Original source

Summary

Investigative deep-dive into how OpenAI signed simultaneous secret deals with Samsung and SK Hynix on October 1, 2025, locking up approximately 900,000 DRAM wafers per month — roughly 40% of global supply. Neither supplier knew about the other’s agreement. The resulting supply shock triggered 156-172% DDR5 price increases within weeks, devastating consumer hardware markets headed into the holiday season.

Key Points

  • OpenAI signed deals with both Samsung and SK Hynix on the same day (Oct 1, 2025) for a combined 900K wafers/month of DRAM capacity.
  • Neither supplier was informed of the other deal — each believed they were fulfilling a large but manageable order. Together, the commitments consumed ~40% of global DRAM output.
  • DDR5 prices surged 156-172% within weeks. A 32GB DDR5 kit went from ~$130 to $330.
  • The market had zero safety stock heading into the crisis. Three factors converged: tariff uncertainty had suppressed advance ordering, falling memory prices in mid-2025 discouraged stockpiling, and stalled equipment transfers to China reduced total available capacity.
  • OpenAI purchased raw wafers rather than finished DRAM modules, suggesting a stockpiling or competitive denial strategy rather than immediate consumption needs.
  • DDR5 lead times extended to 13 months, meaning the supply shock would persist well into 2027.
  • The deal structure resembled a Chokepoint Control play — one buyer cornering a critical input to gain strategic advantage over AI competitors who also need memory capacity.
  • Consumer hardware manufacturers (PC, gaming, smartphones) bore the brunt of the squeeze, as enterprise AI procurement crowded them out.

Newsletter Angles

  • Chokepoint Control in semiconductors: This is a textbook case of a single actor weaponizing supply chain concentration. One company’s procurement strategy inflicted collateral damage across the entire consumer electronics ecosystem. The analytical hook: when does aggressive procurement become market manipulation?
  • Infrastructure Warfare goes corporate: The pattern of chokepoint exploitation previously seen in state actors (oil, shipping lanes, rare earths) is now being replicated by private companies in the AI race. OpenAI treated DRAM the way a state might treat a strategic commodity.
  • The AI bubble’s real-world costs: The speculative AI buildout is now directly raising prices for ordinary consumers. DDR5 for a gaming PC doubled because of AI data center demand that may never materialize at the projected scale.
  • Supply chain fragility as systemic risk: The convergence of tariff chaos, price-driven destocking, and China equipment restrictions created the conditions for a single buyer to break the market. The vulnerability was structural before OpenAI exploited it.

Entities Mentioned

  • OpenAI — orchestrated the dual DRAM deals, central actor in the supply crisis
  • Samsung — one of two DRAM suppliers who signed the Oct 1 deal, unaware of the SK Hynix agreement
  • SK Hynix — second DRAM supplier, also unaware of Samsung’s parallel deal
  • Sam Altman — CEO of OpenAI, architect of the procurement strategy
  • Micron — third major DRAM manufacturer, not party to the deals but affected by market dynamics

Concepts Mentioned

Quotes

Notable: DDR5 prices rose 156% in under a month after the deals were signed, with some SKUs seeing 172% increases.

Notes

  • Source is Moore’s Law Is Dead, a tech YouTube channel / blog with strong semiconductor industry sources. Author “Tom” has a track record of accurate hardware industry reporting but writes with editorial flair.
  • The claim that neither Samsung nor SK Hynix knew about the other’s deal is sourced to industry contacts but not officially confirmed by either company.
  • The “raw wafers, not finished modules” detail is significant — it suggests OpenAI was not buying for immediate use but for strategic positioning or competitive denial.
  • Published November 2025, covering events from October 2025. This is the earliest and most detailed account of the DRAM crisis origin among the sources in this wiki.