Micron just posted $23.86 billion in quarterly revenue. Earnings per share came in at $12.20 adjusted — against Wall Street's $9.31 expectation. Profits rocketed 682% year-over-year.
The stock barely budged. And that reaction tells you everything you need to know about GPU prices for the rest of this year.
The reason analysts shrugged isn't that the numbers were bad. They were extraordinary. The issue is that Micron's CEO Sanjay Mehrotra went on CNBC the next morning and said something quietly catastrophic for anyone waiting to buy an RTX 50-series card at a sane price: "We are only able to supply, for our key customers in the midterm, about 50% to two-thirds of their requirements."
That's the entire 2026 situation in one sentence.
What Actually Happened With Micron's Earnings
Micron's Q2 FY2026 results, reported March 18th, weren't just a beat — they were a structural signal. Revenue hit $23.86 billion, nearly 3x the year-ago $8.05 billion. GAAP profit reached $13.79 billion, which is a 771% year-over-year jump. The company's entire 2026 high-bandwidth memory output is sold out under binding long-term contracts.
DRAM spot prices are up roughly 700% year-over-year.
That last number is the one that should make you close the "RTX 5080 price drop predictions" tab you have open right now.
Note
What is HBM? High-Bandwidth Memory is the stacked DRAM that sits directly on AI accelerators like NVIDIA's H100 and Blackwell GPUs. It's different from the GDDR memory in gaming cards — but the same fabs make both. When AI demand eats HBM capacity, GDDR availability shrinks too.
The HBM Crunch Is Structural, Not a Blip
Here's what people get wrong about the current GPU shortage: they treat it like a launch-day supply crunch that resolves itself in three months. It's not.
SK Hynix and Micron — the two dominant HBM suppliers — are both sold out through the end of 2026 under binding contracts. Samsung is racing to complete HBM4 validation, expected sometime in Q2 2026, but even optimistic timelines put meaningful volume relief in late 2026 at the earliest. New fab capacity takes 18 to 24 months to come online, and the $50+ billion combined investment the three memory giants announced won't translate to actual chips until 2027 or 2028.
Micron confirmed it's planning large-scale HBM and capacity expansions — for 2027 and 2028. Not this year. Not next quarter.
And NVIDIA locked down this supply deliberately. Analysis of their SEC filings shows a strategy some analysts are calling "strategic capacity capture" — NVIDIA secured long-term HBM agreements that effectively starved AMD and Intel of allocation for the current cycle. Whether that's genius supply chain management or anti-competitive behavior is a separate argument. The practical result is the same: the memory that would otherwise flow into consumer GPUs is committed to data center accelerators for AI training and inference.
Warning
The 30-40% production cut is already priced in. NVIDIA officially confirmed it's cutting RTX 50 Series production by 30-40% due to the memory crisis, and that supply impact extends through fiscal 2027. Waiting for a price correction in 2026 means waiting for something NVIDIA has explicitly said won't happen.
What This Means for RTX 50 Series Pricing
The RTX 5090 launched at $1,999 MSRP in January 2025. Custom AIB models currently sell for $2,900 to $3,400 at mid-range and well past $5,000 for premium variants. Founders Edition cards sit closest to MSRP but barely exist in the wild.
The RTX 5080 situation is similar. The 5070 Ti has been marginally more findable, but only because NVIDIA explicitly deprioritized everything above the 5060 lineup — they're pushing production resources toward the entry-level segment where volume is higher. The flagship cards are effectively phantoms.
Tom's Guide reported in February that NVIDIA won't release a new gaming GPU "for the first year in three decades" due to the RAM shortage. That's not spin. That's a product cadence disruption caused directly by memory allocation constraints.
The GPU Poet market report from March 1st puts the RTX 5090 still sitting 40% above MSRP on secondary markets. The RTX 5060 has dipped below MSRP in spots — exactly what you'd expect given NVIDIA's production prioritization. But the cards most enthusiasts actually want? Still wildly overpriced and often unavailable at any price.
So What Do You Actually Buy Right Now?
Two real options exist in March 2026, and both require accepting something you probably don't want to accept.
Option 1: RTX 40 Series (if you can find it at sane prices)
The RTX 4070 Super and 4080 Super represent the last generation where supply was mostly normalized before the memory crisis hit. The 4080 Super retails around $959 for reference models, which looks almost reasonable compared to the RTX 5080 situation — though that price has drifted upward as used market demand tightened.
The problem is that the RTX 40 series is already two years old at this point and you're buying into a generation with no upgrade path. DLSS 4 is a real differentiator, though, and if you're a creative professional or running AI workloads on your local machine, the NVIDIA ecosystem advantages matter more than they do for pure gaming.
Option 2: RX 9070 XT — the actual best value on the market
The AMD RX 9070 XT launched at $599 MSRP in March 2025 and is currently retailing between $729 and $789 depending on the variant. The ASRock Challenger sits at $729. The Gigabyte Gaming OC came in at $739 to $749 on Amazon as recently as yesterday.
And crucially — it's actually in stock.
The performance numbers are harder to dismiss than AMD's previous generations. The 9070 XT scores 29,992 in 3DMark synthetic benchmarks against the RTX 4080's 28,063. That's a 6.9% advantage, at a retail price that's 144% lower. The 9070 XT uses 304W versus the 4080's 320W — more efficient too.
Against the RTX 4080 Super, the comparison is even more lopsided in AMD's favor. XDA Developers ran the full comparison and concluded the 9070 XT is simply the GPU to buy in this price class, despite the Nvidia card nominally sitting in a higher performance tier on paper.
Tip
The 9070 XT dodges the HBM crisis entirely. It uses GDDR6 memory, not HBM. While NVIDIA's Blackwell cards are memory-constrained by the AI supply war, AMD's RDNA 4 architecture made a deliberate choice to stay on GDDR6 for consumer cards — which means supply is comparatively unconstrained. That's also why you can actually buy one.
The 2027 Argument
If neither option excites you — and I get that, neither is perfect — the actual patience play is waiting until mid-to-late 2027.
HBM4 validation completes sometime this quarter. Mass production scales through 2026. New fab capacity from the combined $50B+ investment starts coming online in 2027. NVIDIA's Rubin platform, which will use HBM4, is the next major architecture — and if HBM supply genuinely expands before the Rubin launch, you might finally see consumer card pricing that resembles reality.
Might. There's a real scenario where AI infrastructure demand just keeps absorbing every memory die that comes out of these new fabs, and consumer GPU pricing stays elevated indefinitely. Micron's CEO didn't sound like a man who expected the crunch to end soon. He sounded like a man who was very happy about his gross margins.
The $50B capex expansion wasn't charity toward PC gamers. It was investment to capture more of the most profitable memory market in semiconductor history.
The Honest Verdict
Anyone who told you RTX 50 series prices would normalize in 2026 was wrong, and Micron's Q2 report is the final confirmation of that. $23.86 billion in quarterly revenue built on sold-out HBM contracts doesn't create conditions for GPU price relief. It creates conditions for another quarterly beat.
Buy the RX 9070 XT now if you need a GPU upgrade. It's genuinely excellent, it's available, and it will hold performance value well into 2027 when the market might actually reset. Or buy into RTX 40 series if NVIDIA's software ecosystem is non-negotiable for your workflow.
Waiting specifically for RTX 50 series prices to fall to something reasonable before 2027 is a bet against structural supply economics. And the memory industry just reported its most profitable quarter in history. That's not the earnings report of an industry about to drop prices.
See also: Best GPU Deals for Local LLMs →
Also worth reading: NVIDIA's 4B parameter model — what it means for local AI inference