AI Bubble Alarm: Profits Still Missing

Hand holding digital AI and ChatGPT graphics.

Wall Street’s AI gold rush is keeping money-losing startups alive long enough to chase an IPO—even as their compute bills and burn rates explode.

Quick Take

Major AI firms are posting rapid revenue growth but still reporting massive losses and heavy cash burn.
Startups are increasingly dependent on continuous fundraising and hyperscaler infrastructure deals to keep operating.
Hyperscaler capital spending is projected to exceed $500 billion in 2026, a scale that amplifies both opportunity and risk.
Investors are becoming more selective, favoring revenue-linked software/platform plays over pure infrastructure hype.

Big Valuations, Bigger Losses: The “Growth First” AI Model

OpenAI and Anthropic illustrate the defining tension of late-2025 AI markets: eye-popping revenue trajectories alongside sustained unprofitability. Research summarizing reported figures describes OpenAI losing about $5 billion on roughly $4 billion in 2024 revenue, then reaching an $18–$20 billion run-rate estimate by late 2025 while still carrying heavy losses. Anthropic’s revenue reportedly surged from a modest base to multi‑billion run rates in 2025, yet forecasts still point to significant burn ahead.

Those numbers help explain why many conservatives see echoes of past boom cycles: when the story is “growth will cover it later,” discipline can slip and accountability gets postponed. The research does not claim any formal bailout or government rescue for these startups. Instead, it points to a private-market reality where huge rounds, lofty valuations, and optimism about future dominance can sustain operations even when profits remain elusive and timelines to break-even are unclear.

Compute Commitments and Hyperscaler Dependence Raise the Stakes

AI isn’t just software; it’s an industrial-scale infrastructure bet. The research highlights multi-year compute and cloud commitments that can dwarf current earnings power, including references to massive partnerships and spending plans tied to leading labs. That dynamic strengthens the hand of hyperscalers—Microsoft, Amazon, and Alphabet—because they control the scarce resource every frontier model needs: compute at scale. For startups, that reliance can become a structural vulnerability if revenue slows or capital markets tighten.

Goldman Sachs’ analysis, as summarized in the provided research, frames 2026 as a year when AI-related investment could exceed $500 billion, with hyperscalers’ capital expenditures projected around $527 billion. That kind of spending is not a culture-war headline, but it matters to family budgets and retirement accounts: capex waves can inflate valuations, concentrate market power, and increase systemic exposure if expectations outrun reality. The research also notes that prior capex forecasts have been underestimated.

Investors Are Starting to Demand Proof, Not Promises

Even with the hype, the research describes a shift in investor behavior. After a period of broad enthusiasm—highlighted by large totals raised by AI startups in 2025—capital is portrayed as rotating toward companies that can connect AI to measurable revenue rather than raw infrastructure spending. Stock dispersion and lower correlation are cited as signs that markets are differentiating winners and losers more aggressively. In plain terms: investors may still fund AI, but they increasingly want unit economics and a believable route to profitability.

This matters because the “kept afloat” phenomenon is not magic—it’s liquidity. If investors decide the next round depends on margins, not momentum, unprofitable firms can face hard choices: cut costs, raise prices, or seek strategic deals that dilute independence. The research contrasts this environment with historical survivors that paired innovation with operational discipline. It also points to profitable counterexamples, suggesting that firms with proven cash flows and durable enterprise adoption may be better positioned if the easy-money phase ends.

What This Means for the Economy in 2026—and for Ordinary Americans

The research does not provide a single smoking-gun event showing “bailouts,” and it does not claim a crash is inevitable. It does, however, outline conditions that create real downside risk: high burn, debt or expensive commitments, dependence on constant fundraising, and profitability timelines that remain vague. For Americans still frustrated by years of inflation and fiscal mismanagement, the lesson is straightforward: when markets reward spending without discipline, someone eventually pays—often workers, shareholders, and consumers.

For policymakers under President Trump’s second term, the best guardrails are transparency and fair competition rather than politically driven picking of winners and losers. The research underscores that hyperscalers look structurally resilient because they can absorb risk with diversified cash flows, while pure-play AI labs may be more fragile if funding conditions change. Limited 2026 real-time public data is available in the provided materials, so readers should watch for audited financials, clearer profit paths, and the terms of major compute deals before treating valuations as reality.