The AI Money Pit: How Much Are OpenAI, Anthropic, and xAI Actually Losing?
A data-driven look at how much the major AI companies are spending, losing, and projecting — and how their financials compare to the dot-com era companies that either survived or went to zero.
The AI industry is burning money at a pace that makes the dot-com era look quaint. OpenAI is projecting $14 billion in losses for 2026. xAI is burning $1 billion a month. The four biggest hyperscalers are planning to spend $470 billion on AI infrastructure this year alone.
But underneath the staggering numbers, the real question is simple: does the math work? I dug into the financials of every major AI player and compared their trajectories to the companies that lived and died during the dot-com bubble. Here’s what the numbers actually say.
The Current Burn Rates
Before diving into the numbers, it’s worth clarifying what “burn rate” actually means — the term gets thrown around loosely in tech press and it refers to a few different things.
Gross burn rate is total cash going out the door per period, regardless of revenue. When Bloomberg reports that xAI is “burning $1 billion a month,” that’s gross burn — salaries, compute costs, data center leases, everything combined.
Net burn rate is cash out minus cash in — the rate at which a company is actually draining its bank account. If xAI spends $1B/month but brings in ~$42M/month in revenue, the net burn is ~$958M/month.
Burn rate as a percentage of revenue is the metric I use throughout this post to compare companies against each other. It’s calculated as (net cash loss / revenue) × 100. When I say OpenAI’s burn rate is “70% of revenue,” it means for every $1 they earn, they lose an additional $0.70 — they’re spending $1.70 for every $1 coming in.
This is the number that matters most for comparing companies at different scales, because it tells you whether revenue is growing faster than costs (burn percentage shrinking) or costs are outpacing revenue (burn percentage growing). Here’s how the three major AI startups compare:
| Company | Revenue | Net Loss | Burn as % of Revenue |
|---|---|---|---|
| OpenAI (2025) | ~$13B | ~$9B | ~70% |
| OpenAI (2026-27) | growing | growing | ~57% |
| OpenAI (2028) | growing | accelerating | ~75% |
| Anthropic (2026) | ~$26B | ~$8.6B | ~33% |
| Anthropic (2027) | growing | shrinking | ~9% |
| xAI (2025) | ~$0.5B | ~$13B | ~2,600% |
One last term: runway — how many months of cash a company has at its current net burn before hitting zero. This is the number that determines life or death. It’s what killed Pets.com in 2000 (less than one quarter of runway) and what keeps OpenAI dependent on continuous fundraising today.
OpenAI
OpenAI is the biggest name and the biggest spender. Their current financials paint an aggressive picture:
| Metric | Number |
|---|---|
| 2025 Revenue | ~$13B |
| 2026 Projected Loss | $14B |
| Cash Burn Rate | ~70% of revenue |
| Cumulative Losses (2023-2028) | $44B |
| Total Planned Spend (through 2029) | $200B |
The company’s internal documents project $100 billion in annual revenue by 2029 — roughly matching what Nvidia makes today with a near-monopoly on AI hardware. To get there, they need to 8x revenue in four years while losses are simultaneously tripling.
An HSBC analysis projects that OpenAI still won’t be profitable by 2030, even if ChatGPT’s user base grows to 44% of the world’s adult population. They estimate a $207 billion funding shortfall just to keep the lights on.
Making matters worse, ChatGPT’s market share is already eroding — web traffic share fell from 86.7% to 64.5% in a single year, with Google Gemini capturing most of the difference. Harder to hit $100B when your moat is shrinking.
Anthropic
Anthropic tells a meaningfully different story:
| Metric | Number |
|---|---|
| 2025 Revenue | ~$10B |
| Current Annualized Run Rate | $14B |
| 2026 Revenue Target | $26B |
| 2028 Revenue Target | $70B |
| Cash Flow Positive | 2027 (projected) |
| 2028 Projected Cash Flow | $17B |
The key differentiator is efficiency. Anthropic’s burn rate is projected to drop to ~33% of revenue in 2026 and just 9% by 2027. Compare that to OpenAI’s 57% burn rate holding steady through 2027, then spiking to 75% in 2028.
