AI Investment Bubble? Dot-Com Comparison for India 2026

Michael Burry (The Big Short) warns the AI market "feels like the last months of the dot-com bubble." In 2026, Big Tech is spending $660–725 billion annually on AI infrastructure—yet AI companies lose $17–35 billion yearly. OpenAI burns $17B (2025), projects $35B loss (2027). Startup valuations trade at 20-30x revenue vs. SaaS's 2.5-7x. Philadelphia Semiconductor Index surged 65% YTD 2026, beating dot-com peaks. Only hardware vendors (Nvidia, AMD) and cloud providers profit; everyone else building models or AI apps is unprofitable. This guide walks through the warning signs, compares dot-com's crash pattern, and shows where India fits in the coming correction.

The Math Doesn't Add Up: $660B Spent, $100B Revenue

Big Tech plans $660–725 billion in AI capex for 2026 across data centers, chips, power infrastructure. Meanwhile, global AI revenue is ~$100 billion annually. The capex-to-revenue ratio is 6.6:1. For context, that's unsustainable. No technology has ever been built at this scale with such poor unit economics.

Who's burning cash? OpenAI ($17B loss in 2025, targeting $17-35B loss in 2026-2027), Anthropic ($3B loss), Perplexity (164% of revenue spent on compute), Cursor (unprofitable despite VC hype). The only winners: Nvidia (75% gross margin on chips), AMD (recovering), cloud providers (AWS, Azure, Google Cloud capturing value by renting infrastructure to loss-making AI companies).

Profitability Delusion: OpenAI's $207 Billion Shortfall

OpenAI needs $207 billion more in capital through 2030 to fund compute for training larger models. At current revenue trajectory ($13B annualized, August 2025), OpenAI won't be profitable until 2030 at earliest—and that assumes no major new competitors burn the model economics further. HSBC report: OpenAI is unprofitable until 2030 despite 44% of world's adults using it.

The problem: scaling large language models is a capital arms race. Competitors (Meta, Google, Anthropic) are each burning $3–15B annually to stay in the race. The first to profitability might win; everyone else could collapse.

AI startup carnage (already happening)

966 AI startups failed in 2024 (up 25.6% YoY). Of the 53% of global VC funding that went to AI in H1 2025, most startups will burn through it and fail. Median time to failure: 3.2 years. The graveyard will grow through 2026-2027.

Valuation Multiples: Dot-Com Levels

In 2000, dot-com stocks traded at Nasdaq P/E over 90x. Today, AI stocks are approaching those levels:

• AI startups (median): 20–30x revenue (vs SaaS 2.5-7x baseline)
• Premier AI infrastructure: 23.4–35x revenue
• LLM vendors: Up to 78.2x EBITDA
• OpenAI: 167x projected 2025 revenue (extreme outlier)
• Nvidia (peak): 70x forward earnings

54% of global fund managers call AI stocks "bubble territory" (2025 survey). Yet prices keep rising due to momentum (Burry: "stocks up because they've been going up").

The Dot-Com Parallels

The timing rhymes. Dot-com took 5 years to build, 2.5 years to collapse. AI took 3 years (2023-2026) and shows early cracks (startup failures up 25%+, valuation resets expected). The crash could accelerate through 2026-2027.

Who Actually Profits? The "Shovel Sellers"

In a gold rush, the surest profits go to shovel sellers, not miners. In AI:

• Nvidia: FY25 revenue $215.9B (+65% YoY), 75% gross margin. Profits will keep soaring as demand for chips stays robust.
• AMD: Server CPU growth forecast 35% next 3-5 years (doubled from 18%). EPS +62% (2026) expected.
• Cloud providers (AWS, Azure, Google Cloud): Renting expensive compute to AI companies at 2-3x markup. Net profitable.
• Model developers (OpenAI, Anthropic, others): Unprofitable. Burning billions. Valuation multiples don't match fundamentals.
• AI app builders (Perplexity, Cursor, specialized tools): 90% will fail. Survivors might hit profitability by 2030.

Lesson: In 2026, own semiconductor, cloud, and infrastructure plays. Avoid unprofitable model developers and AI apps unless they're diversifying into services (rare).

India's AI Boom vs Global Correction Risk

Google, Microsoft, AWS committing $15–17 billion to India's AI infrastructure (data centers, skilling, operations). India's data center capacity: 1.93 GW (2025) → 4 GW target (2028). Government incentives: 100% tax holiday for foreign cloud companies using Indian data centers (until 2047).

India's advantage: structural, long-term. If global AI correction happens (likely), India's data center investments continue because they're part of broader supply chain moves away from China. But Indian startups building AI models or apps will face the same profitability squeeze as global players.

For India investors: Bet on infrastructure (data centers, telecom, power) not AI apps. The correction will be severe for unprofitable AI startups; cloud & infrastructure providers will emerge stronger.

When Will the Crash Come?

Michael Burry's timing has been wrong before (2021 crash prediction didn't materialize). But the setup is there: 54% of fund managers see a bubble, profitability is deteriorating, and startup failure rates are accelerating. Likely timeline: 2026-2027.

Catalyst: A major model developer fails or raises at lower valuation. Enterprise customers realize ROI is weaker than promised. Startups' funding dries up as LPs demand profitability. Valuations compress 50-70% over 18-24 months.

The safest play: Hardware & infrastructure

Nvidia, AMD, and cloud providers will survive and thrive even if AI model economics crack. Own these. Avoid unprofitable AI startups unless you see a clear path to $10M+ revenue and positive unit economics.

The AI boom is real. The infrastructure being built is real. But the economics being sold are fantasy. 2026 is the year reality meets hype. Prepare accordingly.

For cautious AI investment strategies, see our investment advisory service.