In 2026, AI capital expenditure is projected to reach $765 billion, surpassing oil and gas spending at $681 billion for the first time in history. On October 5, CME Group will launch compute futures contracts — cash-settled derivatives on H100 and B200 GPU rental prices — putting GPU capacity in the same asset class as crude oil and natural gas. Compute is no longer just an engineering input. It is a traded commodity with a forward price curve, a political constituency, and enemies.
📖 Read the full version with charts and embedded sources on The Arc of Power →
The problem is that a $765 billion buildout does not happen in a vacuum. It happens on land that someone owns, through grids that someone else maintains, in communities where people vote. And those voters — 71% of them, according to Gallup — do not want a data center in their neighborhood. More Americans are now comfortable living next to a nuclear reactor than a server farm.
This week, the collision between capital and consent produced its most revealing moment yet. Texas Governor Greg Abbott — who in November 2025 declared Texas the "epicenter of AI development" and offered major sales tax exemptions to attract Google's $40 billion investment — went on ABC and said that AI companies have "basically dug their own grave." Within hours, President Trump fired back, calling opposition to data centers a "mistake" and declaring that the "smart ones" welcome them.
A Republican governor and a Republican president, publicly disagreeing about the value of the defining infrastructure project of the decade. That is not a policy debate. That is a fault line.
The Scale of the Collision
The numbers tell a story of acceleration without precedent. Goldman Sachs projects $765 billion in annual AI capital expenditure in 2026, growing to $1.6 trillion annually by 2031. The hyperscalers — Microsoft, Google, Amazon, Meta — are committing roughly $700 billion to compute, data centers, and supporting systems this year alone, a sharp acceleration from the $400 billion deployed in 2025.
But the infrastructure required to absorb that capital has not kept pace. Only about one-third of the roughly 12 GW planned for 2026 is actively under construction. Up to half of projects face delay or cancellation. The binding constraint is no longer chips or capital — it is the up-to-five-year wait on transformers, switchgear, and batteries. The money is ready. The grid is not.
In Virginia, where data centers account for nearly 40% of all electricity consumed, wholesale electricity prices have increased 267% over five years in data-center-heavy zones. A Carnegie Mellon study estimates that data centers could push average electricity bills up 8% nationally by 2030 — and up to 25% in hotspot markets like northern Virginia. Virginia legislators responded by proposing a dedicated rate class to shift costs from residential customers to large-scale users — projected to cut residential rates by 3.4% while raising data center rates by 15.8%.
This is the arithmetic of backlash: when an industry's infrastructure demands show up on voters' utility bills, it stops being a technology story and starts being a kitchen-table issue.
The Abbott Reversal: A Case Study in Political Gravity
Governor Abbott's transformation from data center evangelist to skeptic is the most instructive development in AI politics this year — not because it reveals hypocrisy, but because it reveals rationality.
In November 2025, Abbott actively courted the industry, offering tax exemptions and streamlined permitting. Google announced a $40 billion Texas investment. xAI broke ground on a Memphis-style gigafacility near Austin. Abbott positioned Texas as the alternative to permitting-hostile blue states.
By June 2026, Abbott had directed regulators to make data centers pay the full cost of electrical infrastructure they require and called for phasing out tax incentives. By August, he ordered regulators to audit every data center seeking to connect to ERCOT — the Texas grid — before allowing projects to proceed.
What changed was not Abbott's ideology. What changed was the political math. The Echelon Insights poll showing 62% opposition — including 54% of Republicans — arrived on the desk of a governor who governs an electricity-deregulated state where blackouts have ended political careers before. The February 2021 winter storm that killed over 200 Texans and caused $195 billion in damage remains the sharpest memory in Texas energy politics. When voters see data centers as a threat to grid reliability, a Republican governor does not need to be persuaded by environmentalists. He needs to count votes.
