Why ARR Growth Without Profitability—and an Infrastructure Crisis—Signal a Reckoning Ahead

Executive Summary

The artificial intelligence sector is experiencing a classic speculative bubble, and the warning signs are unmistakable. While global AI investment has doubled to $275 billion and companies report breathtaking Annual Recurring Revenue (ARR) growth, a critical examination reveals a troubling reality: AI companies are growing revenue rapidly while simultaneously hemorrhaging cash, operating with structurally inferior gross margins, and avoiding the transparency that public market scrutiny would demand.

Compounding these application-layer problems is a brewing infrastructure crisis. According to the International Energy Agency, global electricity demand from data centers could reach 945 TWh by 2030—more than Japan’s total electricity consumption today. Bloomberg’s analysis reveals that wholesale electricity costs have risen as much as 267% in areas near data centers since 2020, with those costs being passed directly to American households. The AI boom is building on a foundation of strained power grids, seven-year construction timelines, and uncertain demand forecasts that could leave billions in stranded assets.

1. The Gross Margin Problem: AI’s Fundamental Economic Flaw

Behind the impressive ARR headlines lies a structural economic problem that threatens the entire AI investment thesis. According to Bessemer Venture Partners’ State of AI 2025 report, the fastest-growing AI startups—what they call “Supernovas”—operate on average with only 25% gross margins. Compare this to traditional SaaS companies, which routinely achieve 70-85% gross margins.

This isn’t a minor difference—it’s a fundamental economic divergence. As TechCrunch has reported, “every AI startup that is stuck reporting gross margins in the 50s and low 60s will find itself in the basement of SaaS company lists.” And 25% gross margins? That’s not even on the same chart. These companies are generating less cash flow than a pure SaaS startup of similar scale and should be valued far more conservatively.

The root cause is structural: AI companies face compute costs that scale with usage. Every customer query, every inference call, costs real money. Unlike traditional software where marginal costs approach zero at scale, AI companies face GPU bills that can grow faster than revenue. This is not a problem that “will be solved” by falling compute costs—it’s baked into the architecture of how these products work.

2. ARR Without EBITDA: The Revenue Quality Crisis

The AI sector’s obsession with ARR growth obscures a more important question: where is the profitability? OpenAI, the poster child of the AI boom, lost $5 billion in 2024 despite generating $3.7 billion in revenue. The company isn’t having trouble growing revenue—it’s having trouble making money. Microsoft reported a quarterly loss of roughly $4 billion attributable to its share of OpenAI’s losses.

As Marina Davidova of DVC has acknowledged, “The last two years have been wild. We see some of the fastest-growing companies in history, but it’s never been harder for VCs to see through the noise of bloated numbers.” The admission that investors themselves struggle to evaluate these companies should concern anyone considering AI investments.

Mikael Johnsson, General Partner at Oxx, puts it bluntly: “For the majority of AI-native companies, ARR isn’t really ARR as we know it, but rather a hotchpotch of one-off, credits-based, performance-based or outcome-based revenue contracts.” This revenue quality problem means that even the impressive ARR figures being reported don’t represent the stable, predictable income streams that traditionally define valuable subscription businesses.

3. The Missing IPOs: Why AI’s Best Companies Won’t Go Public

As Bloomberg has noted, “Artificial intelligence meets the pre-conditions for a tech bubble better than any technological innovation cycle in the last quarter-century—except for one: a publicly-traded pure play that primarily operates in the new innovative space.” None of the main protagonists driving the AI euphoria are publicly traded AI enterprises.

This is not a coincidence—it’s a strategy. Sam Altman himself has said he’s “0% excited” about running a public company. Why? Because going public means answering hard questions about unit economics, profitability timelines, and sustainable business models. It means quarterly scrutiny from analysts and regulators demanding answers about the path to profitability.

The fact that AI startups can raise virtually unlimited private capital allows them to avoid this accountability. OpenAI raised $40 billion—the largest amount ever raised by a private tech company—precisely because it could. But this creates an information asymmetry that historically precedes market corrections. Without public financial reporting, investors are operating on projections and promises rather than audited results.

4. The Power Grid Crisis: AI’s Insatiable Energy Appetite

The AI boom is placing unprecedented strain on America’s electrical infrastructure. According to Pew Research Center, a typical AI-focused hyperscaler annually consumes as much electricity as 100,000 households. The larger facilities currently under construction are expected to use 20 times as much. In 2023, data centers consumed 26% of Virginia’s total electricity supply, with significant shares in North Dakota (15%), Nebraska (12%), Iowa (11%), and Oregon (11%).

The costs are being passed directly to American consumers. Bloomberg’s investigation found that wholesale electricity costs have risen as much as 267% since 2020 in areas near data centers. In the PJM electricity market stretching from Illinois to North Carolina, data centers accounted for an estimated $9.3 billion price increase in the 2025-26 capacity market. The average residential bill is expected to rise by $18 a month in western Maryland and $16 a month in Ohio. A Carnegie Mellon study estimates data centers could lead to an 8% increase in the average U.S. electricity bill by 2030—potentially exceeding 25% in northern Virginia.

