THE GRID
A Gigawatt Now Costs $100 Billion
Jensen Huang stood on stage in Taipei on Monday and said a number that stopped me cold.
One gigawatt of AI factory. Up to $100 billion to build.
That's not a data center. That's a small country's GDP, packed into one campus. Huang said these costs started at $20–30 billion. Then $50–60 billion. Now the sticker price is racing toward nine figures per gigawatt.
Trump’s Last Desperate Move Could Change Everything
Trump's approval is at 36%. Midterms are seven months away. He needs a move that changes everything overnight.
He has one. It doesn't require Congress. It could add over $1 trillion to the government's balance sheet and potentially make a large number of everyday Americans very wealthy.
A president used this same move once before in 1934. It created generational fortunes. A free report explains what it is and how to get positioned before he plays it.
And he wants more of them. Nvidia's Vera Rubin platform is in full production. The supply chain is twice the size of the last gen. Huang told the crowd that every watt of compute is now revenue. Every idle watt is loss.
Meanwhile, the grid is racing to keep up. NERC's summer assessment, out May 19, found that 58 GW of new power capacity came online since last year. That sounds huge. Solar, batteries, gas — all surging.
But demand grew by 11 GW. That's faster than last year. And NERC's own director said the improved picture should not be read as falling risk.
The math is simple. Nvidia is building machines that eat gigawatts. The grid is adding capacity. But every new AI factory eats the surplus before it arrives.
Bloomberg reported this week that 30% of data center power doesn't even go to compute. It goes to cooling, lighting, overhead. Nvidia's new DSX platform tries to claw some of that back. But waste at this scale is a financial problem, not just an engineering one.
If a single gigawatt campus costs $100 billion and still wastes a third of its power, the real cost of AI isn't in the chips. It's in the grid behind them.
VOLTAGE
The 49-Point Swing Nobody Saw Coming
Nine months ago, Americans were split on data centers. Roughly even. Support and opposition in a dead heat.
Not anymore.
A new Heatmap Pro poll of 4,118 voters, published this week, found 71% now oppose a data center near their home. An outright majority — 55% — said "strongly oppose." Just 21% would support one.
This isn't a left-right split. Gallup's March poll confirmed the 71% number independently. Among Trump voters, 63% would fight a local project. Among Harris voters, 78%. Independents? Seventy-two percent opposed.
The backlash has teeth. At least 20 projects were cancelled after local pushback in Q1 alone, per Heatmap Pro data. That wiped out $41 billion in planned investment and 3.5 GW of power demand.
States are responding fast. Twenty-three states have now approved at least one large-load tariff for data centers. Oregon's PUC just signed off on Portland General Electric's framework, shifting more costs to hyperscalers. Colorado ordered Xcel to do the same.
One utility CEO put it plainly. Eversource's Joe Nolan said data centers are of "no value to our residential customer." That's the head of a major New England utility, on an earnings call, saying the quiet part out loud.
When 71% of voters oppose your project and utility CEOs won't defend it, no amount of lobbying changes the timeline. The permits slow down. The tariffs go up. And the power map shifts.
WIRED IN
Risks the Market Isn't Pricing
Data centers are tripping the grid — literally. On May 4, NERC issued a Level 3 alert, its most urgent tier, after data centers in Virginia and Texas dropped over a gigawatt of load in seconds. Multiple times. Without warning. These weren't storms or equipment failures. Protection circuits inside the data centers sensed a grid wobble and yanked the load offline instantly. That sudden drop sends frequency and voltage haywire. NERC says these events can damage generators and transformers. The deadline for grid operators to respond is August 3. But here's the contrarian angle: NERC's alert is non-binding. Nobody has to do anything. And the data centers causing the problem aren't even registered with NERC. The biggest loads on the grid have no obligation to follow its rules. That gap is a blackout waiting to happen.
Shoulder seasons are the new danger zone. NERC's summer outlook sounds good on the surface. Plenty of reserves for a normal summer. But the risks have migrated. Spring heat hit early this year, catching grids during planned maintenance windows. When generators are offline for tune-ups and a heat wave rolls in, reserves vanish fast. Three regions — New England, Saskatchewan, and the Pacific Northwest — still face elevated shortfall risk under extreme conditions. The market prices summer risk. It doesn't price a May heat wave that lands during spring outages.
The demand forecasters can't agree. NERC's January long-term assessment projected 224 GW of new summer peak demand by 2035 — a 70% jump over last year's forecast. Then Grid Strategies, a consulting firm, pushed back in March, arguing NERC overestimates how much data center load will actually show up. Meanwhile, multiple utilities trimmed their near-term data center forecasts after seeing slower-than-expected interconnection timelines. The truth is nobody knows. And when $100 billion factories depend on power that may or may not exist on schedule, the forecasting gap itself is a risk.
