Micron's record AI-memory earnings drive stock surge

Memory has always been a commodity business with a punishing rhythm: glut, price collapse, losses, recovery, repeat. The single most consequential thing in Micron's report is not the revenue number - it is the legal structure underneath it. Micron disclosed 16 multi-year Strategic Customer Agreements that lock in roughly $100 billion in minimum contracted revenue, backed by $22 billion in upfront customer cash ^[1]. These are take-or-pay deals: the buyer must pay for the committed volume whether or not it actually takes delivery, the contracts cannot be canceled, and they carry floor pricing and upfront deposits ^[2].

That is a different animal from how memory has historically been sold. In a normal cycle, demand softens, customers walk, spot prices crater, and the manufacturer eats the downside. Here, HBM3E and HBM4 are already fully booked through calendar 2027 with demand pushing into 2028, and a large slice of that revenue is contractually floored rather than exposed to the spot market ^[2]. The bull thesis is that this converts Micron from a price-taker into something closer to an infrastructure utility with visibility years out. Tech Times framed the contracts explicitly as a signal that the AI memory cycle is breaking from its old pattern ^[1]. Whether the floors hold under stress is the whole debate - but the structure itself is the news.

The number that made analysts blink was the margin. Micron posted a GAAP gross margin of 84.6% (84.9% non-GAAP), up from roughly 39% a year earlier, and guided next quarter to about 86% ^[3]. At that level, a memory maker is now more profitable per dollar of revenue than Nvidia, which runs near 75%, and Meta, near 82% - prompting CNBC to crown Micron tech's new "margin king" ^[4]. The revenue trajectory is just as steep: $41.46 billion in the quarter against $9.3 billion a year earlier, with Q4 guidance of roughly $50 billion versus a $36 billion consensus, and data center revenue alone surging more than sevenfold to $11.5 billion ^[5].

The striking part is the valuation disconnect this opens up. Micron briefly carried a $1.398 trillion market cap, topping Meta's $1.392 trillion ^[6], yet it is throwing off margins that exceed Nvidia's while trading at a fraction of Nvidia's multiple. Bulls read that as a re-rating still in early innings; skeptics read peak margins as a classic late-cycle tell. Either way, a memory company out-earning the two companies it just passed in market value is the kind of statistic that forces a second look at how the market prices the AI supply chain.

To understand why customers are signing non-cancelable checks, look at what the next generation of AI hardware actually requires. Nvidia's Vera Rubin racks are specified for up to 400TB of memory capacity and up to 300TB/s of aggregate bandwidth each ^[7]. Feeding that takes enormous volumes of HBM4 and DRAM, and Micron's 12-high HBM4 delivers over 2.8 TB/s of bandwidth - a 2.3x jump over the prior HBM3E generation ^[7]. In other words, every new Rubin system is a memory-hungry machine, and the amount of high-bandwidth memory per accelerator keeps climbing.

This is the mechanism behind the "memory wall" that Susquehanna's Mehdi Hosseini described, where customers have no choice but to pay a premium ^[2]. Compute has outrun the ability to feed it data, so the scarce input shifts from GPUs to the memory stacked next to them. Micron's own CEO said the company does not yet have line of sight on when supply will catch up to demand ^[5]. Crucially, the same physical scarcity that lets Micron charge record prices also constrains its rivals: SK Hynix and Samsung are certified for Rubin too, but all three are fighting over limited fab capacity, which is why the supply-side discipline analysts cite isn't just a promise - it is a manufacturing reality.

Not everyone buys the structural story, and the skeptics make a specific, mechanical argument rather than vague caution. Memory cycles have historically ended the same way - revenue declines of 25-40% and stock drops of 50-60% - and record margins have often marked the late innings, not the early ones ^[8]. As recently as fiscal 2023, Micron's revenue roughly halved and the company posted a net loss near $5.8 billion ^[9]. The contrarian read is that take-or-pay floors look durable on paper until the cash to honor them has to actually exist.

That is where community sentiment splits sharply from the headline euphoria. On the bullish side, investors are framing the print as a structural AI-infrastructure re-rating and circulating the guidance gap - roughly $50 billion in next-quarter revenue - as proof the market mispriced memory as merely cyclical. The harder-edged skepticism lives on Reddit, where the load-bearing worry is financial plumbing rather than demand: a take-or-pay obligation is only as good as the counterparty's balance sheet, and several voices argue the money to pay for all that booked HBM does not yet exist, with one blunt formulation that bankrupt AI companies don't honor purchasing contracts. A second strand worries about spillover - rising memory prices pushing up RAM costs and forcing consumer-hardware price hikes - which would make the same boom that lifts Micron a tax on everyone downstream. Bank of America's Vivek Arya splits the difference, crediting genuine supply-side discipline for a "more durable cycle" without declaring the cycle dead ^[2]. The unresolved question is whether floor pricing survives a real downturn in AI capex, or whether it is the most elegant version yet of the same trap memory always springs.