Part 2 of 6:
The Capitalist identified three threats to AI six months ago. Updated data shows all three are now worse—and compounding each other.**
High Country Advocate Investigative Team*
In September 2024, The Capitalist published an analysis identifying three existential problems facing AI infrastructure: not enough power to run it, not enough money to build it, and no viable revenue model to pay for it. The article warned these problems would compound each other.
Six months later, updated data shows the warning was prescient. All three problems have worsened. Power prices jumped another 10% in key markets. The funding gap widened to $1.5 trillion. MIT confirmed that 95% of AI projects fail to deliver measurable returns.
These aren’t three separate issues. They’re interconnected crises that reinforce each other—and no clear solutions are visible even as spending accelerates toward $5 trillion.
Problem 1: Not Enough Power
U.S. data centers consumed 176 terawatt-hours in 2023—4.4% of total electricity. By 2030, Goldman Sachs projects 426-1,050 TWh, potentially 12% of demand. The IEA forecasts global consumption will triple by 2035.
Current global capacity sits at 17 gigawatts. By 2030, that needs to reach 130 GW—130 large nuclear plants worth. Grid demand will nearly triple from 50.5 GW in 2024 to 134.4 GW by 2030.
Virginia leads with 12.1 GW in 2025, up from 9.3 GW in 2024. Texas hit 9.7 GW. Oregon exceeded 4 GW. Energy analyst Mike Silverman: “Where is that load growth coming from? The answer is data centers.”
The grid can’t keep up. Northern Virginia projects face delays. Ohio’s American Electric Power cut its pipeline from 30 GW to 13 GW after regulators ordered new tariffs. Europe’s grid connection requests “risen exponentially.”
Goldman Sachs estimates $720 billion in grid expansion needed globally by 2030. U.S. utilities forecast $1.1 trillion through 2029. But permitting delays, supply chains, and inflation slow progress.
Ratepayers absorb the costs. Residential prices jumped 6% nationally year-over-year. Virginia: 13%. Illinois: 16%. New Jersey: 20%. Maryland: $18 more monthly. Carnegie Mellon projects 8% average increases by 2030, with high-demand markets hitting 25%.
Bloomberg found wholesale costs up to 267% higher than 2020 near data centers. Over 70% of price nodes showing increases sit within 50 miles of data center activity.
The PJM Interconnection—serving 13 states plus D.C.—provides the starkest example. Capacity market costs hit $2.2 billion in 2022, $14.7 billion in 2024 (up 569%), and $16.1 billion in 2025. Watchdog Monitoring Analytics determined data center demand accounted for $9.3 billion—63% of costs. The agency concluded: “Data center load growth is the primary reason for recent and expected capacity market conditions.” These costs transfer directly to bills.
Political consequences followed. Virginia’s new governor won partly on promises to make data centers “pay their fair share.” New Jersey’s governor promised to freeze bill increases. Maryland’s People’s Counsel David Lapp warned: “If we don’t get the answer right, which is that data centers should be fully responsible for all the costs they are causing, it’s going to get much worse.”
Problem 2: Not Enough Money
Morgan Stanley calculated in July 2025 that global data center spending will reach $2.9 trillion by 2028. Internal cash flow can cover $1.4 trillion—leaving a $1.5 trillion gap.
McKinsey projects $3.7-7.9 trillion needed through 2030 depending on demand scenarios, with $5.2 trillion as the base case. Add $720 billion for grid infrastructure, and total requirements exceed $6 trillion.
Big Tech allocates $405 billion in 2025—94% of operating cash flows, up from 76% in 2024. Microsoft: $91.3 billion. Amazon: over $100 billion. Meta: $70-72 billion. Google: $85 billion.
Creative financing emerges. Meta disclosed $27 billion in off-balance-sheet financing. OpenAI raised $57.9 billion across 11 rounds, reaching a $300 billion valuation, while committing to $1.4 trillion in infrastructure deals including the $500 billion Stargate Initiative.
On November 6, 2025, OpenAI CFO Sarah Friar mentioned federal “backstop” or “guarantee” during a presentation. Trump’s AI Czar David Sacks responded: “There will be no federal bailout for AI.” Friar backpedaled. Sam Altman clarified: “We do not have or want government guarantees.” But the CFO felt the need to mention it.
Burn rates compound the problem. OpenAI lost $5 billion in 2024 on $4 billion revenue. First-half 2025: $2.5 billion cash burn, $6.7 billion R&D spending. Projected cumulative losses 2023-2028: $44 billion. Anthropic lost over $5 billion in 2024 with $2.7 billion in development spending.
Michael Burry—who predicted 2008—filed November 11, 2025 disclosures showing $1.1 billion in put options against AI companies: $187 million against Nvidia, $912 million against Palantir. His thesis: AI chips have 2-3 year lives but depreciate over 5-6 years, creating $176 billion in overstated earnings through 2028. He compared the accounting to Enron.
The circular financing pattern: Nvidia invests in OpenAI. OpenAI buys Nvidia chips. OpenAI’s valuation rises. Nvidia’s investment looks sound. Repeat. Free cash flow declines. Debt climbs. Depreciation charges could jump from $150 billion to $400 billion annually over five years.
Problem 3: No Viable Revenue Model
Current AI application revenue totals $20-30 billion in 2025. Justifying $5.2 trillion in infrastructure by 2030 requires $2 trillion—100x growth in five years while currently losing billions and facing 95% failure rates.
