Push Comes to Shove on the Global Memory Crisis. ARD #108

Today’s theme: push comes to shove on the global memory crisis — the supply initiatives are running far slower than the demand-shortage consequences now building through the whole electronics economy. Three events I’d like to dig into for the AI Tech Wave — each with my Take first, then my Overall Take.

(1) Memory-Chip Price Impact on Smaller Tech — An ‘Existential Crisis’

MP TAKE: Despite general awareness and digestion of the big-company price increases by Apple, Microsoft, Sony and others, it’s still not appreciated at a mainstream level just how broad and deep the impact is on electronics of all types — far beyond PCs, smartphones and game consoles. And how existential it is for a whole host of devices mainstream audiences take for granted — CNBC frames the squeeze as an outright ‘existential crisis’ for the smaller players who can’t absorb or pass through the cost the way Apple and Microsoft can.

Think about the $20-30 gadgets, adapters and gizmos you buy on Amazon without a second thought — often made by companies of a dozen people, sourcing one or two components needing 8 or 16 gigs of DRAM. When memory goes from ~5% of the bill of materials to 40-60%, entire categories simply become unviable. Even a successful, well-funded outfit like Valve sees its Steam Machine console jump from a planned ~$500-600 to $1,100+ without the controller — and there’s nothing they can do, because they’re not even on the customer list of the three trillion-dollar memory makers (Micron, SK Hynix, Samsung), each sold out for four-to-five years at 85%+ gross margins.

The scale of the move is the part that hasn’t sunk in: Jefferies is warning memory prices surge ~50% in Q3 2026 and another ~40% in Q4 — with no relief until 2028. To widen the lens: we’ll only miss these products when they’re gone — and the forces driving toward that outcome are gathering strength, not easing. This is RAMageddon not just pushing, but shoving many smaller tech companies — with their plethora of electronic products — out of business. And that’s not yet understood in the market.

Sources, in narrative order: CNBC — The memory shortage shaking Apple and Microsoft is an ‘existential crisis’ for smaller players. WCCFTech — Jefferies warns memory prices will surge 50% in Q3 2026 and another 40% in Q4, with no relief until 2028. For longtime readers: ‘Memory Chip Supply Constraints and Price Hikes a major AI headwind ahead’ in AI-RTZ #951.

(2) Big Memory Oligopolies Promise Bigger Investments — A Mixed Bag

MP TAKE: Although this announcement looks positive on the surface — Samsung and SK Hynix planning a massive chipmaking expansion (the FT pegs it at ~$600 billion; Korea’s own plan for a southwestern semiconductor hub runs in the $500-600 billion range) — some chunk of this is allocated to AI data centers in South Korea, which end up being an additional drag on memory supply rather than relief. So this announcement, while positive, is a mixed bag.

There’s also nothing to guarantee the new capacity flows to the side where relief is actually needed. It can just as easily go to the higher-margin ‘mainframe AI’ side of the market — much more lucrative for Samsung and SK Hynix — rather than the local-memory side, the other side of the AI barbell, the gadget side. Also, the timing here is measured in years — so no immediate relief for any of the parties around the world. To widen the lens: this is the supply side of the same gating factor we flagged when SK Hynix listed and when we called semiconductors a key AI bottleneck — capacity is coming, but it lands on a multi-year lag while demand compounds now.

There’s a lot of pushing out of investment dollars — shoving even, especially into sovereign AI data centers. But the question is the time it’ll take to make a real difference. Good to see nevertheless — it’s the right direction, just not the right timeline for the near-term squeeze.

Sources, in narrative order: FT — Samsung and SK Hynix plan $600 billion chipmaking expansion. The Korea Times — Samsung, SK Hynix pledge for southwestern chip hub. For longtime readers, in narrative order: ‘Semiconductors Remain a Key AI Gating Factor’ in ARD #104; and ‘SK Hynix Ready for its AI Close-up’ (2024) in AI-RTZ #557.

