For decades, consumer memory was a dependable, if unglamorous, business. Buy DRAM, sell DRAM, ride the cycles, repeat. Micron's Crucial brand — launched in 1996 to serve PC enthusiasts building their own machines — was a beloved institution. Reliable, well-priced, and everywhere.
Then AI broke everything.
A typical server has 32GB to 128GB of memory. An AI server can have up to 1TB. Do that math across thousands of hyperscale data centers, and you start to see the problem. Training and inference workloads for large language models — think OpenAI, Meta, and Anthropic — require memory-to-compute ratios that dwarf those of traditional cloud or gaming systems. Memory wasn't just in demand. It was being consumed at a pace the industry had never seen.
Tech giants like Microsoft, Google, and ByteDance were racing to secure memory supplies from chipmakers including Micron, Samsung, and SK Hynix amid tight supply. SK Hynix reportedly locked in multi-year wafer agreements with AI titans, effectively ring-fencing supply. The consumer market wasn't just deprioritized. It was being quietly starved.
The day Crucial died
On December 3, 2025, Micron announced it would completely exit the consumer memory market, discontinuing its 29-year-old Crucial brand by February 2026.
The language from Micron's leadership was blunt.
The AI-driven growth in the data center has led to a surge in demand for memory and storage. Micron has made the difficult decision to exit the Crucial consumer business in order to improve supply and support for our larger, strategic customers in faster-growing segments.
-- Sumit Sadana, Micron's chief business officer
As expected, Micron was abandoning its consumer business due to reallocation of its 3D NAND and DRAM output and production capacity to enterprise-grade SSDs, high-bandwidth memory for AI accelerators, and server-grade memory modules.
Translation: every wafer going to a DDR5 kit for a PC builder was a wafer not going to a hyperscaler paying vastly more. The math was brutal and simple.
What this did to your wallet
The fallout for everyday consumers was severe and measurable. DRAM spot prices increased 172% year-over-year as of Q3 2025, with retail prices for 32GB DDR5 modules jumping 163–619% in global markets.
The RAM you were buying in 2024 for under $100 doubled, and in some cases tripled in price within 12 months. One of the most bizarre data points from this crisis? DDR4 — old memory — started costing more per gigabit than cutting-edge HBM3e AI chips: $2.10 per gigabit vs $1.70. That's a market inversion that signals something deeply broken in supply and demand dynamics.
The high cost of memory sideswiped the technology industry, causing server vendors to admit their quotes are guesstimates and depressing sales of PCs and smartphones. Nobody was immune: Microsoft used the RAM panic as cover for fixing Windows 11's memory gluttony, and Sony suspended orders for compact flash and SD cards because it couldn't buy the chips to build them.
Meanwhile, Micron itself was thriving. The company reported record fiscal 2025 revenue of $37.38 billion, representing nearly 50% year-over-year growth driven primarily by data center and AI applications, which accounted for 56% of total revenue.
Enter Google, stage left
Then something unexpected happened. On March 24, 2026, Google Research quietly posted a paper. TurboQuant is a compression algorithm that modifies the way data is stored and processed, using a two-step technique: first, it simplifies activation vectors, reducing the complexity of instructions that an AI needs; second, it eliminates errors through the QJL technique, which allows each vector address to require significantly less memory space.
The headline result: a 70B parameter model that previously required 80GB of VRAM can now run on just 12GB. That's not an incremental improvement. That's a structural shift in the economics of AI infrastructure.
Memory maker stocks fell hard. Samsung dropped 6% and nearly 5%, respectively, in South Korea. Japanese flash memory company Kioxia dropped nearly 6%. Micron and WD also fell in the US.
And for consumers? DDR5 prices saw their first notable decline in months. Some 32GB kits fell $40 to $100 at major retailers within days of the announcement. It wasn't a crash. But it was the first breath of relief in months.
The truth about TurboQuant's limits
Here's where we have to be honest, because the hype has outrun the reality.
TurboQuant is still, as of today, a research paper. Google has not released official code, and TurboQuant is still a research paper at this stage. Independent developers have already created early implementations for PyTorch, Apple's MLX framework, and llama.cpp, but Google's official release is expected around mid-2026.
More importantly, TurboQuant is believed to have an effect on the stock market, lowering the price of memory providers as traders look to the release as a sign that less memory will be needed — though analysts note this may be incorrect, given Jevons' Paradox. Jevons' Paradox is the inconvenient economic truth that making something more efficient tends to cause people to use more of it, not less. If running AI gets cheaper, more AI gets run. Demand might not fall — it might explode further.
Analyst firm TrendForce predicts TurboQuant will lower the cost of AI infrastructure, and by doing so "spark massive long-sequence application demand, comprehensively driving structural growth and specification upgrades for high-bandwidth, main, and flash memory across cloud and edge platforms." In other words: TurboQuant might actually create more memory demand over the long run, not less.
And there are supply-side complications that algorithms can't fix. The war in the Persian Gulf has damaged the supply chain for helium — a vital component in semiconductor production — which could mean chipmakers can't make all the RAM they assumed would inflate their revenue and profits.
So can Micron come back to consumers?
This is the speculative part — and it's complicated.
The financial case against it is powerful. Micron posted a blowout second quarter for fiscal year 2026, reporting quarterly revenue of $23.86 billion and adjusted earnings per share of $12.20, effectively tripling its performance from the same period two years prior. The enterprise pivot is working spectacularly. Why would Micron voluntarily step back into a low-margin, price-volatile consumer market?
The infrastructure case is even more daunting. Micron's VP of Marketing emphasized that the ongoing memory shortage is unlikely to ease before 2028, given the complexity of fab expansions and certification processes. You can't just flip a switch to start making consumer RAM again. Fab retooling takes time, certifications take time, and every committed wafer is a foregone enterprise dollar.
But the structural argument for a return exists. TurboQuant — if it delivers even half of its promised efficiency gains at scale — genuinely reduces the memory per AI inference workload. If hyperscalers need fewer modules per server, wafer capacity frees up. Micron still maintains significant market share through OEM partnerships, supplying memory modules — such as LPDDR5 — to PC makers like Dell and ASUS for integration into their own products. The supply chain relationships haven't been entirely severed.
There's also a competitive pressure angle. Kingston, ADATA, and other consumer-focused brands are now absorbing the demand Micron abandoned. Every month Micron stays out, those brands entrench themselves further. Returning to a market you've exited is always harder than staying in it.
Our honest verdict
The RAM prices dropping right now are real but fragile. They're driven partly by a research paper that hasn't shipped production code, partly by cooling hyperscaler expectations (OpenAI pulling back from some procurement commitments has also played a role), and partly by natural market recalibration after an extreme spike. Analysts warn that the underlying supply shortage driven by AI training demand remains, and prices are still far above their historic lows.
Micron could re-enter the consumer market — but not through Crucial. That brand is dead for good (RIP) and the retail relationships have wound down. Any return would be quieter: OEM channels, B2B partnerships, or a new consumer sub-brand launched if and when capacity genuinely allows.
The most likely path? Micron watches TurboQuant's real-world deployment carefully. If inference efficiency gains materialize at scale and hyperscaler demand pressure genuinely eases, the calculus changes. If Jevons' Paradox wins instead — if efficient AI just means more AI — then consumer memory remains a luxury Micron can't afford to produce.
The RAM war isn't over. It's just entering a new, stranger phase. And the consumers who got caught in the crossfire are watching very closely.