Newsletter Draft — The IOU That Broke the Memory Market

Working Title Options

1. “The IOU That Broke the Memory Market” — focuses on the LOI → panic → binding deal cascade
2. “Five Reasons Your RAM Costs $400 (And When It Won’t)” — consumer-facing, listicle structure
3. “The Bluff, the Paradox, and the Strait: Why the RAM Crisis Won’t End” — literary, evocative
4. “How One Man’s Fake Purchase Order Reshaped a Global Supply Chain” — narrative, profiles Sam Altman as the catalyst
5. “RAMmageddon: A Perfect Storm in Five Acts” — dramatic framing

Recommended: Title 1 or 4 for the newsletter’s analytical audience. Title 2 if optimizing for click-through.

Core Thesis (One Paragraph)

In October 2025, Sam Altman signed a non-binding letter of intent for 40% of the world’s memory supply. He wasn’t legally committed to a single dollar. But the panic his bluff triggered caused Apple, Amazon, Microsoft, and AMD to fly executives to South Korea and lock in real, binding multi-year deals at peak prices. DDR5 went from $120 to $400. Then five independent forces — Jevons Paradox neutralizing a 6x efficiency breakthrough, a helium shortage from the Strait of Hormuz threatening the factories that make 70% of global DRAM, a Samsung strike over windfall profits not reaching workers, half of US data centers getting cancelled, and Chinese competitors still two years from volume production — created a perfect storm that keeps prices frozen even as the original demand signal collapses. OpenAI cut spending in half. Stargate is dead. Sora is dead. And DDR5 is still $400.

Proposed Structure

Act 1: The Price That Won’t Move (~300 words)

Hook: Open with the disconnect. DDR5 was $120. Now it’s $400. Every signal says it should be falling — OpenAI is scaling back, data centers are cancelling, Google found a way to use 6x less memory, Chinese stockpilers are panic-selling. So why is the price still $400?

Data to use:

• DDR5 32GB kit: ~$120 (mid-2025) → $400+ (April 2026)
• HP: memory now 35% of PC build costs (was 15-18%)
• Gartner: sub-$500 laptops may become “financially unviable” within 2 years
• Micron exited its consumer “Crucial” brand entirely
• IDC: phone prices to rise 8-9%

Act 2: The Bluff (~500 words)

The LOI mechanism: Explain what Sam Altman actually did. Not a purchase — a letter of intent. Non-binding. No legal obligation. But the market impact was identical to real demand.

Key details:

• Oct 1, 2025: Simultaneous deals with Samsung AND SK Hynix for 900K wafers/month (~40% of global DRAM)
• Neither supplier knew about the other’s deal
• OpenAI purchased raw wafers, not finished modules — stockpiling/competitive denial, not consumption
• The LOIs triggered 156-172% price increases within weeks (MLID data)
• Cumulative peak: ~700% increase over the following year (Telegraph)

The pivot: OpenAI’s subsequent retreat — $1.4T → $600B spending (57% cut), Stargate cancelled (Oracle financing collapse), Sora shuttered, $8B burn on $13.1B revenue, acquired a podcast for $150M+ while cutting compute

The Ben Thompson quote (attributed via Big A paraphrase of Stratechery): “OpenAI might be the short bus at the end of the rainbow. There’s supposed to be a pot of gold there, but it never quite seems to materialize.”

Act 3: The Panic (~400 words)

The cascade: How the LOI triggered real binding deals by panicked buyers.

The Apple scene: Executives booking extended hotel stays in Hwaseong, South Korea, near Samsung facilities. Negotiating 2-3 year LTAs. LPDDR5X at $70/unit (230% increase). Samsung becomes Apple’s largest DRAM supplier (60-70% of iPhone 17).

The parade: AMD’s Lisa Su flying to South Korea, meeting Samsung Vice Chairman Lee Jae-yong personally. Signing MOU for HBM4 + DDR5. Amazon, Microsoft doing the same.

The analytical point: The non-binding signal created self-fulfilling binding demand. Even if OpenAI never draws down a single wafer from the LOIs, the contracts Apple/Amazon/Microsoft signed are real and locked in for years. The fake demand became structurally real.

Act 4: The Five Forces (~800 words)

Why the price is stuck — even though the original demand signal is collapsing.

Force 1: Jevons Paradox (~200 words)

• Google’s TurboQuant: 6x compression of LLM context windows (Mar 24, 2026)
• Stocks dropped 5-6%. Brief panic. Then the market applied Jevons Paradox.
• William Stanley Jevons (1865): more-efficient steam engines → more factories, not less coal
• Modern analogy: video compression didn’t reduce bandwidth — we just stream more, at higher quality, on more devices
• AI version: 6x less memory per context window = 6x longer context windows + more AI deployments = same or more memory
• Result: contract prices didn’t budge. Only speculative retail layer corrected.