(3) Big Customers Look for US-Government Help on China Memory

MP TAKE: China memory supply is, at best, negotiating leverage for big US customers against the three memory oligopolies — SK Hynix, Samsung and Micron. Ming-Chi Kuo’s read is that the supply-demand gap keeps widening through 2027 — and that’s the real reason Apple is lobbying the White House (now joined by Microsoft and others) to keep China memory maker CXMT off the Entity List, as a source of short-term DRAM and NAND for the local-AI side. You can see the demand pull on the other side too: Reuters reports CXMT just won a ~$3 billion memory-supply deal with Tencent.

But here’s the rub: Micron’s lobbying leverage in the US remains a significant bipartisan barrier for most US big-tech customers to navigate around — right now even Apple can’t get those approvals. To widen the lens: this is supply-chain politics meeting the model-gating politics we covered yesterday — the same US-vs-China board, different square — and it’s not clear the administration, fixated of late on gating the biggest Anthropic/OpenAI models, has its head around the memory side yet. The second Xi/Trump meeting at the White House this September may move the issue in the right direction, opening more memory-supply options for US buyers.

US companies need to push and shove harder on US regulators to open up sourcing opportunities in China. It’s the relatively quickest way to a solution that provides some timely relief to RAMageddon.

Sources, in narrative order: Ming-Chi Kuo — The memory supply-demand gap will keep widening through 2027 (the real reason Apple is lobbying to keep CXMT off the Entity List). Reuters — China’s CXMT wins $3 billion memory-supply deal with Tencent. For longtime readers: ‘US hamstrung on memory-chip supply via China’ in AI-RTZ #1128.

MP OVERALL TAKE

This is a broader issue that, over time, will become as politically relevant in the mainstream as food and gas prices are today. Politicians will start paying a lot more attention to electronics prices and shortages — sooner than currently envisioned — once the squeeze shows up in the checkout price of everyday devices, not just flagship phones and consoles. Come the holiday period, when people find that the electronics they’re buying are going up in price or disappearing, the pressure will build.

But again, the solutions will take years and billions — along with the time needed for true regulatory responses that actually move the needle. It’s the mirror image of the build-out euphoria — the same multi-trillion AI-infrastructure wave that’s lifting the ‘mainframe AI’ side is draining the memory the mainstream electronics economy runs on. The markets — OpenAI, Anthropic, the hyperscalers — are already leaning into the white-hot enterprise market for AI coding (Claude Code, OpenAI’s Codex). But what the AI market truly needs is affordable AI compute, inference especially, on the mainstream side — on people’s phones, laptops, AI devices and all kinds of electronics. That requires not just memory chips but the AI packaging technology like CoWoS we discussed yesterday, and as much administration attention, globally, as GPU reshoring in Arizona has gotten.

To put it in gas-price terms: in electronics, we’re effectively moving toward $7, $8, $9 gas over the next couple of years — a real political issue not yet discounted in the broader AI Tech Wave. The mainframe AI side got its historic momentum on a lot of pushing and shoving. We need some of that same pushing and shoving on the other side of the barbell — on local AI computers and devices. And we need it fast.

Gadget AI — E-Bikes Before EVs Could Be a Thing in the US

MP Take: It’s a notable trend on relatively small volumes: roughly 1.7 million e-bikes sold in the US last year vs about 1.3 million EVs — so e-bikes are outselling EVs by unit volume, and that curve is widening, with better battery tech and more AI managing the throttle and assist systems. It gets less attention than self-driving cars, but it’s a striking signal for a country as car-centric as the US. To widen the lens: don’t over-read the volumes — an e-bike and an EV aren’t substitutes, and prices for both are likely to rise over the next few years on the very chip-and-memory shortages that are today’s theme. But the directional signal matters — even The Verge’s own writer went from e-bike hater to believer. E-bikes have long had more traction in Europe; now, in both urban and rural settings, they’re coming into their own in the US — a quietly real on-ramp to electrified, lower-friction transport.

