bitdepth.co
LIVE
HBM3E demand +47% q/qNVDA inventory lead time −9dSK Hynix HBM4 risk prod 2026Q3TSMC CoWoS capex +$3.8B
Issue Nº 6 · The pillar edition

Pillars,
not headlines.

Tech moves too fast for daily news to mean anything. We organize what matters into long-running, continuously-updated topic hubs — and let the headline churn happen at the edges.

6Active pillars
13Sub-articles
12Tracked sources
1Editor (me)
ActiveAI · Memory · Edition 14 · Updated 2h ago

The AI memory bottleneck.

The constraint that's quietly reshaping the trillion-dollar AI compute build-out — explained, tracked, and updated as it evolves. Six solution paths, four still unresolved.

Sub-articles9
Reading time22 min
Read the pillar  →
HBM demand vs. supply, 20263.2×
HBM TAM forecast, 2027$84B
Avg CoWoS lead time52wk
HBM market held by top 394%
Solution paths tracked6
§ All pillars

What I'm following.

All 6 pillars →
§ Latest analyses

Long-form, this week.

May 12AI · Memory9 min read

Samsung's HBM3E qualification, finally. The full timeline and what it means for NVIDIA's 2026 allocation strategy.

For 18 months Samsung was the third HBM vendor on paper only. NVIDIA's quiet May qualification changes everything — including SK Hynix's pricing power and Micron's growth ceiling.

May 11Energy · Grid14 min read

The hyperscalers built nuclear deals nobody can finance. Here's what happens when reality hits.

SMR contracts announced in 2024 and 2025 looked clean. The capex curves don't pencil out, and three of seven announced deals are quietly behind. Source-by-source breakdown.

May 10Policy · Geo7 min read

Commerce's HBM3E export guidance is narrower than reported. The HBM4 ambiguity is the real story.

Reading the actual rule text — not the headlines — reveals three exemptions and one critical omission. We walk through every line that matters.

May 09AI · Memory14 min read

Why every AI training run is now a packaging negotiation.

The bottleneck no longer measured in GPUs but in CoWoS slots. How TSMC quietly became the gatekeeper of the entire frontier-AI roadmap — and what the 2027 packaging map looks like.

May 08AI · Memory11 min read

Cerebras WSE-4 is generally available. We ran the benchmarks. The numbers are real.

Wafer-scale inference has spent four years as a press release. With WSE-4 in our hands, the latency-vs-cost economics finally pencil out for a specific, narrow class of workloads.

§ Long-reads

The essay shelf.

All essays →

We use cookies

We use cookies to ensure you get the best experience on our website. For more information on how we use cookies, please see our cookie policy.