Anthropic just closed a $30 billion Series G at a $380 billion valuation. Around 80% of their revenue comes from enterprise customers — stickier than consumer subscriptions. Claude Code alone is doing $2.5 billion in annualized revenue.
xAI
Elon Musk’s xAI has the most alarming ratio of spending to revenue:
| Metric | Number |
|---|---|
| Standalone Revenue | ~$500M annualized |
| 2025 Projected Loss | $13B |
| Monthly Burn Rate | $1B |
| Burn in First 9 Months of 2025 | $9.5B |
| Total Funding Raised | $22B equity + $5B debt |
For every dollar xAI earns, it loses roughly $26. They’re targeting profitability by 2027, but the gap between revenue and spend is by far the widest of any major AI company.
The Hyperscalers
Google, Meta, Microsoft, and Amazon are in a different position — they’re already profitable from existing businesses and are pouring cash into AI as an investment.
But the scale of that investment is eye-watering:
| Company | 2025 Capex | 2026 Planned Capex | Increase |
|---|---|---|---|
| Google/Alphabet | $91.4B | $175-185B | ~100% |
| Meta | ~$72B | $115-135B | ~73% |
| Combined Big 4 | ~$350B | $470B+ | ~34% |
Google is spending nearly half its revenue on capex. Meta is doubling down despite not having a cloud business to monetize the infrastructure. Total AI spending across the industry is expected to surpass $1.6 trillion through 2029.
The $2 Trillion Question
Here’s where the profitability math gets uncomfortable.
The industry needs to generate $2 trillion in annual AI revenue by 2030 to justify current capex levels. Current AI revenues across all companies sit at roughly $20 billion. That’s a 100x gap in four years.
Even with generous 25% gross margins, break-even on infrastructure alone requires $160 billion per year in AI revenue. Solid returns require $400-500 billion.
Big Tech free cash flow could drop up to 90% in 2026 as spending outpaces revenue growth. That’s not a typo — ninety percent.
The Dot-Com Comparison
This Is Bigger
One researcher called the AI bubble “17 times larger than the dot-com bubble.” The share of the economy devoted to AI investment is already a third greater than what was devoted to internet investment at the dot-com peak. The Magnificent 7 now make up over 30% of the S&P 500 — exceeding dot-com era market concentration.
Morgan Stanley analysts describe the capex frenzy as “eerily similar to the telecom bubble,” with each of Meta, Microsoft, and Alphabet spending 21-35% of revenue on capex — more than AT&T at the height of the telecom bubble.
Amazon vs. Pets.com: The Lesson That Matters
The most instructive dot-com comparison isn’t about whether AI is “real” — it obviously is. The question is which companies survive the correction and which don’t. The Amazon vs. Pets.com story tells you exactly what to look for.
| Amazon (survived) | Pets.com (died) | |
|---|---|---|
| Peak burn | $320M/quarter, throttled to $70M when crisis hit | $83.5M/quarter, accelerating |
| Cash reserves | $700M working capital, ~2 years runway | $23M cash, < 1 quarter runway |
| Unit economics | Path to positive margins per sale | Lost $0.26 on every $1 in sales before marketing |
| Time to first profit | ~7 years (Q4 2001) — $0.01/share | Never |
| Stock during crash | Lost 90%, took 10 years to recover | Went to zero |
Amazon survived because it could throttle spending when conditions deteriorated, had enough cash to weather 2+ years of downturn, and — critically — had improving unit economics underneath the losses. Each sale was getting closer to profitable. Pets.com’s core business lost money at every scale.
It took Amazon 14 years from its IPO to accumulate as much total profit as it eventually made in a single quarter. The long game worked — but only because the underlying economics were sound.
Which AI Companies Are the Amazons?
The Amazon/Pets.com lens maps surprisingly well onto today’s AI landscape.