Sources, in narrative order: Google AI Mode — E-bikes significantly outsell EVs in the US by unit volume. The Verge — How I went from an e-bike hater to a believer. For longtime readers: ‘Tesla vibe-coasts on AI Robotaxis’ in AI-RTZ #761.

Questions

Q1 — What’s MP’s favorite feature of e-bikes?

Longer rides — and a little ‘assist’ on those gnarly hills, of course. I’ve liked bikes and e-bikes for a long time; one of my favorite rides is around the park in Manhattan, and an e-bike lets me do two or three loops instead of the usual one. More range means more time outdoors and, with the pedaling you still do, a net workout — a subtle but real upside.

Q2 — What’s the biggest hassle?

Charging still requires extra logistics in houses, offices and apartment buildings — there’s far more charging infrastructure (and regulatory support) for cars than for bikes. The other is regulation: there’s huge variance in how e-bikes are classified and where they can be ridden — far more locally and state-driven than EVs. Some federal-level sanity and infrastructure support on both fronts would help.

Full Source Reading —

For the broader context, see the canonical sources for ARD 108 — in today’s narrative order:

Event 1 — Memory-Chip Price Impact on Smaller Tech

• CNBC — The memory shortage shaking Apple and Microsoft is an ‘existential crisis’ for smaller players

• WCCFTech — Jefferies warns memory prices will surge 50% in Q3 2026 and another 40% in Q4, no relief until 2028

• AI-RTZ #951 — Memory Chip Supply Constraints and Price Hikes a major AI headwind ahead

Event 2 — Big Memory Oligopolies Promise Bigger Investments

• FT — Samsung and SK Hynix plan $600 billion chipmaking expansion

• The Korea Times — Samsung, SK Hynix pledge for southwestern chip hub

• ARD #104 — Semiconductors Remain a Key AI Gating Factor

• AI-RTZ #557 — SK Hynix Ready for its AI Close-up

Event 3 — Big Customers Look for US-Government Help on China Memory

• Ming-Chi Kuo — The memory supply-demand gap will keep widening through 2027

• Reuters — China’s CXMT wins $3 billion memory-supply deal with Tencent

• AI-RTZ #1128 — US hamstrung on memory-chip supply via China

Gadget AI — E-Bikes Before EVs

• Google AI Mode — E-bikes significantly outsell EVs in the US by unit volume

• The Verge — How I went from an e-bike hater to a believer

• AI-RTZ #761 — Tesla vibe-coasts on AI Robotaxis

Shorts Clips from today

Clip 1 — Affordable AI Compute Inference Needed

Watch on YouTube Shorts

The AI market is leaning hard toward the white-hot enterprise coding market (Claude Code, OpenAI’s Codex). But what AI truly needs is affordable inference compute on the mainstream side — on people’s phones, laptops and AI devices.

MP Take: That’s the other side of the barbell, and it’s starved. It needs not just memory chips but AI packaging tech like CoWoS — and as much administration attention, globally, as GPU reshoring in Arizona has gotten. The mainframe AI side got its momentum from a lot of pushing and shoving; we need that same push and shove on local AI, fast.

Clip 2 — E-Bikes: More Than Just a Ride

Watch on YouTube Shorts

E-bikes get less attention than self-driving cars, but in the US they outsell EVs by unit volume — about 1.7M vs 1.3M last year, and the curve is widening with better batteries and more AI in the throttle systems.

MP Take: For a country as car-centric as the US, that’s a striking directional signal. Prices for both will likely rise on the same chip-and-memory shortages hitting everything else — but micro-mobility is a quietly real on-ramp to electrified, lower-friction transport, in both urban and rural settings.

Clip 3 — E-Bikes: The Next Big Thing in Gadget AI

Watch on YouTube Shorts

Why is an e-bike a ‘Gadget AI’ story? Better battery tech and more AI managing the throttle and assist systems are making e-bikes more affordable and capable — and even tough mainstream audiences are getting converted (The Verge: hater to believer).

