Investor Briefing · azizsaif.com · July 2026
The greatest bet in history — or the greatest bubble ever inflated?
Big Tech is spending $725 billion in one year to build AI. The industry it's feeding earns a fraction of what it costs. Here's the case — in numbers, fact-checked against current market reporting.
Source video below · figures verified July 2026
▶ Watch the full breakdown, then scroll for the verified data dashboard.
$725B
2026 Big Tech AI capex (AMZN, MSFT, GOOG, META)
▲ 77% vs 2025 · 8× vs 2020
$650B
Revenue AI must earn to justify the spend
JPMorgan / Sequoia "$600B question"
~$75B
What AI model makers actually earn today
— 9–10× short of target
$5.2T
Nvidia market cap at the June 2026 peak
Then shed ~$320B in 3 days
01 · The Spend
Capex went up 8× in six years
In 2020, before ChatGPT existed, the four biggest US tech firms spent a combined $90B on capex. In 2026 that figure is $725B — almost entirely AI infrastructure.
Combined AI capex — 4 hyperscalers
Amazon · Microsoft · Alphabet · Meta ($ billions)
Verified: ~$410B (2025) → $725B (2026), +77%. Analysts project $1T+ in 2027.
Who is spending the $725B (2026)
Planned capex by company ($ billions)
Verified: Amazon ~$200B · Microsoft ~$190B · Google $175–185B · Meta $115–135B.
Big Tech capex is set to consume 94% of operating cash flow over the next two years.
For every $100 earned, ~$94 goes back into AI infrastructure — leaving ~$6 for dividends, buybacks and everything else. That ratio was ~40% in 2023. (PIMCO estimate)
02 · The Revenue Gap
The $600 billion question
To justify the spend at a bare-minimum 10% return, AI needs ~$650B a year. It earns roughly $75B. That's the gap nobody knows who fills.
What AI must earn vs. what it earns
Annual, 2026 ($ billions)
Verified: JPMorgan pegs required revenue at $650B; Sequoia's David Cahn calculates a ~$600B annual gap, widening in 2026.
AI revenue vs. spend, side by side
2026, the four numbers that matter ($ billions)
Model-maker revenue ≈ $75B; losses ≥ $17B; giants spend $725B. Divergence ~46% vs ~32% in the 2001 telecom cycle.
Does "enterprises will pay" hold up?
73%
of enterprise AI deployments miss projected ROI (McKinsey)
5%
of companies see substantial ROI from AI (BCG)
95%
failure rate in achieving measurable financial return (MIT)
And when the bill lands, buyers move. One SF startup (Lindy) switched 100% of traffic off a premium model and cut costs ~90%. Uber's CTO admitted it burned its entire annual AI budget in 4 months. Palantir's CEO: customers say they're "paying for tokens that create no value."
03 · The AI Tax
Why your laptop got more expensive
The three firms making 90%+ of the world's memory shifted ~93% of production to AI memory (HBM), which pays up to 10× more per module. Consumer prices followed.
Memory (DRAM) price spike
Selected verified 2026 moves (% increase)
Verified: DRAM ASP ~+90% QoQ in Q1 2026, +50–60% again in Q2; DRAM up ~171% YoY; 1GB went $0.43 → $2.39 in 6 months (Dell).
Apple's mid-year price hike — a first
June 25, 2026 · % increase at entry price
Verified: MacBook Air $1,099→$1,299 (+18%); iPad Pro $999→$1,199 (+20%). Apple stock fell ~6% that day.
Apple — the richest, most vertically integrated tech company on earth — said it "has never seen a component price increase this much this quickly" and can no longer absorb the cost. That's how the AI build-out reaches your wallet.
04 · The Mechanism
The capital cycle behind every bubble
If ROI is broken and buyers are fleeing, why keep spending more? Because capital always chases the highest return — until it overbuilds.
1
High returns attract capital
A real, world-changing technology pulls in a flood of money.
2
Capital overshoots
Money keeps flowing until far too much capacity is built.
3
Over-capacity → collapse
Supply dwarfs real demand; prices and valuations crack.
4
A few survivors win big
Wreckage is bought cheap; the infrastructure powers the next era.
05 · The Precedent
The dark-fiber mirror
After the 1996 Telecom Act, firms poured $500B+ (mostly debt) into fiber, certain demand would explode forever. It grew — just not fast enough.
| What happened, 2000–2002 | Figure |
|---|
| Fiber actually carrying data by 2002 | 2.7% |
| Installed fiber left dark / unused | ~95% |
| Bandwidth price collapse | up to −90% |
| Telecom market value wiped out | ~$2 trillion |
| Global Crossing valuation → bankruptcy | $47B → $0 |
| WorldCom hidden expenses → biggest US bankruptcy | $3.8B |
The technology was real. YouTube, streaming, cloud and smartphones eventually used exactly that overbuilt fiber. The tech survived; most of the companies that built it did not. In 2026, America is building $725B of data centers a year — and no one yet knows how much will be used.
06 · The Verdict
Bubble — or the greatest bet ever?
The honest answer sits in between. Unlike the telecom players, today's spenders are the most profitable companies in history — and valuations, while stretched, are nowhere near 2000.
Valuation: 2000 peak vs. today
Nasdaq-100 forward P/E
Verified: ~60× at the March 2000 peak vs. ~26× today. High, but not 1999 insanity.
This time, the spenders earn real money
Why a telecom-style wipeout is less likely
Nvidia earned ~$120B net income last year; Microsoft, Google & Amazon are among the most profitable firms ever — funded by cash, not just debt.
HIGH probability of a bubble
…but NOT a certainty
Path 1
The bubble pops
NASDAQ corrects, spending slams shut, jobs are lost. That spend feeds India's IT & services sector — so the pain spills far beyond Silicon Valley.
Real risk
Path 2
It doesn't pop — AI gets expensive
To justify trillion-dollar valuations, firms race to profit. Token prices rise; only giants afford AI; today's cheap tools become a luxury.
Also plausible
Path 3
A miracle
Token costs crater, enterprises finally pay, and everyone makes money at once.
Teeny-tiny possibility
We're not betting on whether AI changes the world. We're betting on whether the price makes sense.
The technology is real. The revenue is real. The question is the multiple.