For 15 years, the playbook was simple:
- Build software.
- Zero marginal cost.
- Infinite scalability.
- 80% margins.
- Buybacks.
- Multiple expansion.
This was the "asset-light" model. It made Silicon Valley the most profitable rent-seeking machine in human history. It convinced investors that tech companies weren't capital-intensive—they were intellectual property empires printing cash from servers.
That narrative died this week.
The $740 Billion Truth Bomb
Here's what actually happened while you were watching Bitcoin reject $69K:
Four companies—Google, Amazon, Microsoft, Meta—just guided 2026 capital expenditures to $650 BILLION.
Add Oracle and CoreWeave. $740 billion .
Let me contextualize that number for you:
- Up 70% year-over-year
- Double what the market expected (35% growth)
- Nearly equals the ENTIRE operating cash flow of these companies
- Approaches 1.4% of U.S. GDP—dot-com bubble territory
This is not a "capital expenditure cycle." This is a structural regime shift.
---
The Cash Flow Collapse You Haven't Priced In
Goldman Sachs: "If this spending holds, Mag 7 will have zero free cash flow for buybacks in 2026."
Bank of America: "Excluding Microsoft, every hyperscaler's free cash flow goes to zero—or negative—even if they completely halt buybacks."
Meta: Already signaling a move from "net cash neutral" to "net debt positive."
This is the part the CNBC anchors aren't screaming:
The companies that defined "asset-light" are now the most capital-intensive enterprises on earth.
A data center isn't software. It's concrete, copper, and cooling towers. It depreciates. It consumes power. It requires ongoing maintenance CAPEX.
This is not a pivot. This is a permanent transformation.
$BERA The Debt Bubble Nobody Is Watching
When internal cash flow isn't enough, you go to the bond market.
Oracle: $25 billion bond issuance last week. $129 billion in orders—5x oversubscribed. Stock down 15% anyway because the market realized they're borrowing to survive, not to grow.
Google: $20 billion dollar bond—**$100 billion in orders**. Largest in company history. Then immediately turned around and issued:
- £5.5 billion sterling bonds (largest ever in UK corporate history)
- CHF bonds (broke Roche's record)
- A 100-YEAR BOND—first by a tech company since 1999
Let's sit with that.
Google—$125 billion in cash, $90 billion in annual FCF from advertising—is issuing century bonds.
Why? Because they don't believe the cash will be there when they need it. They're pre-funding the next decade of losses today, at today's rates, because they know the cost of capital only goes up from here.
AI-related investment-grade debt now accounts for 14% of the entire U.S. IG bond market.
That's larger than the banking sector.
The market has quietly shifted from equity-sponsored growth (buybacks, multiple expansion) to debt-sponsored survival (leverage, interest coverage, refinancing risk).
That changes everything about how you value these companies.
$BTC The Prisoners' Dilemma: Why They Can't Stop
Here's the part that should terrify you:
Every CEO knows this is irrational.
Goldman ran the math. To justify $500-600B annual CAPEX, these companies need to generate $1 TRILLION in annual profits by 2028 .
Current consensus for 2026: $450 billion.
That's a $550 billion profit gap.
Explain to me how $30/month ChatGPT subscriptions and enterprise API calls close that gap. You can't. The math doesn't work.
So why do they keep spending?
Because the game theory is brutal:
This is a Nash equilibrium. Every player acts rationally in their own interest, and the collective outcome is collective self-destruction.
As Goldman put it: "Even if near-term returns are compressed, continuous capital expenditure remains rational at the individual level."
Translation: We know we're building a bridge to nowhere. But if we don't build it, someone else will—and we'll be stranded on this side forever.
The Second-Order Effect: AI Is Eating Its Own Children
Here's the part the "AI bull" narrative completely misses:
The same AI infrastructure these giants are building is actively destroying the valuation of their own customers.
Software companies—SaaS, enterprise software, vertical applications—are getting obliterated because investors realize AI makes most of them obsolete .
Why pay $50/seat/month for Salesforce when an AI agent can write CRM entries, schedule calls, and generate reports automatically?
Why pay $200/user for Adobe when Midjourney + Claude does 80% of the work?
This is not hypothetical. Software company bonds are getting hammered. Leveraged loan prices in software: down 4% YTD .
And here's where the systemic risk lives:
Private credit funds (BDCs) have 20% of their portfolios in software debt.
- 50% of software loans are rated B- or lower
- 26% are CCC—junk by any definition
- 46% of software debt matures in the next 4 years
If AI replaces software headcount and software revenue collapses, those loans don't get refinanced. They default.
And when BDCs start taking 20-30% losses on 20% of their portfolio, the credit cycle turns.
The AI giants aren't just spending themselves into debt. They're engineering the collapse of the ecosystem that buys their cloud compute.
The Two Endgames
Scenario A: Cloud 2.0 (The Bull Case)
AI adoption follows the AWS trajectory:
- 3 years to breakeven
- 10 years to 30% margins
- $1.5 trillion in backlogged cloud orders eventually converts to revenue
In this world, today's $740B/year CAPEX looks cheap in 2032. Google's 100-year bonds trade at a premium. The debt bubble was actually "pre-funding a productivity revolution."
Scenario B: Global Crossing 2.0 (The Bear Case)
The 1990s fiber optic bubble wasn't built by dumb money. It was built by rational actors overestimating demand.
Global Crossing laid cable across the Atlantic because everyone knew the internet would need bandwidth. They were right. They were just 20 years early.
The company went bankrupt. The bonds went to zero. The infrastructure got bought for pennies on the dollar.
The same dynamic applies today.
Is AI overestimated by 20%? Or 200%?
If it's the latter, the $4 trillion BI projects through 2030 doesn't become profits. It becomes stranded assets .
And the bond market wakes up.
$ZRO What This Means For You
For Bitcoin:
Tech debt bubbles eventually break risk assets. If the bond market closes for Mag 7, liquidity evaporates everywhere. $64K support becomes $52K. Watch credit spreads, not NFP.
For Equities:
The "asset-light" premium is gone. You are now valuing capital-intensive infrastructure businesses trading at software multiples. That math doesn't work. Expect multiple compression to accelerate.
For Credit:
The IG market is now 14% levered to AI CAPEX. The BBB tranche is the canary. If spreads blow out here, the entire corporate debt stack reprices.
For Your Thesis:
The question is no longer "Will AI change the world?" It will.
The question is: "Will the companies spending $740B/year be the ones who capture that value—or will they be the fiber optic backbone that someone else profits on top of?"
History suggests the latter.
The asset-light model didn't die because CEOs made bad decisions.
It died because AI is physically impossible to deliver without assets.
You cannot run GPT-7 on "intellectual property." You need nuclear reactors, rare earth minerals, fiber optic cable, and cooling systems the size of football fields.
That's not software. That's infrastructure.
And infrastructure has never traded at 25x sales.
The re-rating has just begun.
What's your exposure?
⬇️ Are you still holding Mag 7 through this transformation, or rotating into something that doesn't need to borrow for 100 years to survive?
💬 If you're long Google or Microsoft here—defend the thesis. I want to hear it.
#AI #CreditMarkets #AssetLight #Macro #NotFinancialadvice