Trump's response was equally revealing. By calling communities that embrace data centers the "smart ones," he framed the issue in his characteristic zero-sum terms: AI infrastructure as a loyalty test. He claimed data centers were "not taking power from the grid" because companies were "building their own power plants." This is technically aspirational — some hyperscalers are pursuing on-site generation, but the vast majority of currently operating and under-construction facilities draw from the public grid. The distinction matters because it exposes the gap between federal techno-nationalism ("we must beat China in AI") and state-level populism ("my voters' electricity bills went up").
View original article on Forbes →
Three Lessons from the Fault Line
Lesson 1: Compute Is the New Oil — Including the Political Baggage
Chamath Palihapitiya's framing of AI capex surpassing oil and gas is more than a financial milestone. It is a political category shift. When CME launches compute futures in October, GPU capacity will have a visible forward price curve — the same kind of number that cable news anchors put on chyrons and politicians cite in stump speeches.
Consider what happened to oil. The moment crude oil prices became a daily-tracked, publicly quoted number, they became a political football. Presidents got blamed for high gas prices. Senators held hearings when futures traders drove up volatility. States competed for refinery investment while communities fought refinery construction. The same dynamic is now arriving for compute.
H100 one-year contract rates rose from roughly $1.70/hr in October 2025 to approximately $2.65/hr by March 2026 — a 56% move in five months. When those rates become a published index, every price spike becomes a talking point. Every talking point becomes a hearing. Every hearing becomes a regulatory action that affects where and how fast data centers get built.
ℹ️ CME Group and Silicon Data will launch two compute futures contracts on October 5, 2026 — one for the NVIDIA H100, one for the Blackwell B200. These are the first standardized derivatives for AI compute, making GPU rental prices as tradeable as natural gas.
The oil analogy carries further. Chamath describes the AI investment stack as "LPS" — Land, Power, Shell — which is literally the same vocabulary the energy industry uses. The $765 billion is not being spent on intelligence. It is being spent on real estate, electricity, and cooling systems. When an industry's cost structure looks like energy, it will be regulated like energy.
Lesson 2: The Bipartisan Backlash Is Not a Bug — It Is the Dominant Political Equilibrium
The data center backlash is now the most bipartisan issue in American politics. That sentence would have been absurd twelve months ago. It is verifiable now.
The numbers are striking. According to the Annenberg Public Policy Center, 61% of Americans oppose new data centers in their communities — up from 49% just five months earlier. That 12-point swing is the "largest shift on any AI question we have measured," according to APPC's Shawn Patterson. The partisan breakdown is the tell: 69% of Democrats oppose, 54% of Republicans oppose, 53% of Independents oppose. There is no party that owns this issue. There is no party that can ignore it.
View original article on Newsweek →
The legislative response has been staggering in its breadth. In 2026, more than 300 data center bills have been filed across 30-plus states. Arizona, Illinois, and Ohio have paused data center tax incentives. New York imposed a one-year moratorium on facilities drawing over 50 megawatts. Eight states have enacted legislation rolling back data center tax subsidies. In the first three months of 2026 alone, local opposition blocked or delayed 75 projects worth $130 billion — matching the total for all of 2025.
The issue is now reverberating through all six Senate races rated toss-ups by the Cook Political Report. In Wisconsin, Republican Tom Tiffany attacked his opponent as "Data Center David Crowley." In Texas, Democrat Gina Hinojosa ran ads linking rising electric bills directly to data centers. Pennsylvania Governor Josh Shapiro — widely discussed as a 2028 presidential prospect — issued an executive order removing fast-track permitting and declared that data center companies "can afford to be good neighbors."
This is what a political consensus against an industry looks like before it becomes law.
Lesson 3: The Safety-Capacity Paradox Is the Real Fault Line
Beneath the populist backlash, a more consequential debate is playing out between David Sacks and Dario Amodei — one that connects AI safety, open-source access, and data center permitting into a single political logic.
Sacks's argument is structural: Amodei's calls for AI safety regulation — framed as necessary guardrails against existential risk — function as a "DMV for AI" that would slow model releases and create compliance barriers that only the largest labs can clear. On the All-In Podcast, Sacks predicted the endgame bluntly: "An open source ban is coming. They're not going to call it that."
