"Charts of the Week — Memory to the..." — a16z editorial team
Why this is in the vault
Two threads the founder is actively positioned on: (1) the AI productivity-acceleration claim that anchors the Innermost Loop demand side, and (2) memory stocks as the chips-layer beneficiary inside that thesis. a16z's curation is directionally aligned with our framework, which means it's also a useful sentiment check — when the VC narrative arrives on chart-of-the-week status, the trade has already been seen by everyone.
⚠️ Sponsorship
a16z is a venture firm with structural portfolio bias toward bullish AI-adoption narratives. No explicit per-article sponsor block, but the standard disclaimer ("any investments or portfolio companies mentioned are not representative of all investments in vehicles managed by a16z") confirms portfolio overlap is undisclosed by default. Treat all forward-looking claims here as supporting the firm's book, not neutral analysis. Cross-check tickers against the public a16z investment list before any position decision.
Issue contents
Thread 1 — AI productivity acceleration (three charts)
- Affirm pull requests per week. Inflection after the team "re-tooled software engineering to be truly AI-first." PRs more than doubled; ~2/3 are now agentic. Source: Affirm investor disclosures. Single-company anecdote, not a sector study.
- Output per employee, by industry. Claims an upward inflection in 2025, "especially amongst the High AI Industries." a16z's own caveat: "productivity is notoriously hard to measure in the aggregate."
- Agent adoption by sector. "Nearly 1 in 5 firms are using agents." Manufacturing surprisingly accounts for ~18% of agent usage despite <10% of firms. Source: Microsoft data.
Thread 2 — Memory stocks ("to the moon")
- Companies: Samsung, SK Hynix, Micron. Western Digital + SanDisk spinout mentioned in the broader storage sub-thread.
- Y-axis: Operating income and net income — not stock price. This matters: the chart is an earnings chart, not a returns chart, so any "the run played out" question requires going back to price/EV multiples separately.
- Period: 2025 actuals + 2026 projections.
- Headline claim: Memory makers expected to "sextuple their operating income in 2026." Micron earned more in Q1 2026 alone than in any single full year before 2025.
- Hook: Memory has shifted from commoditized component to AI infrastructure bottleneck with pricing power. PC and phone prices projected to rise ~10-20% due to memory costs. Whether this is a step-change or a one-time spike is framed as a "multi-hundred billion question" — a16z explicitly does not commit.
- Adjacent mentions: Google + Amazon custom chip development, NVIDIA GPU dependency as the demand pull.
Mapping against Ray Data Co
On productivity acceleration (Q1)
The Affirm chart is the strongest evidence in the piece — it's a single-company case study with a clear measurement (PRs/week) and a clear treatment (AI-first re-tooling). It reinforces the [[2026-05-14-treybig-how-agents-use-systems-differently]] observation that engineering throughput is one of the first surfaces to inflect.
The aggregate "output per employee by industry" chart is softer. a16z's own caveat is the right read: aggregate productivity stats are notoriously lagging and noisy. Brynjolfsson / Aghion / StanfordHAI work over 2024-2025 has shown narrow-task productivity gains (20-50% on customer-support, coding, writing tasks) but the macro signal is still ambiguous. So: real at the firm level, hype-cycle-peak at the macro narrative level.
The Microsoft "1 in 5 firms using agents" data point is consistent with [[2026-05-15-innermostloop-singularity-optimizing-optimizer]]'s stage-marker reading — adoption is real and broad, but "using" sets a low bar (one prompt counts).
On memory stocks (Q2) — the cyclicality discipline
This connects directly to [[01-projects/investing/theses/2026-05-12-innermost-loop-ai-infrastructure]] where Micron is named in the chips layer. The piece supports the multi-year setup but does not address the cyclicality risk that disciplines any memory thesis:
- Memory is historically the most cyclical semiconductor sub-sector. The 2017-2018 NAND glut produced 60-70% drawdowns from peak in 12-18 months. The 2022 DRAM downcycle did similar damage.
- HBM tightness through 2026-2027 is real (SK Hynix dominates HBM3E, Micron + Samsung chasing), and AI capex is the demand pull. But "sextuple operating income in 2026" is also a peak-earnings forward call — earnings charts that go vertical are usually within 2-4 quarters of the cycle top.
- The fact that this is now an a16z chart-of-the-week is itself a late-cycle attention marker. The narrative has arrived at the retail surface.
The real question for sizing isn't "buy/don't buy" — it's: (a) is this a position-trade (where the next 2-4 quarters of earnings revisions matter and you need an exit discipline) or a thesis-hold (where you're underwriting the 2027-2030 HBM-on-every-rack story and willing to sit through the next cyclical drawdown)? Those are two completely different position sizes and stop-loss rules.
[[01-projects/investing/candidates/README]] should add Micron / SK Hynix as named candidates with the cyclicality red flag explicitly disclosed, parallel to how Cerebras got 4 S1 red flags surfaced.
Related
- [[01-projects/investing/theses/2026-05-12-innermost-loop-ai-infrastructure]] — names Micron in chips layer
- [[06-reference/2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]] — source thesis
- [[06-reference/2026-05-15-innermostloop-singularity-optimizing-optimizer]] — recent adoption stage markers
- [[06-reference/2026-05-14-treybig-how-agents-use-systems-differently]] — agentic engineering throughput
- [[06-reference/2026-04-17-dwarkesh-jensen-case-for-chips-china]] — Jensen on chip supply
- [[01-projects/investing/candidates/README]] — candidate roster (Micron/Hynix should be added here)