The $80 Billion Bet: How the AI Capital Arms Race Is Minting Winners — and Quietly Killing Everyone Else

There is a number that keeps coming up in conversations among investors, analysts, and executives right now, and it is so large that it has started to lose meaning. One trillion dollars. That is the rough annual capital expenditure figure that the world's biggest technology companies are barreling toward by 2028 — money being poured into data centers, graphics processors, fiber, power substations, and the cooling systems needed to stop all of it from melting. And this week, that number got a face: Alphabet, the parent company of Google, announced it would raise $80 billion through stock sales to fund what it called its "world-class AI compute infrastructure."
The markets, it turned out, were not entirely sure what to make of this.
Alphabet shares fell nearly 4% on Monday as Wall Street processed the deal. On the surface, the drop looked paradoxical. Here was one of the most cash-generative companies on the planet, fresh off a quarterly earnings report that had impressed even the skeptics, raising money from Berkshire Hathaway and a syndicate led by Goldman Sachs, JPMorgan, and Morgan Stanley. If anything, it should have been a signal of confidence. But the reaction revealed something more complicated sitting underneath the headline: the AI buildout has become so expensive that even the companies winning the race are starting to look financially stretched, and investors are beginning to ask questions they were not asking twelve months ago.

The Deal Itself

The structure of the Alphabet raise is worth examining, because it is not a simple equity offering. The $80 billion package breaks into three distinct tranches. First, $10 billion comes via a private placement to Berkshire Hathaway — split evenly between Class A and Class C shares at prices of $351.81 and $348.20 respectively. Second, $30 billion will come from underwritten public offerings, half of which takes the unusual form of "mandatory convertible preferred stock." Third, and perhaps most notably, Alphabet plans a $40 billion at-the-market program — meaning the company will sell shares directly into open trading from the third quarter onwards, drip-feeding dilution into the market over time rather than absorbing it in one large hit.
The Berkshire piece of the puzzle has drawn particular attention. Warren Buffett's successor as CEO, Greg Abel, has been building a position in Alphabet since the third quarter of last year, and the conglomerate's stake before this week's deal was already worth roughly $20 billion. The additional $10 billion commitment is one of the largest single equity investments Berkshire has made in recent memory, and it signals something meaningful: the most famous value investor on the planet has decided that Google's position at the center of AI — spanning search, cloud, and infrastructure — is worth paying up for.
"The company is experiencing strong demand for its AI solutions and services from enterprises and consumers, at levels that are exceeding the company's available supply," Alphabet said in its official statement. That is a candid admission buried in corporate language: even with billions already deployed, Google cannot build fast enough to meet what people are asking of it.
The funds will go toward what Alphabet describes as scaling its AI compute infrastructure to meet "unprecedented customer demand." In April, the company had already updated its full-year capital expenditure guidance to as much as $190 billion — a figure that would have seemed science fiction just three years ago. This latest fundraise is essentially a bridge over the gap between what the company can fund internally and where it needs to be.

A Race Nobody Can Afford to Lose — or Afford to Win
Alphabet is not alone in this position. It is not even the most extreme case.
Amazon has guided for roughly $200 billion in capital expenditure for 2026, a number that materially exceeds its operating cash flow and has the company returning to debt markets in size. Microsoft is doing the same, albeit at a somewhat more measured pace. Meta has committed to a capital spending program that analysts project will push its free cash flow negative. Oracle, after signing a $300 billion compute capacity deal with OpenAI last year, raised its own capex guidance by 43% and is bridging the shortfall with bonds.
Across the five largest cloud and AI companies combined, analysts at multiple firms now project 2026 capital expenditure in the range of $635 to $690 billion — more than double what these companies spent just two years ago. Nvidia CEO Jensen Huang, speaking on an earnings call last month, suggested that AI capex could eventually reach $3 to $4 trillion annually. Even if that is half right, the numbers involved dwarf every previous technology investment cycle in history, including the buildout of the internet itself.
The debt markets have taken notice. Morgan Stanley and JPMorgan estimate that technology sector debt issuance to finance AI infrastructure will exceed $1 trillion over the next few years. BlackRock's investment strategists flagged this week that hyperscalers are increasingly financing their buildouts through debt as interest rates remain elevated, calling it a deliberate leveraging-up strategy rather than a sign of distress. The distinction matters, but it does not make the underlying dynamic any less consequential: the AI race is now running on borrowed money, and the interest clock is ticking.
The question investors keep circling is whether the revenues will follow. Goldman Sachs Research has noted a growing divergence in the market, with investors rewarding companies that can demonstrate a clear line between what they are spending and what they are earning, while rotating away from those where operating earnings growth is under pressure but the capex tap remains wide open. Alphabet, for its part, has been among the better performers on that measure — its cloud division has been growing rapidly, and its AI products have captured genuine enterprise adoption. The $80 billion raise, in that context, is less a plea for survival and more an attempt to lock in infrastructure capacity before competitors do.

