows up on a bill before the next election. Rice's permit reform is years from lowering anyone's rate. Khosla's tax needs a 2028 candidate and a Congress. The Ratepayer Protection Act, which still needs the Senate, only tells states to "consider" standards making loads over 100 megawatts pay the full incremental cost of their grid upgrades. Sivaram's flexible load is the only offer a utility commission can test this winter, which makes his fourth-quarter facility the thing to watch. If it cuts power on command during a cold snap without breaking its customers' workloads, he has a working deal. If the ribbon cutting slips again, all three pitches are still just forecasts.