Memory · Power · Custom Chips · Optics — Four Bottlenecks Converging Simultaneously
Four bottlenecks converging simultaneously — prioritize true global #1 by sector; recommend Korean names only where Korea genuinely leads.
Big Tech 4 2026 CapEx raised to $700B+; Microsoft CFO stated '83% of increase is component costs,' confirming memory price pass-through structure.
Winners — SK hynix, Doosan Enerbility, TSMC, Marvell, Lumentum. Caution — Hanmi Semiconductor (share loss), OE Solutions (awaiting profitability).
SK hynix quarterly OPM trend, Doosan new order disclosures, TSMC CoWoS expansion schedule, Lumentum InP capacity expansion, Apple quarterly memory-cost guide commentary.
Combined 2026 investment plans increased from $645B to ~$700B, up approximately $60B. The headline looks like "AI investment grew again," but the actual meaning is different.
| Company | CapEx Change | Key Comment |
|---|---|---|
| +$50B | Sundar Pichai: "Cloud revenue would have been higher if we could meet all demand" | |
| Microsoft | +$300B | Amy Hood: "$250B of the $300B increase is component price inflation" |
| Meta | +$100-150B | Zuckerberg: "Most of the increase is component costs, especially memory" |
| Amazon | No change | Andy Jassy: "We have $225B in Trainium revenue commitments" |
The most important statement came from Microsoft CFO: "$250B of the $300B increase is component price inflation." They're spending more for the same quantity of servers because memory prices have risen.
> Amy Hood (Microsoft CFO): "Our 2026 capital expenditure of $190B includes approximately $25B of component price increases. Despite this additional investment, we will remain capacity constrained throughout at least 2026."
All four companies used the same phrase in their earnings calls: "compute constrained" — not enough computing capacity.
The most significant change this quarter is that custom chips (ASICs) are no longer just "internal cost-saving tools." They're now generating external revenue.
Amazon is the clearest example. CEO Andy Jassy disclosed "$225B in Trainium revenue commitments" — Anthropic at 5GW, OpenAI at 2GW, plus Uber and Meta.
> Andy Jassy (Amazon CEO): "Our custom AI chip business has entered the top 3 datacenter chip businesses globally. We now have over $225B in Trainium revenue commitments."
Google also announced that they will "deliver TPU hardware directly to certain customers' datacenters" — effectively turning TPU into an external revenue stream.
Sign in with Google — your tier will be checked automatically and access granted if eligible.
Sign in with GoogleThis site runs on ads — the tier system rewards community contributions.
This report is provided for informational purposes only and does not constitute a recommendation to buy or sell any financial instrument. Investment decisions should be made based on your own judgment and responsibility. The analysis and opinions contained herein are based on information available at the time of writing and are subject to change.