"RAMageddon": The Other Side of the Equation
While the hyperscalers burn through capital in their race to build, the commodity markets that sit beneath all of this infrastructure are experiencing a reckoning that is destroying smaller companies in real time.
Memory chips — specifically DRAM and NAND flash storage — are at the center of it. AI data centers require enormous quantities of high-bandwidth memory, and the three companies that dominate global production, Samsung, SK Hynix, and Micron, have responded to that demand by reorienting their factories. Production of the kinds of memory chips used in cameras, consumer laptops, cars, and smartphones has been sharply reduced in favor of the higher-margin, higher-specification chips that hyperscalers need. Supply to consumer electronics companies has fallen. Prices have exploded.
Some in the industry have taken to calling it "RAMageddon."
For GoPro, the California-based maker of action cameras popular among extreme sports enthusiasts and adventure travellers, the timing could not have been worse. The company was already navigating a difficult stretch — revenue had declined nearly 20% in 2025, competition from DJI had eroded its market share to the point where its Chinese rival now controls roughly two-thirds of the action camera category, and a series of costly missteps had left it with barely $50 million in cash reserves. Then, in the last week of March, memory chip suppliers informed GoPro that prices would rise between 80% and 115% — not gradually, not over a year, but immediately. In April, the same suppliers announced reductions in the actual volume of memory they would sell to the company, cutting into GoPro's ability to manufacture at all.
On June 1, GoPro filed an 8-K with the Securities and Exchange Commission disclosing "substantial doubt about the company's ability to continue as a going concern." Its auditor, PricewaterhouseCoopers, echoed the warning. The company's stock fell as much as 14% on the news to close at $1.10 — a price that requires some historical context to fully appreciate. GoPro went public in 2014 at a $3 billion valuation. Its shares peaked above $90 that same year. It now has a market capitalization of approximately $168 million.
The memory price shock compounded an already dire financial situation. First quarter 2026 revenue came in at $99 million, down 26% year-over-year and well below analyst forecasts of $137 million. Unit shipments fell from 385,000 cameras in the first quarter of 2025 to 267,000 — a collapse of nearly a third in a single year. The company has already laid off 23% of its workforce and received covenant waivers from its lenders after failing to comply with minimum liquidity requirements. The loan agreements carry cross-default provisions, meaning a breach on one facility could trigger immediate repayment demands across all of them. GoPro has confirmed it is in active discussions with its lenders — Farallon, Wells Fargo, and Yorkville — and has hired advisors to evaluate a potential sale or merger.
No bankruptcy filing has been made, and the company insists it is pursuing its options. But the going-concern warning is not a press release. It is a legal disclosure that management itself cannot guarantee the lights stay on.

The Invisible Collateral Damage

GoPro is the most visible casualty of the memory reallocation, but it is not likely to be the last. The same dynamic is touching car manufacturers, smartphone makers, and laptop producers. Dell's chief financial officer said last November that the company had "never seen costs move at the rate" they currently are. Raspberry Pi raised prices in December and described the situation as painful. Lenovo began stockpiling chips as a precautionary measure. Samsung has raised prices for some memory products by as much as 60%, and analysts at both Samsung and SK Hynix have warned that AI-driven memory shortages could persist through 2027 and beyond.
Consumer RAM prices have more than doubled in some segments. SSDs have followed. The electronics industry is expecting product price increases of 5% to 20% across the board in 2026, with some categories facing steeper increases depending on memory intensity. The inflation being created here is not abstract. It will be felt by anyone buying a new laptop, a phone, a car, or in GoPro's case, a camera.
The irony embedded in all of this is sharp enough to cut. The same AI boom that has driven Alphabet's share price to record levels and handed Berkshire Hathaway a paper gain on a $30 billion tech bet is also the reason a 24-year-old company that essentially invented the action camera is now contemplating bankruptcy. The trillion-dollar AI infrastructure buildout has created enormous wealth for the three memory makers, the chip designers, and the hyperscalers they supply. For the consumer electronics companies on the other side of that supply chain, the math no longer works.

What Comes Next

The immediate questions are fairly clear. For Alphabet, the market will be watching whether the $80 billion in new capital translates into revenue acceleration through the back half of 2026 and into 2027. Morgan Stanley has already projected Alphabet could spend up to $250 billion in capital expenditure in 2027 alone. Sustaining investor confidence through that spending cycle will require the company to keep delivering on cloud growth and enterprise AI adoption at a pace that justifies the dilution.
For the broader hyperscaler complex, the more uncomfortable question is the one BlackRock's strategists have started asking openly: at what point does the debt load accumulated to fund this buildout begin to matter? Interest rates have not come down as much as the market anticipated at the start of the year. The ECB is reportedly considering a June rate hike. The Federal Reserve has adopted an explicitly data-dependent posture for the third quarter. Financing hundreds of billions of dollars of infrastructure construction in that environment is manageable for the Alphabets and Amazons of the world — barely — but it is not without risk.
For GoPro, the window is narrowing. A sale or merger remains the most plausible exit that preserves the brand and its products. Its subscriber base, its library of user-generated content, and its brand recognition among action sports communities still have value to a larger acquirer with a deeper balance sheet. Whether any such acquirer will move quickly enough — and at a price that satisfies GoPro's lenders — is another question entirely.
What is already clear is that the AI arms race has produced a new kind of collateral damage: not the science fiction scenario of machines displacing workers, but something more mundane and more immediate. Companies that make physical things people hold in their hands, that depend on commodity components to do so, and that lack the negotiating power of the hyperscalers — those companies are being squeezed by a dynamic they did not create and cannot escape. They are competing for raw materials against customers who are spending $200 billion a year.
That is not a fair fight, and GoPro is learning what losing it looks like.