Everything About AI
NextFin News - Alphabet's Google will spend $17 billion to build a data center campus in Wyoming, formally putting a price tag on the project two days after voters across the United States delivered a broad rebuke to the development of massive artificial-intelligence facilities. The commitment, announced on Wednesday, sets up a direct collision between the hyperscaler's accelerating AI infrastructure buildout and a wave of local political resistance that swept the midterm elections. It also exposes the central tension of the AI era: the companies building the technology have the capital to keep going, but they no longer have a public grid — or a patient public — to build it on.
A $17 Billion Bet Placed Against the Political Wind
The timing is the point. On Tuesday, American voters sided with limits on data centers in 40 of 41 local ballot measures, according to Ballotpedia's tally of the 2026 election cycle. In Texas, Democrat Gina Hinojosa won the governor's race after promising a data center moratorium, and successful Democratic candidates for governor in Georgia, Ohio and Pennsylvania campaigned on stricter regulation of these projects. Two days later, Google attached a $17 billion price tag to Project Tembo in Laramie County, Wyoming.
The backlash is unusually broad. Nearly 40 data center-related measures appeared across 10 states, turning the midterms into a sprawling referendum on the AI buildout. About 77% of voters rejected data center development, and more than half of the initiatives were first introduced by citizens rather than sponsored by elected officials. In Joliet, Illinois, nearly 80% of voters moved to restrict data centers. The tally across all ballots ran 44 to 1 against new projects.
The complaints behind those votes are specific and repeated: data centers consume large amounts of electricity and water, add little employment relative to their footprint, and — in the most politically damaging version — push up household utility bills while local residents see few of the benefits. That last point is the one Google has tried hardest to engineer away.
Yet Google is not retreating. It is relocating. Wyoming is an energy-producing state with low population density, existing fossil-fuel infrastructure, and county commissioners who already approved the site plan unanimously in January. The state also has a standing sales and use tax exemption for data center equipment, administered by the Wyoming Business Council, which an industry group estimated in April 2026 had attracted nearly $7 billion of activity to the state while forgoing about $198 million in sales and use tax revenue from 2010 to 2024. The $17 billion figure is not an outlier either: Alphabet raised its full-year 2026 capital expenditure guidance to $195 billion to $205 billion, and the Wyoming campus represents roughly 8.5% of a single year's planned spending. The company has signaled that capital expenditure will increase significantly again in 2027.
The market took the news in stride. Alphabet's Class A shares fell 1.95% to $283.31 on Wednesday, and Class C shares dropped 1.68% to $282.48, both moving with a broader market decline rather than on the announcement itself. Black Hills Corp., the utility that signed the power deal, slipped 1.08% to $83.28. As of Wednesday's close, Alphabet's market capitalization stood at about $3.46 trillion.
The Real Buildout Is Under the Ground
The most important detail of the Wyoming project is not the $17 billion. It is the 2.7 gigawatts of power.
Before Google ever announced its involvement, the project had a power problem to solve. Originally pitched in July 2025 by Crusoe and Tallgrass Energy as a 1.8-gigawatt campus called Project Jade, it won county approval in January 2026. By June, Crusoe said it had "paused its development activities" on the site "at the request of our customer." Jupiter Star Holdings, a Google subsidiary, then took over as the construction entity and rebranded the project as Project Tembo, scaling the plan to 2.7 gigawatts with a potential buildout to 10 gigawatts.
On the power side, the structure is already locked in. Black Hills Corp. announced definitive agreements effective September 30, 2026, with terms running through 2048. The utility will provide up to 590 megawatts of grid-connected energy service and manage approximately 2.1 gigawatts of Wyoming-based, third-party contracted resources through a privately managed microgrid. To support the project, Black Hills plans to invest $1.8 billion between 2027 and 2029 to construct 564 megawatts of company-owned natural gas generation. Energy service is planned to begin in late 2027 and ramp to the project's peak load in 2030.
The financial structure is designed to defuse the single most potent political attack on data centers: that they raise everyone else's electricity bills. Linn Evans, president and CEO of Black Hills Corp., said in the October 6 release: "Importantly, these agreements and supporting regulatory mechanisms are structured so that Google bears all costs associated with serving the planned data center throughout the life of the project."
