The timeline on this is genuinely insane. October 2025: Sam Altman flies to Seoul and signs simultaneous deals with Samsung and SK Hynix for 900,000 DRAM wafers per month. That's 40% of global supply. Neither company knew the other was signing a near-identical commitment at the same time. Those deals were letters of intent. Non-binding. No RAM actually changed hands. But the market treated them as gospel. Contract DRAM prices jumped 171%. A 64GB DDR5 kit went from $190 to $700 in three months. December 2025: Micron kills Crucial, its 29-year-old consumer memory brand, to reallocate every wafer to AI and enterprise customers. The company explicitly said it was exiting consumer memory to "improve supply and support for our larger, strategic customers in faster-growing segments." Translation: the AI demand signal was so loud that selling RAM to PC builders stopped making financial sense. March 2026: Google publishes TurboQuant, a compression algorithm that reduces AI memory requirements by 6x with zero accuracy loss. Cloudflare's CEO called it "Google's DeepSeek." The entire thesis that AI would consume infinite memory forever just got a six-month expiration date on it. Same month: OpenAI and Oracle cancel the Abilene Stargate expansion. The $500 billion data center vision that justified the RAM deals couldn't survive its own financing terms. Bloomberg attributed the collapse partly to OpenAI's "often-changing demand forecasting." MU is now down ~33% from its post-earnings high. Revenue up 196% year over year, EPS up 682%, and the stock is in freefall because the company restructured its entire business around a demand signal that came from non-binding letters and is now being compressed out of existence by a research paper. Micron bet the consumer division on Sam Altman's signature. The signature was worth exactly what the paper said: nothing binding.
This chart-style infographic is DRAMeXchange’s DRAM market index (DXI) visualization showing a clear upward trend in the memory price/index over time. It directly supports the tweet’s claim about a dramatic DRAM price rally by visualizing the steep rise in DRAM pricing that underlies the described supply squeeze and market volatility.
Source: DRAMeXchange (TrendForce)
Research Brief
What our analysis found
On October 1, 2025, OpenAI CEO Sam Altman traveled to Seoul and signed letters of intent with both Samsung Group and SK Group to secure memory chips for the Stargate AI data-center initiative. Press reports described OpenAI's projected demand as up to 900,000 DRAM wafer starts per month, a staggering figure that rippled through global memory markets. Industry tracker CTEE reported that contract DRAM prices surged approximately 171.8% year-over-year by Q3 2025, and retail DDR5 kits that had sold for under $200 were listed at $700 or more by early 2026, according to PCPartPicker aggregations and tech outlet reporting.
The downstream consequences were swift. On December 3, 2025, Micron announced it would shutter Crucial, its 29-year-old consumer memory brand, to redirect all wafer capacity toward enterprise and AI customers, specifically citing faster-growing strategic segments. When Micron reported fiscal Q2 FY2026 results on March 18, 2026, the numbers were extraordinary: revenue of $23.86 billion (up roughly 196% year-over-year) and adjusted EPS of approximately $12.20 (up about 682%). Yet the stock had already begun falling sharply from its post-earnings peak.
Two developments in March 2026 undercut the demand thesis that had driven these moves. Google Research published TurboQuant on March 24, a quantization method that compresses LLM key-value caches to roughly 3 bits, reducing KV-cache memory requirements by approximately 6× with no reported accuracy loss. Days earlier, Bloomberg reported that Oracle and OpenAI had shelved the planned expansion of the Abilene, Texas Stargate campus, citing financing difficulties and OpenAI's "often-changing demand forecasting." Together, these events raised serious questions about whether the AI memory supercycle Micron had restructured around would materialize at the scale originally projected.
Fact Check
Evidence from both sides
Supporting Evidence
LOIs signed with Samsung and SK Group in Seoul
Multiple contemporaneous reports from TechCrunch, Yonhap, Korea Times, and Forbes confirm that on October 1, 2025, OpenAI signed letters of intent or MOUs with both Samsung Group and SK Group during Altman's Seoul visit, and that the stated DRAM demand projection reached up to approximately 900,000 wafer starts per month.
