In the late 1990s, a handful of telecom companies spent $750 billion laying fiber-optic cable across the ocean floor. They were right about the internet. They were wrong about the timeline. Most of them went bankrupt. The cable, however, changed the world.

I want you to hold that thought while I give you a number: $650 billion. That's what four companies — Alphabet, Amazon, Meta, and Microsoft — plan to spend on capital expenditures in 2026. Not revenue. Not market cap. Cash out the door, in a single year, mostly on data centers, GPU clusters, cooling systems, and the electrical infrastructure to power them. Amazon alone is spending $200 billion. Alphabet: $185 billion. Meta: up to $135 billion. Microsoft: $105 billion.

To put that in plain English — the combined 2026 capex of these four companies exceeds the total spending of America's 21 largest automakers, defense contractors, railroads, wireless carriers, and energy giants combined. Those 21 firms will spend roughly $180 billion. Four tech companies will spend 3.6 times that amount.

Think about what that means for the economy.

Every dollar of that $650 billion flows somewhere real. Semiconductor fabs running at maximum capacity. Construction crews building hyperscale campuses the size of small towns. Meta's Hyperion facility in Louisiana covers 3,650 acres — twice the size of New Orleans' main airport. Fiber-optic cable manufacturers are seeing demand they haven't experienced since the original telecom buildout. CoreWeave, a GPU cloud provider most people have never heard of, is spending $30 to $35 billion on its own capex this year.

And here is the part that should make you nervous. Wall Street analysts are warning that free cash flow across the Big Four could plummet by as much as 90% in 2026 as spending dramatically outpaces revenue growth. After Meta raised its capex guidance, the stock dropped 6%. Microsoft fell 2.5% on its spending disclosure. The market is starting to ask the question that matters: what if the return on this investment takes longer than the balance sheets can absorb?

As Buffett has said, only when the tide goes out do you find out who's been swimming naked. The AI tide is still rising. But the undertow of capital destruction is getting stronger.

If you own these stocks — and through index funds, you almost certainly do — you need to understand that you're riding the largest coordinated private infrastructure bet in the history of the global economy. The companies are probably right about AI. They may be wrong about the timeline. And the difference between "right about the technology" and "right about the investment" has destroyed fortunes before.

Most investors think the AI story is about software and models. It's not. It's about concrete, copper, kilowatts, and cooling. The constraint isn't intelligence. It's physics

Here is a number that explains why the DOE just committed $17.5 billion to nuclear reactors: a single hyperscale AI training cluster draws 100 megawatts — enough electricity to power a small city.

Data centers accounted for roughly 50% of all electricity demand growth in the United States last year, according to the International Energy Agency (IEA). Not 50% of data center growth. 50% of all new electricity demand in the country — dwarfing residential, industrial, and transport growth combined. The IEA projects this will continue through 2030. By 2028, data centers could consume 6.7% to 12% of total U.S. electricity, up from 4.4% in 2023.

The grid wasn't built for this. In Northern Virginia — home to the world's largest concentration of data centers — facilities already consume almost 40% of the state's total electricity. In July 2024, a voltage fluctuation triggered the simultaneous disconnection of 60 data centers, forcing a 1,500-megawatt emergency adjustment to prevent cascading outages. AEP Ohio has paused all new data center interconnections. Communities in Oregon, Georgia, Indiana, and Missouri are fighting proposed facilities. Maine's legislature just approved a statewide moratorium on new data centers.

And one more thing. The PJM Interconnection — the regional grid serving 65 million people across 13 mid-Atlantic and Midwestern states — saw capacity auction prices surge to a record $333.44 per megawatt-day. That's a 12-fold increase from two years ago. Analysts estimate this will add 10–20% to average electricity bills across those states.

So why does this matter to you? Because electricity bills have already risen 42% since 2019, according to Brookings. Utilities requested $31 billion in rate hikes in 2025 alone. Goldman Sachs estimates data center-driven electricity demand will add 0.1 percentage points to core inflation in both 2026 and 2027. In a world where the Fed is debating rate hikes because inflation is stuck at 4.1%, that 0.1% isn't trivial — it's the difference between holding and hiking.

The AI boom isn't just a tech story. It's an energy story, a utility story, and increasingly, an inflation story. If you own utility stocks, you're in the right sector. If you pay an electricity bill, you're subsidizing it.

Moody's chief economist Mark Zandi published a note this week that every investor should read. Here's the headline: the top 20% of American households — those earning $200,000 or more — now account for 60% of all personal consumer outlays. In the year ending Q1 2026, their spending grew 6.5%, or 4% after inflation.

During the dot-com bubble, that same cohort accounted for 50% of spending. Today it's 60%. And the engine driving their spending is the same engine that drove it in 1999: surging stock prices. The Fed's distributional accounts show that nearly 90% of corporate equities and mutual funds are held by the top 20%.

In other words, the entire U.S. consumer economy — GDP grew at a 2.1% annual rate in Q1 — is riding on the portfolio values of the wealthiest fifth of households.

Zandi doesn't call it a bubble. He calls it "overvalued, bordering on speculative." Price-to-earnings multiples are at 19x. He notes that AI stocks have soared "for strong fundamental reasons" but have also received a boost from index funds that mechanically buy whatever gets bigger — regardless of valuation. Sound familiar? It should. The same dynamic drove the Nifty Fifty in the 1970s and the dot-com concentration in 2000.