Force 2: Helium from the Strait of Hormuz (~200 words)

• Qatar’s Ras Laffan: ~33% of global helium, offline since March 2026 (Iranian strikes + Strait closure)
• South Korea: 65% of helium from Qatar. Samsung + SK Hynix = 70% of global DRAM.
• Ultra-high-purity helium is irreplaceable in chip fab: cooling, lithography, etching, CVD, EUV
• South Korean fab helium reserves: ~6 months (through June 2026)
• If helium runs out → Samsung/SK Hynix prioritize high-margin AI memory (HBM) over consumer DDR5
• LNG supply also disrupted → compound energy + process gas squeeze

Force 3: Samsung strike (~150 words)

• 90K unionized workers, 18-day strike vote from May 21
• Could halt 50% of Pyeongtaek production
• Samsung profits 8x year-over-year; workers want 7% raise + profit-sharing
• Tesla poaching chip designers
• One of three companies controlling 95% of memory → massive labor leverage

Force 4: Data center bottleneck (~150 words)

• 50% of 2026 US data centers delayed/cancelled (Sightline Climate/Bloomberg)
• Only 4GW of 12GW actually under construction
• Bottleneck: electrical components from China (<10% of costs but gating)
• Odd paradox: the demand collapse isn’t releasing supply because contracts are already locked

Force 5: Chinese factories still two years away (~100 words)

• CXMT: Shanghai mega-fab (2-3x Hefei), targeting 2027
• YMTC: 3rd Wuhan fab, also 2027, pivoting from NAND to DRAM
• Both filing IPOs to fund the expansion
• Until Chinese volume comes online, the big three face no competitive pressure to lower prices

Act 5: The Timeline and the Bigger Picture (~400 words)

When does it end?

• Now → end 2026: crisis pricing. $400 RAM. Expensive phones, laptops, consumer electronics.
• Mid 2027: gradual relief as new factory supply comes online
• 2028: potential glut — Chinese factories enter volume, Samsung/SK Hynix new capacity also arrives

The bigger picture (this is where the newsletter earns its keep):

• One man’s non-binding IOU reshaped a global supply chain. What does this tell us?
• The DRAM market’s vulnerability: three companies control 95% of supply. A single credible bluff can break the entire market.
• Chokepoint Control isn’t just geographic — it’s structural. OpenAI exploited a market chokepoint the way Iran exploits a geographic one.
• The AI hype cycle has real-world costs. Speculative demand projections → real supply lockups → real consumer price increases → real economic damage. Even when the speculation collapses, the damage is locked in by binding contracts.
• Jevons Paradox as the ultimate bull case for resource demand: efficiency gains don’t reduce consumption, they expand markets. Applies to AI memory, energy, bandwidth, and (potentially) labor.

Key Quotes to Use

“If you want to sell your shares, I’ll find you a buyer. Enough.” — Sam Altman to investor questioning $1.4T spending commitment

“If one piece of your supply chain is delayed, then your whole project can’t deliver.” — Andrew Likens, Crusoe Energy

“HBM is the gating component in modern AI accelerators. If supply tightens at the memory level, the effects cascade almost immediately into server builds and data-center deployments.” — Kevin Hein, Tirias Research

“It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth.” — William Stanley Jevons, 1865

What’s Missing (Source Gaps)

• Samsung official earnings: Need to verify the “8x profit growth” claim against Samsung’s actual Q1 2026 report
• Contract price data: TrendForce reports contract stability, but specific contract price levels are not public
• OpenAI LOI status: Are the LOIs being drawn down, renegotiated, or quietly abandoned? No reporting on this.
• Helium recovery timeline: When does Strait of Hormuz reopen for helium shipments? Depends on Iran situation.
• Consumer impact data: Beyond HP and Gartner projections, are there actual sales figures showing consumer electronics demand destruction?

Tone Notes

• This piece should be analytical, not polemic. The tone is “here’s what’s actually happening and why” — not “AI bad” or “gamers sad.”
• Use Jevons Paradox as the intellectual anchor — it’s the most counterintuitive and most revealing of the five forces.
• The LOI → binding deal cascade is the narrative engine. It’s a story about how signals become reality in concentrated markets.
• Avoid the gamer framing (that’s Big A’s audience, not yours). Frame it as a supply chain / market structure story with consumer impact, not a consumer grievance piece.