Deals were non-binding letters of intent
Press coverage from Korea Times and other outlets consistently described the agreements using LOI and MOU language, confirming the tweet's characterization that these were preliminary, non-binding commitments rather than finalized supply contracts.
Contract DRAM prices surged approximately 171.8%
Industry data from CTEE (Commercial Times Economic & Financial News), widely cited by outlets including TweakTown, reported a year-over-year contract DRAM price increase of roughly 171.8% as of Q3 2025, supporting the tweet's price-spike claim.
Micron exited Crucial consumer memory
Micron's own investor press release dated December 3, 2025, confirms the company announced it would wind down Crucial consumer shipments by approximately February 2026, explicitly stating the move was to prioritize enterprise, HBM, and AI customers in faster-growing strategic segments.
Micron's fiscal Q2 FY2026 financials match the tweet
Micron's earnings release on March 18, 2026, reported revenue of $23.86 billion (approximately +196% YoY) and adjusted EPS of roughly $12.20 (approximately +682% YoY), closely matching the tweet's cited figures.
Abilene Stargate expansion was shelved
Bloomberg reported in early March 2026 that Oracle and OpenAI decided not to proceed with a planned expansion lease at the Abilene, Texas Stargate campus, citing financing challenges and OpenAI's often-changing demand forecasting as contributing factors.
TurboQuant achieves approximately 6× memory reduction
Google Research published TurboQuant on March 24, 2026, demonstrating quantization methods that compress LLM KV caches to roughly 3 bits, achieving approximately 6× memory reduction with no reported downstream accuracy loss in their benchmarks, confirming the tweet's description of the algorithm's impact.
Contradicting Evidence
The 900,000 wafer figure was a projected multi-year target, not an immediate commitment
Press reports described the 900,000 wafer-starts-per-month figure as a long-term projection of OpenAI's potential demand, not a single immediate order. The tweet presents it as though Altman locked up 40% of global supply in one stroke, but the LOIs outlined aspirational, phased capacity needs over multiple years.
Neither company knew is unverified
The tweet claims Samsung and SK Hynix were unaware the other was signing a near-identical deal simultaneously. No sourced reporting has confirmed this mutual ignorance; given that Samsung and SK Group are major Korean conglomerates operating in overlapping circles and that the Seoul visit was publicly reported, the secrecy claim appears to be speculation rather than established fact.
The 171.8% DRAM price surge was year-over-year, not solely caused by the LOIs
The CTEE-reported price increase reflected a full year-over-year comparison through Q3 2025, encompassing many demand and supply factors across the broader DRAM market, including HBM demand from multiple AI companies and cyclical industry recovery. Attributing the entire price spike to the OpenAI LOIs oversimplifies a complex market dynamic.
The $190-to-$700 retail price jump is selectively sourced
While some retail DDR5 kit listings did reach elevated prices, the specific trajectory from $190 to $700 in three months appears drawn from individual product listings or worst-case examples rather than a market-wide average. Retail pricing varied significantly by region, retailer, and kit specification.
TurboQuant targets KV-cache memory specifically, not all AI DRAM demand
Google's TurboQuant paper addresses key-value cache compression for large language model inference, a specific subset of AI memory usage. The tweet implies it undermines the entire thesis that AI consumes vast amounts of memory, but HBM demand for model training, parameter storage, and non-LLM AI workloads remains unaffected by KV-cache quantization techniques.
The Abilene cancellation was an expansion, not the full Stargate project
Bloomberg's reporting specified that Oracle and OpenAI shelved a planned expansion lease at the Abilene campus, not the entire Stargate initiative. The tweet frames this as the collapse of the $500 billion vision, but other Stargate facilities and partnerships continued, and Oracle later clarified aspects of the reporting.
Micron's stock decline reflects multiple factors beyond the demand-signal narrative
While MU did fall roughly 33% from its post-earnings high, semiconductor stocks broadly experienced multiple headwinds in early 2026 including tariff uncertainty, broader market rotation, and valuation compression across the AI supply chain. Attributing the decline solely to the unwinding of Altman's LOIs and TurboQuant conflates correlation with causation.
Report an Issue
Found something wrong with this article? Let us know and we'll look into it.