- Author argues DRAM companies offer AI-critical, substitute-poor demand, concentrated four-player supply, and record capital return, while valuations price a sharp 2027–28 cyclical decline; margins could stay above prior peaks if AI demand holds.
- Reported Google interest in securing memory through 2035 supports durable demand. Elon’s Terafab reportedly targets SRAM/logic volume production in 2H28—likely 2029–30—not traditional DRAM.
- Memory manufacturing requires difficult yields, economics, and proprietary IP. The author favors an eventual MU–Elon partnership but warns premature capacity expansion could recreate oversupply before Terafab demand materializes.
Investment pitches
filter:
61 72 - Negative hyperscaler free cash flow reflects heavy reinvestment, not weak economics: operating cash flow is at record levels and accelerating.
- Semis recognize system revenue upfront, while hyperscalers monetize datacenters over a decade or longer; divergent FCF profiles reflect timing and business models.
- The author sees semis as especially attractive: NVDA is cited at roughly 100% growth and 10x, while fears of unsustainable profits look misplaced.
- Hyperscalers also appear undervalued given durable franchises, accelerating growth, and their essential role alongside semis in the AI buildout.
81 - Long-running, token-heavy agentic workloads could make memory the market’s biggest blind spot: long decode raises capacity and bandwidth needs while compression and quantization lose effectiveness.
- Author favors HBM/DRAM names including MU and SKHY, plus Samsung; current low-to-mid-single-digit P/Es look dislocated if demand outruns supply, sustaining pricing, margins, earnings, and NAND spillover.
- Historical supply-cycle analysis is a counterweight, but the author expects this cycle’s demand growth to exceed forecast supply growth for longer.
- NVDA’s integrated compute, networking, interconnect, rack architecture, and software may deliver superior full-system goodput as prefill/decode complexity makes isolated chip benchmarks less relevant; the author remains bullish.