The AI capital expenditures of Alphabet, Amazon, Meta, Microsoft, and Oracle are projected to exceed $800 billion in 2026 and reach $1.1 trillion in 2027, accounting for approximately 3.2% of U.S. GDP.

Author and source: 0x9999in1, ME News



TL;DR

• The Kobeissi Letter estimates: The AI capital expenditures of Alphabet, Amazon, Meta, Microsoft, and Oracle will exceed $800 billion in 2026 and reach $1.1 trillion in 2027, accounting for approximately 3.2% of U.S. GDP.
• If realized, AI's annual capital expenditure will exceed U.S. defense spending for the first time, with defense spending projected to account for approximately 2.7% of GDP next year.
• This is not a distant prophecy. In the first quarter of 2026, four of these giants alone spent over $130 billion in capital expenditures for the quarter, on track to reach $700 billion to $785 billion for the full year.
• The money has already run out. According to Epoch AI, these companies' cash capital expenditures are expected to exceed operating cash flow around the third quarter of 2026—shifting from "investing with earnings" to "investing with borrowed money."
• The Bank for International Settlements (BIS) has officially classified AI infrastructure investment as a systemic risk, specifically citing "circular financing."
• NVIDIA invests in OpenAI, and OpenAI buys NVIDIA chips—same money going in a circle, making demand appear to double.
• This is a high-stakes gamble, with the fate of the nation on the line. The only question is: who will pay the bill?

A number that makes you sit down.

$1.1 trillion.

Written out, it is 1.1 trillion. It is ten thousand billion.

This is 2027, and the amount five U.S. companies are expected to spend on AI infrastructure—not market capitalization, not revenue, but capital expenditures—on land, building data centers, stockpiling chips, laying power lines, and installing cooling systems.

Spending it in a way that many things would be depreciated and written off within five years.

Approximately 3.2% of the U.S. GDP.

What does this mean? Next year's U.S. defense budget is approximately 2.7% of GDP.

In other words, if Kobeissi Letter’s calculations are accurate, the amount spent by five tech companies on GPUs would exceed the total U.S. government spending on aircraft carriers, missiles, and military salaries.

Five companies. Facing off against the military of a superpower.

You read that right. I also stared for three seconds when I first saw it.

This is not a prediction; it is happening now.

Some might say that 2027 is still far off—predictions are just meant to be listened to.

No. The terrifying thing about this is that it’s not a fantasy—it’s the elevator beneath your feet, and it’s already moving upward.

Look at this year: In the first quarter of 2026, Alphabet, Amazon, Meta, and Microsoft alone reported capital expenditures exceeding $130 billion for a single quarter. What’s the full-year target? According to JPMorgan, it’s over $700 billion; Moody’s has raised its 2026 forecast for the six largest tech firms to $785 billion, approaching $1 trillion in 2027.

Looking back further: In 2025, these companies’ AI capital expenditures accounted for approximately 1.5% of GDP. This year, 2026, it jumps directly to about 2.5%.

One year, up by two-thirds.

Epoch AI created a chart showing that since the release of GPT-4 in March 2023, capital expenditures by hyperscale companies have grown by approximately 70% annually. Extending this trend forward, the $1.1 trillion figure for 2027 is not an aggressive assumption—it’s the inevitable outcome of continuing at this pace.

So the question isn't "Will it reach $1.1 trillion?"

The question is, "What do we do after reaching $1.1 trillion?"

Why are four companies with completely different businesses suddenly all doing the same thing?

This is the part I find most intriguing.

Microsoft sells cloud services. Google focuses on search and advertising. Meta earns revenue from social media feeds. Amazon combines e-commerce with AWS. Oracle holds onto enterprise databases.

Five businesses, five logics, five sectors.

But at the same moment, they reached the same conclusion: if you don’t invest in computing power now, there won’t be a seat at the table later.

Why?

Because they all believe in one thing—AI is not just a feature, but a platform-level shift, much like the transition from PC to the internet, and from the internet to mobile. Whoever falls even one step behind in infrastructure risks being left behind for an entire generation.

Nokia wasn't lacking effort either; it just didn't invest enough money in the right direction.

So it's not greed—somewhat, it's fear.

The fear of missing out outweighs the fear of wasting.

Management’s calculation is clear: spending an additional 100 billion yuan could at worst drag down profits for a few years through depreciation. Spending 100 billion yuan less could at worst mean surrendering the next decade.

Which is more terrifying?

For a CEO, the answer isn't hard to choose.

I'm running low on money.

But the accounts must be settled.

There's a signal here that deserves even more caution than the $1.1 trillion figure.

Epoch AI estimates that the cash capital expenditures of these five companies are expected to exceed their operating cash flow around the third quarter of 2026.

In the past, they would make $100 and invest $80 of it. Now, they’re making $100 and investing $120.

Where does the extra money come from?

Borrow.

Bond issuance, loans, and various off-balance-sheet "shadow lending" structures.

This is why the Bank for International Settlements—the institution known as the "central bank of central banks"—has officially classified AI infrastructure investment as a systemic risk. It specifically named three factors: deferred depreciation pressure, "shadow lending" through private credit channels, and the most subtle of all—recycling financing.

Systemic risk. The central bank does not use these four words lightly.

The last time I heard similar wording was around 2008.

That spinning trillion

Recurring financing. This one deserves a proper explanation.

Here's the story: NVIDIA invested a massive amount of money—reportedly in the hundreds of billions—into OpenAI. OpenAI used that money to buy chips—specifically, NVIDIA’s chips. NVIDIA then recorded the revenue from selling those chips as its own income.

