Why Building AI Data Centres Isn’t Working Anymore

The video argues that the $650 billion being poured into AI data centers in 2026—part of a $9 trillion planned buildout—has become a speculative bubble with little chance of turning a profit, as nearly half of U.S. projects are delayed or cancelled while companies like Microsoft fail to break ground. It highlights that the economy's apparent growth is dangerously dependent on this spending, with 92% of GDP growth tied to AI infrastructure while the rest grew only 0.1%. Beyond the shaky financials, the boom has created severe local backlash: communities face skyrocketing electricity bills, infrasound noise, polluted water, and environmental destruction for relatively few jobs, leading to project cancellations and even an FBI warning about anti-AI extremism. Ultimately, the episode concludes that the money invested may never be recovered, and the trillion-dollar data center buildout could end up being a historic waste of resources.

• ▶ 0:00 The episode opens with its central claim: four major tech companies are expected to spend $650 billion on new data centers in 2026 alone.
• ▶ 0:11 To put that in perspective, $650 billion in $100 bills would form a stack 710 km high—more than 300 km above the International Space Station's orbit.
• ▶ 0:25 Despite a $9 trillion planned investment by 2030, nearly half of U.S. data centers scheduled to open this year have been delayed or cancelled.
• ▶ 0:38 Bloomberg reports roughly 140 U.S. projects (12 GW) were planned, but only a third are actually being built; the rest exist only on paper.
• ▶ 1:12 Microsoft's case shows a wider pattern: none of its data centers announced since 2023 have broken ground, and even "fully complete" sites like Fairwater are less than half built per satellite imagery.
• ▶ 1:31 Public resentment toward data centers is rising due to infrasound noise, polluted water, higher energy costs, and other problems, with people saying they have "had enough."
• ▶ 1:44 The cost of data centers falls on ordinary people, homeowners, families, and communities living near them, not on shareholders or companies receiving tax breaks—and these residents never got a say.
• ▶ 1:59 The central question emerges: if massive sums are being spent on AI infrastructure, why is it falling apart before it is even built, hinting at a possible collapse or fundamental flaw.
• [2:21 – 2:25] The discussion reframes the debate from AI spending itself to the financial returns needed to justify it.
• [2:26 – 2:28] The guest bluntly concludes the money will never be recovered: “They'll never get it.”
• [2:29 – 2:43] Trillion-dollar data center buildouts are described as throwing money away, with some firms spending more cash than they have.
• ▶ 3:06 The US economy has become dangerously dependent on AI data center spending, with an estimated 92% of GDP growth coming from it; excluding AI infrastructure, the rest of the economy grew just 0.1%.
• ▶ 3:24 Speculative excess is rampant—shoe company Allbirds pivoted to leasing AI data center equipment despite making no sense, yet its stock soared 580%.
• ▶ 3:57 More money has been committed to data centers in six years than the combined cost of the Marshall Plan, Manhattan Project, Apollo program, and ISS—plus $120 billion extra—all for a business plan that isn't solid yet, and GPUs rapidly become outdated.
• ▶ 4:26 AI's long-term potential isn't dismissed, but profitability within the timeframe needed to prevent funding collapse is "looking less likely."
• ▶ 4:36 There is a real, plausible risk that the entire massive spending effort could turn out to be a "huge waste of money."
• ▶ 4:47 While Claude has a clear niche (e.g., programming), most other AI models lack a distinct, profitable purpose—raising doubt about their massive investments.
• ▶ 4:54 At the boom's peak, more than two new data center facilities were built every week, with states like Virginia, Texas, Georgia, and Arizona aggressively courting developers.
• ▶ 5:05 Massive tax incentives fueled the boom: Texas gave over $1 billion for the Stargate project, and Virginia granted nearly $1 billion in tax savings across 56 projects in one year.
• ▶ 5:38 The promised benefits didn't match reality: property tax abatements left communities absorbing infrastructure costs, Oregon schools lost $275 million, and even the largest data centers employ fewer than 150 permanent workers.
• ▶ 6:22 Revolut is pitched as an all-in-one finance app for simplifying spending, used by over 75 million people.
• ▶ 6:50 Key features include single-use virtual cards for online payment security, money management, and international use across currencies.
• ▶ 7:28 The segment offers a $40 sign-up incentive, customizable cards with Apple Pay/Google Pay support, and a premium metal plan option.
• ▶ 7:46 Hyperscalers are now effectively net debtors after exhausting cash reserves, and investors are pulling back from AI amid fears there's no real profit to be made — making the bubble's burst seem "utterly inevitable."
• ▶ 8:31 Power and supply chain bottlenecks are crippling data center buildout: a single facility can use as much electricity as a city of 200,000 homes, while critical components like transformers are in short supply and largely imported from China, so one delayed part can stall an entire project.
• ▶ 10:12 The true environmental and community costs are massive: residents already breathe exhaust from roughly 10,000 diesel generators in Virginia, a Utah project would expel heat equivalent to 23 atomic bombs daily, and Meta's Louisiana data center — nearly one-fifth the size of Manhattan and using 5GW — is consuming irreplaceable farmland for only about 100–500 jobs.
• ▶ 13:03 Communities near data centres in Virginia, Georgia, and Texas report low-frequency hums, sleep disruption, $200,000 spent on insulation with no relief, and sediment appearing in tap water.
• ▶ 13:41 Residents face massive electricity bill hikes tied to AI data centres: Georgia Power raised rates six times (24% jump) from 2023–2025, and Oregon's Pacific Power saw a 50% increase since 2020.
• ▶ 14:36 The DRAM/RAM market is in chaos: AI data centres may absorb 70% of global DRAM production in 2026, a 64 GB DDR5 kit surged from $190 to $700 in three months, and Micron's stock dropped 22% after Sam Altman's non-binding purchase promise collapsed.
• ▶ 15:13 Local backlash against AI data centers quadrupled in 2025, with at least 25 projects cancelled after community opposition (up from just 6 in 2024).
• ▶ 15:35 The FBI has classified anti-AI sentiment as an "emerging terrorist threat," warning of potential anti-tech violent extremism within five years.
• ▶ 16:39 Opposition has become a "kitchen table issue": Maine passed a statewide data center ban, 13 states are considering similar measures, and 65% of Americans now oppose data centers in their communities.
• ▶ 18:27 Microsoft and Oracle/OpenAI have slowed or stalled major data center projects, with analysts citing oversupply relative to demand forecasts.
• ▶ 19:11 The "bubble" is a reality check: infrastructure spending assumes exponential AI demand, but returns are speculative—especially as free open-source models undercut the value of $1 trillion frontier-model training.
• ▶ 20:35 Red flags in financing: A-rated data center bonds are paying 8–12% interest (junk-level), indicating risk credit ratings haven't acknowledged, with exposure on private investors rather than banks.
• ▶ 21:08 None of the core assumptions behind the AI data-center boom have held up.
• ▶ 22:01 Real-world project collapses: Project Matador, named after President Trump, lost its CEO and CFO, had no anchor tenant, and saw its market cap fall from ~$20B to $3.4B.
• ▶ 23:01 Compute constraints may force cancellations, community bans, and supply-chain issues, potentially throttling AI model performance over time.