"Importantly, these agreements and supporting regulatory mechanisms are structured so that Google bears all costs associated with serving the planned data center throughout the life of the project."
That sentence is the template. The hyperscaler is no longer just a large commercial customer plugging into a public grid. It is becoming a de facto utility — financing dedicated generation, managing a private microgrid, and ring-fencing its costs from residential ratepayers. The AI capital expenditure cycle is, in practice, an energy capital expenditure cycle, and the companies that can build electrons as fast as they can buy chips will set the pace.
This is not a Google-only adaptation. Across the United States, hyperscalers have increasingly turned to dedicated power arrangements — behind-the-meter gas plants, nuclear restarts, and long-term contracts that isolate data-center load from the retail grid — precisely because the public interconnection process has become the binding constraint. Project Tembo's evolution, from a startup-led proposal to a Google-owned, utility-backed, microgrid-dependent campus, is the pattern in its most advanced form.
Water is the other constraint being managed in advance. Google told Laramie County it is committed to replenishing 120% of the water it consumes through regional watershed projects, and the campus plan includes evaporative wastewater lagoons alongside four data center buildings, a logistics center, two core network facilities and a two-story office building. A Wyoming industry fact sheet put the water argument bluntly: modern data centers in the state use air-cooled or closed-loop systems, and data centers account for about 1.2% of Cheyenne's municipal water supply. Whether that reassures voters in a drought-prone region remains untested.
The Money Is Not the Constraint — Power, Water and Permits Are
Alphabet can afford the $17 billion. The question is whether it can build the project on schedule.
The company's financial capacity is not in doubt. In the second quarter of 2026, consolidated revenue reached $119.8 billion, up 24% year over year — the 12th consecutive quarter of double-digit growth. Google Cloud revenue rose 82% to $24.8 billion, with operating margin expanding to 35.6%. Operating cash flow was $39.1 billion in the quarter and $185.7 billion over the trailing 12 months; the company ended June with $242.5 billion in cash and marketable securities against $98.2 billion of long-term debt. On that call, management updated full-year 2026 capital expenditure guidance to $195 billion to $205 billion, up from $180 billion to $190 billion, citing "an acceleration in the delivery of capacity to meet growing demand." In the second quarter alone, capital expenditure reached $44.9 billion, with approximately 60% of technical-infrastructure investment going to servers and 40% to data centers and networking equipment.
"We continue to expect our CapEx to increase significantly in 2027, and we'll provide more details at a later date."
But the physical constraints are binding in a way that balance sheets cannot override. Grid interconnection queues across the United States stretch into years; in the country's largest data center market, average wait times have reached seven years, and some data center projects face potential delays of up to 12 years. Transformer supply remains tight. Water rights in the arid West are contested. Skilled construction labor is finite. A single 2.7-gigawatt campus consumes electricity comparable to several large conventional power plants running continuously.
The timeline Google has outlined reflects that reality. Eric Murray, Google's land acquisition and development lead, said mass grading could begin later in 2026, followed by vertical construction in 2027. Full buildout is estimated for the first half of 2031, with around 2,500 workers on site at peak construction. That is a five-year horizon for one campus — a marathon, not a sprint, and every permit, every water ruling and every interconnection study sits between the announcement and the first server rack.
The second-order consequence is where the story widens beyond one company. A hyperscaler that finances its own generation does not just solve its own queue problem; it changes the shape of the power market around it. Dedicated gas generation pulls natural gas demand forward and anchors regional load. Private microgrids that manage third-party resources create a new class of contracted capacity outside the traditional utility planning process. And when the largest electricity customers stop paying into the shared grid's fixed-cost base, the question of who funds transmission expansion shifts to the customers who remain. The buildout is not only a capital allocation decision; it is a restructuring of how American electricity gets paid for.
The strategic implication runs deeper than one site. If dedicated power and local permission are the scarce inputs, then the hyperscalers' competitive advantage shifts from who has the best model to who has secured the most electrons in the friendliest jurisdictions. Cheap inference is not just a function of chip design anymore; it is a function of power purchase agreements signed years in advance.