Amodei's response, published in his June 2026 essay "Policy on the AI Exponential," is that open weights simply shift power concentration from model access to whoever controls the most compute. His point is that AI is "structurally a technology that tends to concentrate power, for reasons that have nothing to do with regulation."
Here is where the data center backlash makes this debate concrete rather than philosophical. If communities restrict where data centers can be built, they restrict who can access frontier compute. If compute becomes scarce, the safety argument strengthens — because scarcity creates chokepoints that regulators can control. But if compute becomes abundant and distributed, the open-source argument strengthens — because abundance makes centralized control unnecessary and potentially counterproductive.
As investor Brad Gerstner observed, the exchange between Sacks, Gavin Baker, and Amodei represents one of the most consequential debates in tech policy. Whoever controls the narrative about AI risk controls the permitting environment for AI infrastructure. The Sacks-Amodei debate is not an academic exercise. It is a fight over whether the bottleneck on AI progress will be technical (compute supply) or political (compute permission) — and who benefits from each scenario.
⚠️ The contrarian case: moratoriums do not reduce AI demand — they relocate it. The jurisdictions that say "no" cede economic leverage to those that say "yes," creating a dynamic identical to what happened with oil refining and semiconductor fabs. Trump's "smart ones want them" framing, while crude, contains a real insight: the political cost of hosting a data center may be lower than the economic cost of losing one.
The Money Behind the Message
The political dynamics become starker when you follow the money. Brookings reports that a super PAC called "Leading the Future" launched with approximately $100 million — $50 million from Andreessen Horowitz, $50 million from OpenAI president Greg Brockman — to support AI-friendly candidates. On the other side, grassroots opposition groups like the Guardrails Alliance operate with budgets that are orders of magnitude smaller.
This funding asymmetry creates its own political backlash. Senator Josh Hawley — a Republican — warned about "a handful of companies assembling a concentration of capital, information, and political power without precedent in the American experience." Senator Elizabeth Warren — a Democrat — said tech companies "have too much power over our economy, our society, and our democracy." When a conservative populist and a progressive populist are reading from the same script, the center has moved.
What Comes Next
Three dynamics will shape the next six months of this collision:
First, compute futures will create a visible "price of AI." When CME launches the H100 and B200 contracts in October, GPU rental rates will join crude oil, natural gas, and corn as prices that appear on Bloomberg terminals and cable news chyrons. Politicians will campaign against compute price spikes the same way they campaign against gas prices. This is not speculation — it is the inevitable consequence of commoditizing a resource that an entire industry depends on.
Second, the Abbott-Trump split previews a GOP fracture. Federal techno-nationalism ("we must beat China in AI") requires maximal buildout. State-level populism ("my voters' bills went up") requires restraint. These positions are incompatible, and the midterms will force Republican candidates to choose. The Democrat side is more unified — opposition plays well across the progressive-to-moderate spectrum — which gives them a rhetorical advantage on an issue that matters in every swing district with a proposed data center.
Third, the midterms will produce the first "data center election" results. Data center opposition is already a campaign issue in all six toss-up Senate races. By November, we will know whether voter anger over utility bills and community disruption translates into actual seat changes — or whether the $100 million Leading the Future PAC can buy enough air cover to neutralize the backlash. Either outcome will reshape the permitting environment for the next cycle of AI infrastructure investment.
The $765 billion is real. The political wall is real. The question is no longer whether AI's infrastructure ambitions will meet political resistance — they already have. The question is whether the resistance reshapes the buildout, or the buildout reshapes the politics. History suggests both happen simultaneously, and the process is neither clean nor predictable.
View discussion on Hacker News →
We have been tracking this collision since June, when permitting revolts first threatened the capex supercycle. We covered the moratorium wave at 69 jurisdictions in May. We examined the regulatory capture dynamics in real time. And we analyzed Sanders' escalation just last week.
The grid is saying no. The question is who blinks first.
Originally published at The Arc of Power