MP Take: E-bikes have far more traction in Europe, but they’re coming into their own in the US, urban and rural alike. It’s the kind of mainstream, lower-cost electrification on-ramp that the AI/electronics wave should be enabling — even as RAMageddon threatens to push prices the wrong way.

Clip 4 — Tech’s Hidden Crisis: Ramageddon

Watch on YouTube Shorts

The memory shortage isn’t just an Apple/Microsoft story. CNBC calls it an existential crisis for smaller players; Jefferies warns prices surge ~50% in Q3 and ~40% in Q4, with no relief until 2028.

MP Take: Whole categories of $20-30 gadgets from tiny firms become unviable when DRAM goes from ~5% to 40-60% of the bill of materials — even Valve’s Steam Machine jumped from ~$500 to $1,100+. This is RAMageddon not just pushing but shoving smaller tech out of business, and it’s not yet understood in the market. We’ll only miss these products when they’re gone.

About AI Ramblings Daily (ARD), and AI-RTZ

Both are daily. Both are free. Both are about AI. But they’re different mediums carrying different messages.

AI-RTZ is the morning text — a deeper written take on one idea, published by at least 5 AM EST. Today: post #1133.

AI Ramblings Daily is the afternoon video + podcast — my ad hoc takes and perspective on the day’s AI issues & news flow, around 20 minutes, with short 1-2 minute clips for quick topic views. Today: episode #108.

Subscribe to either or both on michaelparekh.substack.com. They run as separate Sections you can opt into or out of.

Links used in today’s show (already embedded inline above; listed here for reference)

Take 1 — Memory-Chip Price Impact on Smaller Tech:

• CNBC — The memory shortage shaking Apple and Microsoft is an ‘existential crisis’ for smaller players

• WCCFTech — Jefferies warns memory prices will surge 50% in Q3 2026 and another 40% in Q4, no relief until 2028

• AI-RTZ #951 — Memory Chip Supply Constraints and Price Hikes a major AI headwind ahead

Take 2 — Big Memory Oligopolies Promise Bigger Investments:

• FT — Samsung and SK Hynix plan $600 billion chipmaking expansion

• The Korea Times — Samsung, SK Hynix pledge for southwestern chip hub

• ARD #104 — Semiconductors Remain a Key AI Gating Factor

• AI-RTZ #557 — SK Hynix Ready for its AI Close-up

Take 3 — Big Customers Look for US-Government Help on China Memory:

• Ming-Chi Kuo — The memory supply-demand gap will keep widening through 2027

• Reuters — China’s CXMT wins $3 billion memory-supply deal with Tencent

• AI-RTZ #1128 — US hamstrung on memory-chip supply via China

Gadget AI — E-Bikes Before EVs:

• Google AI Mode — E-bikes significantly outsell EVs in the US by unit volume

• The Verge — How I went from an e-bike hater to a believer

• AI-RTZ #761 — Tesla vibe-coasts on AI Robotaxis

Q1 + Q2 — MP’s e-bike feature + hassle:

• (no external sources — MP’s own analyst view)

Companion text:

• AI-RTZ #1133 — Amazon-Anthropic relationship evolving

AI Ramblings Daily on AI-RTZ is here to think through AI and reset. Together.

Today’s AI-RTZ #1133 — Amazon-Anthropic relationship evolving — Amazon, an early Anthropic partner with over $13B invested, is now also partnering with OpenAI on a reported ~$50B deal — so the Amazon/Anthropic dynamic is shifting much like Microsoft/OpenAI did over the last six months. With both OpenAI and Anthropic eyeing IPOs this year or next, these key partnerships are an important lens on their prospects. Recommended as today’s reading post.

Tomorrow — ARD 109 on AI-RTZ 1134.

Thanks for joining us today, AI Curious Folk. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here.)

Subscribe to AI: Reset to Zero for daily AI Ramblings + Sunday Bigger Picture posts.