$700B a Year -- Who Will Be the Next Motorola?

Alphabet just issued a century bond.

Yes, 100 years. The last time a tech company did this was Motorola in 1997 – which also happened to be the last year Motorola was considered a “big company.” Michael Burry immediately posted a warning on X, implying Google might be repeating history.

But rather than debating one bond, I’m more interested in the question everyone is avoiding: from ChatGPT’s launch to now, how much has the world actually spent on AI? And how much has it earned back?

I pulled the data together and ran a full accounting. The results aren’t pretty.

A Staggering Ledger

Start with the investment side.

When ChatGPT launched in late 2022, global hyperscaler (Amazon, Microsoft, Google, Meta, Oracle) annual capex totaled about $157 billion, with AI-related spending around $47 billion. By 2025, the total soared to $443 billion, with AI’s share at roughly $332 billion – a 7x increase.

Goldman Sachs’ tally puts total hyperscaler capex for 2022-2024 at $477 billion, with 2025-2027 projected at $1.15 trillion – more than double. But that was a late-2025 forecast. The latest numbers are even more staggering.

Just last week (February 6), the four major hyperscalers announced 2026 capex guidance: Amazon $200B, Alphabet $175-185B, Microsoft ~$145B (annualized), Meta $115-135B. Those four alone total $635-665B. Add Oracle’s $50B, and the five giants’ 2026 capex will reach roughly $700 billion – a 58% surge from 2025’s $443B. All four major players crossed the $100B mark for the first time.

Layer on VC investment in AI startups ($200B flowed into AI in 2025 alone), and the total gets even more staggering.

Add it all up: from ChatGPT’s launch through the end of 2025, cumulative global AI infrastructure investment is approximately $650 billion. And 2026 alone will add another $500B+.

Now the revenue side.

Generative AI direct market revenue in 2025 is somewhere between $60-130B (depending on whether you count only GenAI software or include the incremental portion of cloud AI services). Cumulatively from 2022 to 2025, direct AI revenue totals approximately $240 billion.

Do the math: for every $1 invested, only $0.36 has been earned back so far.

But the Growth Rate Gap Is What Matters

If you only look at absolute numbers, this looks like a disaster. But look at growth rates and the picture changes completely.

AI capex growth moderated from 2025’s 73%, but the absolute number surged again in 2026 – the five giants’ capex jumped from $443B to $700B, a 58% increase. Meanwhile, AI revenue growth has consistently stayed at 80-100%. OpenAI is the clearest example: ARR went from $2B in 2023, to $6B in 2024, to $20B in 2025 – tripling every year. Anthropic also rocketed from under $100M in early 2024 to $7B in 2025.

Investment is still accelerating in absolute terms, but revenue is growing faster. The “scissors” are closing, just later than previously expected because 2026 capex came in above forecasts.

So When Does It Break Even?

I modeled three scenarios:

Optimistic (late 2029 - early 2030): If AI revenue sustains 50%+ annual growth, cumulative revenue catches cumulative investment around late 2029. This aligns with OpenAI’s CFO saying “cash flow positive by 2029.” Goldman Sachs projects AI cloud revenue could reach $200-300B/year by 2030; if that materializes, industry-wide breakeven is plausible. Note: because 2026 capex surged to $700B above expectations, the breakeven timeline has shifted roughly six months to a year later than prior estimates.

Base case (2030-2031): Revenue growth decelerates to 35-40%, pushing breakeven to 2031.

Pessimistic (2032+): If AI commercialization hits a wall – enterprises find AI tool ROI below expectations, regulations tighten, or an “AI winter” sets in – massive capex becomes sunk cost and breakeven recedes indefinitely. Given that Amazon’s free cash flow is projected to turn negative in 2026 (-$17B to -$28B) and Meta’s FCF will drop 90%, the cost of this scenario would be severe.

One data point keeps me cautious: only 25% of enterprise AI projects have achieved their expected ROI, and AI service revenue accounts for only about 10% of hyperscaler capex. This tells you there’s still a massive gap between “built it” and “adopted it.”

OpenAI: A Mirror for the Industry

Using OpenAI as a microcosm of the entire industry makes things more concrete.

Revenue growth has been nothing short of epic: $2B in 2023, $6B in 2024, over $20B ARR in 2025. Tripling every year. ChatGPT’s weekly active users surpassed 700 million. Enterprise customers exceeded 3 million.

But profits? A $5B loss in 2024, still losing money in 2025. Cumulative losses are projected to reach $44B by 2028. Even the most optimistic estimate puts positive cash flow at 2029.

A company with $20B in annual revenue growing at 300% is still deeply in the red. That’s the portrait of the AI industry today.

How Is This Different from the Last Bubble?

Whenever you see burn rates like this, the instinct is to think of the 2000 dot-com bubble. But several key differences exist.

Goldman Sachs notes that AI capex currently represents 0.8% of GDP, well below the 1.5% of the 1990s telecom bubble. More importantly, AI has real enterprise adoption – 78% of surveyed enterprises are using AI, 71% are using generative AI. This is fundamentally different from the “land grab first, monetize later” internet bubble.