Keep in mind: unemployment is stable at 4.3%. GDP is growing. Hiring averaged 188,000 jobs per month over the last three months. By the standard macro indicators, the economy is fine. But the consumer spending that supports those indicators is balanced — Zandi's word — on "an increasingly precarious set of dominoes."

If you're positioned for continued consumer strength, you're betting that stock prices hold. If stock prices hold, the top 20% keep spending. If they keep spending, GDP looks resilient. If GDP looks resilient, the Fed hikes. If the Fed hikes, stock prices wobble. And if stock prices wobble...

You see the problem. The dominoes run in a circle.

Here is a fact that would have seemed absurd five years ago. Capital expenditure by just five technology companies is now larger than global investment in oil and gas production. The IEA confirmed this in its latest energy report, and the gap is widening.

The spending isn't speculative in the traditional sense — these companies have the cash. Alphabet, Amazon, Meta, and Microsoft generated hundreds of billions in combined revenue last year. They're not borrowing to build. They're deploying profits. But the scale creates second-order risks that nobody modeled.

First, the memory famine. As we've discussed, HBM chips for AI now consume 23% of all DRAM wafer production, starving smartphones and PCs. Global smartphone shipments are forecast to fall 13% this year. PC vendors are passing through 15–20% price hikes.

Second, the power crisis. Gas turbine orders surged 70% in 2025 because data centers need baseload generation that solar and wind can't reliably provide. Gartner predicts power shortages will restrict 40% of AI data centers by 2027.

Third — and this is what Dalio is watching — the geopolitical exposure. These companies are building critical infrastructure that depends on Taiwanese semiconductors, Chinese rare earth magnets, Middle Eastern energy, and American electrical grids that haven't been upgraded since the 1970s. Every chokepoint we track in this newsletter — Hormuz, TSMC, rare earths, the PJM grid — converges on the same thing: keeping the AI factories running.

I urge you to think about AI not as a software revolution but as an industrial revolution. Industrial revolutions require raw materials, energy, and infrastructure. The companies that control those inputs will define the winners. The companies that assume those inputs will always be available are the ones that get disrupted.

In December 1999, the S&P 500 was powered by a simple thesis: the internet would change everything. The thesis was correct. The stocks still fell 49%.

Today, the thesis is that artificial intelligence will change everything. And again — the thesis is almost certainly correct. AI is already reshaping legal work, medical diagnostics, software engineering, and military intelligence. The $650 billion being deployed by Big Tech isn't irrational in the abstract. The underlying technology is real, the demand is accelerating, and the companies spending the money are among the most profitable in history.

But here is what I keep coming back to. The economy is now structured so that a narrow set of outcomes must all remain true simultaneously for the system to hold together.

First, AI stocks must maintain their valuations — because the top 20% of households own 90% of equities, and their spending drives 60% of consumption. Second, the Fed must not hike too aggressively — because with $39 trillion in federal debt and interest payments approaching $1 trillion, an aggressive tightening risks a fiscal crisis. Third, the electrical grid must absorb demand growth that is literally unprecedented in peacetime history — because if power constraints stall the AI buildout, the capex cycle reverses, and the stocks that underpin consumer spending lose their fundamental story. Fourth, the geopolitical supply chains for chips, minerals, and energy must remain intact — because one disruption to TSMC, one expansion of China's rare earth controls, or one breakdown in the Hormuz ceasefire pulls the thread that connects all the others.

No single one of these conditions is implausible. What's fragile is the requirement that all four hold simultaneously, indefinitely.

Howard Marks calls this "second-level thinking" — the discipline of asking not "what will happen?" but "what happens if I'm wrong?" Most investors today aren't wrong about AI. They're wrong about the assumption that the system supporting AI is unbreakable.

If you own index funds, you own this concentration. The top five stocks in the S&P 500 now represent a larger share of the index than at any point since data began. Your "diversified" portfolio isn't diversified. It's a leveraged bet on four companies' ability to turn $650 billion in spending into revenue before the grid, the supply chain, or the consumer buckles.

I don't know which domino falls first. I know there are more dominoes than most people are counting.

Payrolls day. If the number comes in hot — above 200,000 — September rate-hike odds move toward 70% and two-year Treasury yields push higher. If it disappoints below 150,000, the hawks lose their case and gold catches a bid. Either way, this is the most consequential single data point between now and Jackson Hole.

The 12-fold surge in PJM capacity auction prices hasn't fully flowed through to retail bills yet. Watch for utility rate filings in Virginia, Ohio, and Pennsylvania in the coming weeks. If rate increases land in the 10–20% range that analysts project, the political backlash against data centers accelerates — and with it, the risk of construction moratoriums beyond Maine.

The market absorbed $650 billion in capex guidance in February. It hasn't yet seen what that spending looks like on a quarterly cash flow statement. If free cash flow collapses as sharply as analysts fear — down 60–90% year-over-year — the stocks that power the top 20%'s spending could face their first serious test of the cycle.

The Islamabad Memorandum of Understanding established a 60-day ceasefire window. That clock is ticking. If the Doha technical talks don't produce a durable framework before mid-August, the entire fragile edifice — oil prices, shipping insurance, energy inflation — resets to a war footing.

China's October 2025 extraterritorial rare earth measures remain suspended until November. When the suspension lifts, any foreign-made product containing more than 0.1% Chinese-origin rare earths will require a license. If you hold defense, EV, or industrial equities, this is the deadline that determines whether their supply chains survive or fracture.