• ▶ 23:28 Subsea data centers offer a sustainable alternative, with one operator reporting 99% of electricity powering computing versus ~50% in traditional air-cooled facilities.

• ▶ 24:03 Local AI models and commoditized LLMs could drastically cut demand for centralized infrastructure; within three years, most laptops running decent models may make massive data centers less necessary.

• ▶ 25:23 The host's stance: AI is revolutionary for legitimate uses, but building massive AI infrastructure is not a great trade-off if it raises electricity costs for communities just to generate memes or "slop" — the real question is how to build and whether it's worth the payoff.

• ▶ 0:00 The Promised Boom and Early Doubts (0:00 - 7:46) - Sets up the AI data center spending boom, early profitability concerns, and a sponsor break.
• ▶ 7:46 Debt, Power, and Environmental Costs (7:46 - 13:00) - Explains hyperscaler debt, power and supply chain bottlenecks, and the water/environmental damage from massive builds.
• ▶ 13:00 Grievances, Chaos, and Cynicism (13:00 - 15:16) - Covers community grievances, cynical AI quotes, and DRAM/RAM market chaos.
• ▶ 15:16 Backlash and Global Opposition (15:16 - 18:27) - Describes grassroots backlash, FBI warnings, and controversial data center projects in Illinois and Australia.
• ▶ 18:27 Pullbacks and the Bubble Popping (18:27 - 21:11) - Details Microsoft/Oracle/OpenAI pullbacks and why the AI data center bubble is only partially popping.
• ▶ 21:11 Satire and Project Matador (21:11 - 23:28) - Uses satirical examples and the Project Matador story to illustrate the failed buildout.
• ▶ 23:28 Alternatives, Commoditization, and Closing Stance (23:28 - 27:16) - Offers sustainable local-AI alternatives, LLM commoditization, and the host's final view on the trade-off.