But this doesn’t mean there’s no risk. BofA data shows hyperscaler capex as a share of operating cash flow has risen to 94%, approaching the limit. They’ve started issuing debt at scale – $108B in AI-related bonds in 2025, with J.P. Morgan estimating $1.5 trillion in investment-grade bonds needed over the coming years to support AI data center construction. Even more striking: at 2026’s $700B capex pace, Amazon’s free cash flow will turn negative, while the four major hyperscalers’ combined cash reserves total $420B – sounds like a lot, but that only covers about half a year of capex.

This looks more like an infrastructure “slow bull” than a bubble about to pop. But “slow bull” doesn’t mean no correction. If AI revenue growth visibly decelerates in 2026-2027, market sentiment will turn quickly.

Of the Five Giants, Who’s Most at Risk?

Lay out capex, cash flow, and moat side by side, and the differences are stark.

Amazon: $200B capex, the gambler with negative FCF

Amazon is the heaviest bettor in this arms race. 2026 capex of $200B, $20B more than the runner-up. Morgan Stanley projects Amazon’s free cash flow turning to -$17B to -$28B this year – a company with $600B in annual revenue going cash flow negative. But Amazon’s logic is also the clearest: AWS is the world’s largest cloud platform, and AI training and inference demand converts directly to cloud revenue. AWS annual revenue is about to break $100B, growing at 19%. As long as cloud market share holds, this money isn’t wasted. Motorola risk: Low. Amazon’s moat is the infrastructure itself, not dependent on any single product or technology path.

Alphabet: $180B capex + century bond, center of attention

Alphabet’s 2026 capex of $175-185B, combined with the freshly issued century bond, made it Burry’s named target. But Alphabet holds two trump cards: $125B in cash reserves and the search advertising money machine. Google Cloud is growing at 48%, Gemini has partnered with Apple Siri – all monetizing. The real hidden concern is the erosion of search monopoly. If conversational AI search (ChatGPT, Perplexity) continues eating into Google’s search share, the cash cow itself gets wounded. Motorola risk: Medium-low. Won’t collapse, but the core business moat is being probed.

Microsoft: $145B capex, the steadiest and most boring

Microsoft is the most restrained of the five, with the slowest capex growth (+53%). Microsoft also has a unique advantage: Office 365 and Azure’s enterprise customers are natural channels for AI monetization, with Copilot embedded directly into existing products. Barclays estimates Microsoft’s FCF will only decline 28% this year and rebound by 2027. Motorola risk: Lowest. Microsoft is essentially selling shovels to gold miners. No matter whose model wins, Azure makes money.

Meta: $125B capex, the most puzzling one

Meta is the only one of the five without a cloud business. Amazon, Microsoft, and Google can at least sell their AI infrastructure as cloud services to third parties. Meta’s $125B capex can only be consumed internally – improving ad recommendations and feed ranking. Barclays estimates Meta’s FCF will plunge 90% this year. CEO Zuckerberg insists AI investment returns show up in “core advertising business improvements,” but that’s a hard case to make to investors. The last time Meta went all-in on a direction, it was called the Metaverse – we all know how that turned out. Motorola risk: Highest. Not that Meta will go under, but it has the worst mismatch between AI investment and visible returns among the five.

Oracle: $50B capex, smallest but most leveraged

Oracle has the smallest capex in absolute terms, but it’s the most indebted of the five. Net debt of $88B, more than 2x projected EBITDA. Oracle’s AI story is tightly coupled to OpenAI – if OpenAI doesn’t renew or diversifies suppliers, Oracle’s data center utilization faces a test. Motorola risk: Medium-high. Not because the business direction is wrong, but because the financial leverage leaves the least margin for error.

Build Always Runs Ahead of Demand

This “pour into infrastructure first, wait for demand” pattern is hardly new.

The massive fiber optic cables laid in the 1990s sat dormant for years after the bubble burst, then underpinned the entire mobile internet in the 2010s. When Amazon launched AWS in 2006, most people thought a bookstore doing cloud computing was insane – but AWS revenue will exceed $100B this year.

Infrastructure investment returns are never linear. They stay silent for a long time, then suddenly explode. The only questions are “how long is the silence” and “who falls in between.”

$650 billion in cumulative AI infrastructure investment looks like a massive gamble in 2026. This year will add another $500B. But stretch the time horizon to 10 years, and it’s likely just the foundation cost of a new era.

So who will be the next Motorola?

My take: none of the five giants will truly “go Motorola,” but not all of them will emerge unscathed. Meta and Oracle are in the most delicate positions – one has no cloud business to absorb AI infrastructure spending, the other has leverage ratios that should make you nervous. Amazon, Microsoft, and Alphabet are more like three companies doing the same thing in different postures: turning AI into the next generation of cloud infrastructure.

Motorola fell not because it invested too much money, but because it invested in the wrong direction. For today’s Big Tech, AI is almost certainly not the “wrong direction” – but how much to invest, how fast to break even, and whose cash flow cracks first – that’s the real life-or-death question.