MIT’s July 2025 study analyzed 150 interviews, 350 surveys, and 300 public deployments. Result: 95% of generative AI projects fail to deliver measurable P&L impact. Only 5% achieve rapid acceleration. Fortune, Axios, CNBC, Harvard Business Review, and McKinsey confirmed it.
Gartner projects 30% abandoned after proof-of-concept by year-end. Harvard Business Review: 26% developed working products, 4% achieved significant returns. RAND documented 80%+ failure rates.
Failure reasons: poor data quality (85% cite this), inadequate risk controls, escalating costs, unclear value, technical debt, immature applications, data silos, inability to scale.
Customer willingness presents another obstacle. IDC found 39% believe AI should be free in existing subscriptions. Of the 61% willing to pay more, most answered “don’t know” when asked how much. Next most common: 6-10% increases acceptable. Meanwhile, 25% of SaaS buyers will replace apps without AI.
Vendors struggle. A 2025 survey: 41% struggle with cost-effective scaling, 41% struggle balancing costs with pricing, 22% struggle quantifying benefits. Some 70% raised prices in 2024, 77% changed models, but 40% of these changes failed to improve alignment.
The cost structure differs from traditional SaaS. Traditional software: 90%+ margins, near-zero marginal cost, scale equals profit. AI companies: 20-75% margins, real compute costs per user, more users mean more costs. They can’t scale profitably.
OpenAI exemplifies this. Spending $2.25 to make $1. Inference costs may exceed revenue. ChatGPT Pro subscriptions lose money. Standard subscriptions are profitable, but heavy users consume more than they pay.
Companies try various approaches: 29% bundle free, 24% charge premiums, 11% still figuring it out. Models being tested include token-based, subscription tiers, hybrid, outcome-based (rare—too hard to measure), and seat-based.
The 5-16% who succeed report 82% higher revenue, 53% higher gross profit, 136% ROI over three years. But they remain the minority while $405 billion flows annually into infrastructure.
The Compounding Effect
These three problems reinforce each other. Power constraints delay projects, raising costs and widening the funding gap. The funding gap prevents investment in efficiency, keeping power consumption high and increasing grid strain. Revenue failures reduce available cash for deployment, causing companies to miss targets and eroding investor confidence.
Infrastructure reports identify “power availability” as “the primary technological barrier” to building massive AI clusters. Former FERC Chairman Willie Phillips questions “whether or not all of the projections, if they’re real.” Microsoft CEO Satya Nadella warned about “the overbuild of AI infrastructure.” Yet OpenAI commits $1.4 trillion to infrastructure deals while burning $8.5 billion annually.
The Capitalist’s September 2024 analysis predicted these problems would compound. Six months later, the data confirms the prediction. Power crisis: worse. Money crisis: worse. Revenue crisis: worse. The question is no longer whether these problems exist—it’s whether they’re solvable at the scale and speed required to justify $5 trillion in spending.
Article 3 will follow the money to examine who profits from this buildout while application companies lose billions—and what that reveals about where this is headed.
Sources and Documentation
**Power and Grid Data:**
– Goldman Sachs: Grid infrastructure projections and data center power demand
– 451 Research/S&P Global: Grid power demand forecasts 2024-2030
– International Energy Agency: Global data center energy consumption projections
– PJM Interconnection: Capacity market auction results and analysis
– Monitoring Analytics: Independent market monitoring reports for PJM
– Carnegie Mellon University: Electricity price impact projections
– Bloomberg: Wholesale electricity price analysis (September 2025)
– State utility filings and rate case documents (Virginia, Texas, Oregon, Ohio, New Jersey, Maryland, Illinois)
**Financing and Capital Requirements:**
– Morgan Stanley: Data center financing gap analysis (July 2025)
– McKinsey: AI infrastructure spending scenarios (April 2025)
– Company earnings reports: Microsoft, Meta, Google/Alphabet, Amazon (Q4 2024, Q1 2025)
– OpenAI investor documents and funding round disclosures
– Meta 10-K and 10-Q filings: Off-balance-sheet financing
– SEC filings: Michael Burry’s 13F disclosures (November 11, 2025)
**Revenue Model and Project Success:**
– MIT: “The GenAI Divide: State of AI in Business 2025” (July 2025)
– Gartner: Generative AI project abandonment rates
– Harvard Business Review: AI implementation and ROI studies
– RAND Corporation: AI project outcome research
– IDC: Customer willingness to pay surveys
– Various industry surveys: Vendor pricing and monetization struggles (2025)
**Political and Regulatory:**
– Virginia gubernatorial campaign materials and results
– New Jersey gubernatorial campaign materials and results
– Federal Energy Regulatory Commission statements and reports
– Maryland Office of People’s Counsel testimony and filings
**Original Reference:**
– The Capitalist: “Three Problems Facing AI Infrastructure” (September 2024)
Support Independent Local Journalism — High Country Advocate was created as a real alternative to regional media that too often silences dissenting voices while taking sides in the political struggle. Producing in-depth, unflinching reporting like this series and others is expensive: servers, editing, research time, and legal review all add up quickly. If these articles have informed you or given you new perspective, please consider supporting HCA with a paid subscription — every subscriber helps keep this reporting strong and independent. 