Do you see the issue?

The same money circled around and returned to its starting point, yet on paper, it became NVIDIA's "investment," then NVIDIA's "revenue," and finally OpenAI's "compute expansion."

One coin, three blooms. Demand appears to have doubled.

On July 1, 2026, NVIDIA introduced a new financing model combining revenue sharing with credit support, enabling AI cloud providers to pledge future computing power and pre-emptively access capacity. Harvard’s Negotiation Project and multiple research institutions are monitoring such “circular deals” and issuing warnings.

I'm not saying there's fraud here. Most of these transactions are public and compliant.

But I’d like to ask: when the people selling shovels start lending money to the gold miners so they can buy their shovels—does this gold mine still have any gold left?

Prosperity is sometimes real. Sometimes, it's an echo.

So, is this really a bubble?

I know you want to ask about this.

My assessment is: yes and no. This may sound like avoiding a clear stance, but please let me explain.

It’s called a bubble because it exhibits all the classic characteristics: valuation mania—OpenAI raised approximately $122 billion in funding as of March 2026, valuing it at around $852 billion, a figure verified by the Financial Times. Funding is increasingly driven by debt rather than cash flow. Transaction structures are becoming more convoluted and opaque. These are all hallmarks of every historical bubble.

But it’s not entirely a bubble, because it has a fundamental difference from the dot-com bubble of 2000.

Many companies back then had stories but no revenue, presentations but no products.

Today, these five companies—Microsoft, Google, Amazon, Meta—are cash cows that generate real, tangible profits. They fund their spending with money squeezed from their own profits, plus some borrowing. The data centers they build are physical assets capable of running real business operations and generating actual cloud revenue.

Jensen Huang estimates that demand for AI chips alone will reach at least $1 trillion before 2027. U.S. entities account for approximately 80% to 85% of global AI and data center capital expenditures.

This is not pure air.

So a more accurate description might be: this is not a bubble blown out of thin air, but a real demand-driven arms race that may be significantly ahead of its time.

The real赛道. Potentially inflated stakes.

Two things, both true at the same time.

When a company's spending rate catches up to a nation's military budget

Return to that most striking comparison: AI capital expenditures exceeding defense spending.

This sentence hides a shift of an era.

Over the past century, the most expensive and defining indicator of a nation’s power was its military—aircraft carriers, fighter jets, and nuclear arsenals. Steel and gunpowder were the measures of national strength.

And now? The battlefield defining future strength is shifting from the physical world to the world of computational power.

The Pentagon defends territory. Data centers compete for intelligence.

When private companies invest more in "computing power arms races" than governments do in "military arms races," it is itself a signal—the center of power is quietly shifting.

David Sacks said AI capital expenditures may contribute approximately 2.5% to GDP growth this year and over 3% next year.

This means that U.S. economic growth is increasingly being supported by purchase orders from these five companies.

Is this a good thing?

In the short term, yes. It creates demand, drives electricity consumption, boosts construction, and supports employment.

In the long term, I'm not sure.

The foundation of an economy that bets its growth on just a few companies, a handful of chips, and an unproven business model may not be as stable as it appears.

Who will pay?

In the end, someone always has to pay the bill.

This $1.1 trillion is being paid in the short term by shareholders—using profits consumed by depreciation.

In the medium term, creditors bear the risk—if AI’s returns fall short of expectations, who will repay the lent funds?

In the worst-case scenario, if the systemic risks feared by the BIS were to materialize, the entire financial system would bear the cost—just as in 2008, with the final bill falling on every ordinary person.

I'm not being negative.

I just want to say that when everyone is cheering for a number, someone needs to calmly ask: Will these data centers ever recoup their costs?

If possible, this would be the greatest infrastructure leap in human history, on par with the interstate highway system or the moon landing.

If not, this will be the most expensive pile of graphics cards that quickly become obsolete.

No one can give an answer before 2027.

In conclusion

$1.1 trillion. More than the defense budget of a superpower.

Is this ambition or frenzy? Vision or collective delusion?

Maybe all of them.

Great things and foolish things often look identical at the moment they happen. The difference lies only in the outcome, and the outcome always arrives late.

Five companies have pushed chips of national significance to the center of the table.

The card hasn't been flipped yet.

All we can do is remember today’s numbers and patiently wait for the day when the final card is revealed.

By then, it will be clear whether it will be applause or sighs.

Rushing won't help.

History never speeds up for anyone.

Reference materials

1. The Kobeissi Letter, "The AI Spending Boom Is Redefining the U.S. Economy," X (formerly Twitter), June 2026
2. CNBC, "AI boom: Big Tech capital expenditures now seen topping $1 trillion in 2027", April 30, 2026
3. Compute Forecast: "Hyperscaler Capex Forecast for 2026 Raised to $785 Billion" (according to Moody's estimates), 2026
4. Epoch AI, "Hyperscaler CapEx to Exceed Cash Flow by Q3 2026"
5. Forbes, "U.S. Will Spend 2% Of Its GDP On AI This Year—Almost As Much As Defense And Education Budgets," June 5, 2026
6. TradingKey, "AI Capex: The Next Source of Systemic Financial Risk?" (citing the Bank for International Settlements, BIS), 2026
7. Forbes, "Credible AI Lab Critics Pile Up As The Bubble Math Worsens" (citing OpenAI funding data verified by the Financial Times), July 4, 2026
8. Harvard Program on Negotiation, "What Are Circular Deals?"