Anthropic looks most like Amazon:
- Enterprise-heavy revenue (80% B2B) that’s sticky and high-value
- Declining burn rate as a percentage of revenue — the unit economics are improving
- Cash flow positive by 2027, not dependent on moonshot revenue targets
- $30B in fresh funding provides 2+ years of runway at current burn
- Can likely throttle spending if conditions deteriorate
OpenAI carries more Pets.com risk factors:
- Burn rate increasing as a percentage of revenue (57% through 2027, spiking to 75% in 2028)
- Revenue needs to 8x in 4 years while competition intensifies
- Could run out of cash by mid-2027 without more fundraising
- Consumer market share already contracting
- Key talent departing to competitors
xAI is the most exposed:
- $500M standalone revenue against $13B in annual losses — a 26:1 loss-to-revenue ratio
- Revenue is dwarfed by burn rate with no clear path to narrowing the gap
- Entirely dependent on Musk’s ability to keep raising capital
But This Isn’t Exactly the Dot-Com Bubble Either
The bull case for why this is different: unlike the late-90s internet companies, today’s AI players have real revenue, real products, and the hyperscalers backing them are already profitable businesses that can absorb losses. The dot-com crash killed companies with no revenue at all. AI companies have revenue — just not enough relative to spending.
Today’s AI companies also have lower debt-to-earnings ratios than their dot-com equivalents, and AI is already generating measurable productivity gains in enterprises. This isn’t Webvan delivering groceries at a loss.
The bear case: the capex burden is so massive it could crater Big Tech free cash flow by 90%, AI investment as a share of GDP already exceeds dot-com levels, and nobody has demonstrated that inference economics actually work at the scale being bet on. The S&P 500 CAPE ratio now exceeds 40, matching dot-com valuations exactly.
Key Takeaways
-
The raw numbers are staggering. OpenAI projecting $44B in cumulative losses through 2028. xAI burning $1B/month. Hyperscalers planning $470B+ in capex this year. The industry needing $2T in annual revenue by 2030 to justify it all.
-
Not all AI companies are created equal. Anthropic’s path to profitability (2027, declining burn rate, enterprise revenue) looks fundamentally different from OpenAI’s (losses accelerating through 2028, consumer share eroding) or xAI’s (26:1 loss-to-revenue ratio).
-
The dot-com parallel is instructive but imperfect. The technology is real and generating revenue — this isn’t 1999. But the spending levels exceed dot-com peaks by an order of magnitude, and the gap between capex and AI revenue is a $2 trillion problem.
-
Cash reserves and throttle-ability determine survival. Amazon survived the dot-com crash because it could cut spending and had 2 years of runway. Pets.com had one quarter. When (not if) conditions tighten, the AI companies that survive will be the ones that can slow their burn without collapsing.
-
The honest answer is nobody knows yet. The entire industry is betting that AI revenue scales faster than compute costs. If inference costs keep dropping and enterprise adoption accelerates, the optimistic projections become plausible. If costs plateau or competition commoditizes pricing, the $2 trillion gap becomes unbridgeable.
The dot-com era proved that transformative technology can be simultaneously real and overhyped as an investment. The internet did transform everything — but Amazon still lost 90% of its value along the way, and it took a decade to recover. The AI winners are probably in this group somewhere. The question is whether you can tell them apart from the Pets.coms before the correction hits.
Resources
- OpenAI Financial Projections (Fortune)
- HSBC Analysis: OpenAI $207B Shortfall
- Anthropic $30B Series G (CNBC)
- Anthropic $70B Revenue Projection (TechCrunch)
- xAI $1B/Month Burn Rate (Bloomberg)
- Big Tech Needs $2T in AI Revenue (Where’s Your Ed At)
- AI Capex Boom Analysis (Sparkline Capital)
- Big Tech AI Capex and Cash Impact (CNBC)
- AI Bubble vs Dot-Com (Wikipedia)
- Amazon Dot-Com Survival Story (CNBC)
Published: February 2026