Photo: "Memory" by tony_duell, BY via Openverse

The World's Biggest Tech Companies Are Suddenly Bleeding Cash — Memory Chips Are a Big Reason Why

The pressure point in Big Tech's AI boom is not a chatbot, a model, or a data center. It is memory. A shortage in memory chips is one of the biggest reasons AI costs are rising faster than tech's biggest companies expected — and it's now visible on their balance sheets.

Almost four years into the AI boom, the numbers behind Big Tech's promises have turned stark. AI spending among the megacaps is projected to hit $765 billion this year and climb toward nearly $1.2 trillion by 2027, according to Goldman Sachs. Amazon raised its own capital spending forecast to $220 billion on Thursday, the highest of the four hyperscalers, and reported negative free cash flow of $7.6 billion over the trailing 12 months.

That came a day after Meta disclosed a 91% drop in cash generation from a year earlier, and a week after Alphabet said its cash flow turned negative for the first time on record — a milestone for a company that ranks among the most profitable on the planet. Alphabet finance chief Anat Ashkenazi told analysts that free cash flow will "remain under pressure" as the company chases the "AI opportunity."

Memory is the common thread Tesla, Amazon, and Apple all named directly. Demand for AI processors has outpaced what a small group of memory vendors can supply, and the price has followed. Elon Musk called memory pricing "insane" on Tesla's earnings call, while Amazon CEO Andy Jassy said the "inflated price" of memory chips was a direct driver of his company's higher capex guidance. Musk even singled out vendor Micron for giving Tesla "a very significant allocation on reasonable terms" — a notable thank-you in a market he described as running hot. If a shortage severe enough to draw that kind of gratitude from a CEO is also the factor cited by name for Amazon's raised spending guidance, that's a reasonable signal the bottleneck is now a real line item, not background noise, in how these companies plan their AI budgets.

What makes the shortage worth watching beyond hyperscaler spreadsheets is that it's hitting consumers through the same supply chain, just from the other end. Apple, which spends far less on AI infrastructure than its peers, is exposed because memory sits inside nearly every device it sells. The company has already raised prices on Macs and iPads, and many analysts expect iPhone price increases later this year. Apple issued a weaker-than-expected forecast that CEO Tim Cook attributed to "supply constraints," telling analysts, "If you look beyond September, we see the market pricing for memory continuing to increase, which could drive an increasing impact on our business. And we're continuing to evaluate this." Cook is set to step down as CEO on Sept. 1. Read together, Apple's consumer-facing price hikes and the hyperscalers' capex overruns look like two symptoms of one upstream cause — a genuinely different framing than treating "AI is expensive to build" and "gadgets cost more" as unrelated stories, though it's worth being clear this is an inference drawn from the same memory-price dynamic both sides describe, not a connection the companies made explicitly.

Investors, notably, aren't reacting to the AI buildout as a single story either. Tesla shares fell alongside Musk's own warning about memory pricing, Alphabet's fell after its cash flow turned negative for the first time on record, and Meta plummeted on a weak forecast and lingering doubt about its AI monetization plans. Microsoft went the opposite direction, posting its best day on the market since 2008 after pairing strong results with higher capex guidance — a rally that cut its stock's drop for the year to about 7%. Amazon's shares popped too, on the strength of its cloud growth. JPMorgan strategist Dana Harlap asked in a report last week whether the AI trade is really "all one big" trade, noting that markets are becoming "more critical — and more discriminating — across hyperscalers" even when results beat estimates, as Alphabet's did with 82% cloud growth. That split verdict — some capex increases rewarded, others punished, within the same earnings cycle — supports Harlap's read more than it supports any single "AI spending is good" or "AI spending is bad" headline. The market is no longer rewarding AI spending by default. It is asking which companies can turn a memory-constrained buildout into returns before the cash burn becomes the story.