David Crane, CEO of Generate Capital and former Biden administration energy official, warns: “Without mitigation, the data centers sucking up all the load is going to make things really expensive for the rest of Americans.” He warns of potential brownout situations in some U.S. power markets within the next year or two.

5. The Infrastructure Construction Bottleneck

According to BloombergNEF, U.S. data center power demand will more than double by 2035, rising from 35 gigawatts to 78 gigawatts. But the report’s title says it all: “Power for AI: Easier Said Than Built.” BNEF estimates that data center development typically takes about seven years from initial steps to full operation—4.8 years pre-construction and 2.4 years for construction itself.

The bottlenecks are severe. In hotspots like Virginia, grid connection timelines have stretched from a few years to as long as seven years. Four factors compound the slowdown: technical complexity of resilient, high-capacity feeds; upstream grid capacity shortfalls; lengthy lead times for critical electrical equipment; and slow, inconsistent permitting complicated by zoning gaps and local opposition.

CNBC reports that utilities are grappling with a fundamental question: how much of the projected AI data center demand is actually real? Tech companies are shopping the same big projects to multiple utilities looking for quickest power access, making demand forecasting nearly impossible. FERC Chairman David Rosner warned that “the difference of a few percentage points in electricity load forecasts can impact billions of dollars in investments and customer bills.” Utilities spent $178 billion on grid upgrades last year and are forecasting $1.1 trillion in capital investments through 2029—investments that could become stranded assets if demand projections prove inflated.

6. The AI Infrastructure Bubble: Echoes of Telecom 2000

The problems at the application layer are compounded by a parallel bubble in AI infrastructure investment. In a single week in late 2024, Alphabet announced a $40 billion plan for AI infrastructure while Anthropic committed $50 billion for new data centers. Private equity firms, infrastructure funds, and sovereign wealth pools are pouring hundreds of billions into data centers based on exponential growth projections.

The parallels to the late-1990s telecom bubble are striking: massive debt-financed infrastructure buildout based on exponential growth projections; temporal mismatches between fund lifecycles and asset maturity; and slowing enterprise monetization of AI capabilities. According to industry analysis, the majority of private equity investors expect to exit their data center investments within a 5-10 year timeframe—but these assets typically require 10-15 years to generate optimal returns.

The IEA’s Energy and AI report notes that while data center electricity demand growth may account for less than 10% of global electricity demand growth, “data centres, unlike electric vehicles, tend to concentrate in specific locations, making their integration into the grid potentially more challenging.” This geographic concentration amplifies both the infrastructure strain and the risks of overbuilding in specific regions.

7. The Churn Problem: Enterprise Pilots Aren’t Real Customers

According to ChartMogul’s retention research, AI-native companies have even worse gross retention rates (40%) than B2C SaaS companies and comparable net revenue retention (48%) despite higher price points. This “AI churn wave” means that hypergrowth becomes nearly impossible to sustain—companies are constantly replacing churned customers just to maintain their base.

As Paul Anthony, Founder of OpStart, observes, companies face pressure to “change some of their pricing and contract terms in ways that incentivize near-term signups even if they don’t stick around.” Johnsson predicts “a high casualty rate” for AI startups: “A huge portion of companies are significantly overvalued, and it’s going to be a lot harder for them to get funding if they can’t translate early momentum into long-term business.”

8. The Valuation Math Doesn’t Work

Let’s examine the basic math. If SaaS valuations have retreated to 7x ARR (down from 20x in 2021), and those companies have 80%+ gross margins, what multiple should apply to AI companies with 25-50% gross margins? Certainly not the 50-70x ARR multiples that investors are currently paying for companies like Cursor and Lovable.

The median AI startup operates at 25-35% gross margins—terrible compared to traditional SaaS. Only about 0.7% of all AI startups have reached unicorn status. Among AI unicorns, only a handful are actually profitable on a net income basis. OpenAI would need to increase its revenue to $577 billion by 2029 to deliver on its current commitments—a 2,785% jump in four years. This isn’t a business plan; it’s a prayer.

9. Conclusion: Prepare for the Correction

The AI bubble will pop—not because the technology isn’t real, but because the economics don’t support current valuations. At the application layer, companies are burning cash while reporting impressive but low-quality ARR. At the infrastructure layer, power grids are straining, construction timelines stretch to seven years, and utilities face billions in potentially stranded investments. The entire edifice is built on demand projections that even industry insiders question.

As Paul Anthony warns, “When the bar rises, it will be tough. Due diligence gets deeper, and you could lose funding if you’re not showing long-term traction. It’s once the buzz wears off that you find out if there really is a viable long-term business.”

The smart money is beginning to recognize these risks. Don Butler of Thomvest Ventures notes that “a new set of metrics” focused on actual value creation—cost savings, efficiency gains, revenue growth—will replace the ARR obsession. But that transition will be painful for companies that have raised at valuations predicated on infinite growth rather than sustainable economics.

For investors and founders, the message is clear: focus on unit economics, not headline ARR. Demand transparency on gross margins and path to profitability. Be skeptical of infrastructure investments predicated on exponential demand growth. And remember the most dangerous phrase in investing: “This time is different.”

It isn’t.

—Sources: International Energy Agency, Bloomberg, Pew Research Center, BloombergNEF, CNBC, Bessemer Venture Partners, DevelopmentCorporate