The Counter-Thesis: Local Revolt Cannot Stop a National Buildout
The strongest case against reading the ballot results as a threat to the AI buildout is that the opposition is fragmented, local and geographically concentrated. The measures that passed were overwhelmingly municipal or county-level; state-level policy in energy-producing regions remains pro-development. Wyoming courted the project. Rural communities in Texas, Louisiana and the Plains see data centers as jobs, construction work and tax base. The 44-to-1 tally sounds overwhelming, but it aggregates small jurisdictions, and the governorships that flipped toward restriction sit alongside states where development is actively welcomed.
There is also a financial counter-argument, and it comes from the investors footing the bill. Alphabet's shares fell after the second-quarter earnings call on the size of the capital expenditure increase, and peers have seen similar reactions. The bear case is not that the projects will be stopped; it is that the returns will not justify the spending. If AI revenue fails to compound at the pace embedded in a $195 billion to $205 billion annual capital program — if Cloud's 82% growth rate normalizes before the infrastructure pays for itself — then the constraint is not politics or power at all. It is the cost of capital, and no private microgrid can hedge that.
That argument is correct as far as it goes — and it misses the compounding cost of friction. Even if no single ballot measure stops a project, each community fight adds months. Each permit denial forces a redesign. Each water restriction raises the cost of cooling. The buildout continues, but the marginal project becomes slower and more expensive, which is precisely why Google did not wait for the political dust to settle. It locked in a 22-year power contract, a private microgrid and a cost-segregation structure before breaking ground.
The counter-thesis also assumes the backlash stays local. The falsifying line to watch is state-level action. If ballot measures begin passing at the state level in energy-friendly jurisdictions, or if legislatures in Wyoming, Texas or Louisiana impose moratoriums rather than incentives, the jurisdictional-arbitrage model breaks down. So far, the evidence points the other way.
What Comes Next: A Bifurcated Buildout
The cyclical-versus-structural call here cuts two ways, and both forces are structural. The AI compute buildout is a regime shift in corporate capital allocation, not a cycle that will mean-revert on its own. The local political backlash is equally structural: communities now see data centers as concentrated consumers of land, water and power with diffuse benefits. These two structural forces will coexist, and the adaptation layer — private generation, remote siting, no-cost-shifting contracts — is the durable response.
Beneficiaries: independent power producers, natural-gas generation developers, and utilities that can ring-fence data-center costs the way Black Hills has. Black Hills expects the project to contribute approximately $150 million of net income in 2030 and roughly $2.4 billion of unlevered free cash flow through 2048, net of its $1.8 billion generation investment — a template other regional utilities will study closely. Rural jurisdictions with cheap power and permissive zoning gain a new tax base.
Exposed: utilities serving dense grids without cost segregation, where data-center load shows up on residential bills; suburban communities fighting projects without offering alternative sites; and hyperscalers that delayed securing dedicated power and now face longer queues and higher prices.
Time horizons: In the short term, expect more ballot measures, more local fights, and volatility in utility stocks tied to data-center exposure, while hyperscaler shares remain under pressure from capex intensity. Over the medium term, the buildout continues but likely slower than capex guidance implies, and private microgrids become standard for campuses above 500 megawatts. Over the long term, AI infrastructure bifurcates: grid-connected campuses in friendly jurisdictions, and islanded gas-powered campuses where the public grid — or the public's patience — cannot support them.
Scenarios: The base case is that Google energizes Project Tembo in late 2027 and ramps to 2.7 gigawatts by 2030, Alphabet keeps 2027 capital expenditure rising, and local opposition continues but loses more than it wins at the state level. The upside case is that faster permitting in energy states accelerates the buildout and AI revenue — Cloud grew 82% year over year — validates the spending sooner. The downside case is two consecutive quarters of capital expenditure guidance cuts, ballot measures beginning to pass at the state level, or Google missing the late-2027 service date.
The falsifying signal is concrete: if Alphabet walks back its "significant increase" in 2027 capital expenditure at the fourth-quarter 2026 earnings call, or if Project Tembo's late-2027 service date slips by more than two quarters, the thesis that this is a structurally durable buildout shifts toward a capital-constrained cycle.
The $17 billion is not the story. The story is that Google is building its own power grid because the public one — and the public's patience — can no longer be relied upon.
Explore more exclusive insights at nextfin.ai.
