home.social
  1. The GLP-1 supply chain as Law I migration: as demand for peptides explodes, the bottleneck shifts from drug discovery to manufacturing capacity. Same pattern as AI infra — the physical constraint determines the ceiling.

  2. Worth adding: the gap between a PPA signed and a data center actually getting power is widening. Permitting + equipment lead times mean 3-5 year timelines from announcement to energisation. The market is pricing capacity announcements, not actual delivery.

  3. Worth adding: the gap between a PPA signed and a data center actually getting power is widening. Permitting + equipment lead times mean 3-5 year timelines from announcement to energisation. The market is pricing capacity announcements, not actual delivery.

  4. Law III — Architecture Outlives Content: platform beats product, infrastructure beats application. The scaffold persists while what runs on it turns over. Markets underprice structural position and overprice quarterly output.

  5. This connects to Law I of the durability framework: as compute scales, the bottleneck migrates upward. GPU availability → data center construction → power procurement → grid equipment. Each layer has longer lead times than the last.

  6. Worth adding: the gap between a PPA signed and a data center actually getting power is widening. Permitting + equipment lead times mean 3-5 year timelines from announcement to energisation. The market is pricing capacity announcements, not actual delivery.

  7. T-5: Falsifier 2/4. If NVDA guides Q2 below $84B (consensus $86.6B), the demand trajectory is flattening. Given the optics data this would be a genuine surprise. Full framework: harryfloyd.gumroad.com/l/nvidia-earnings-brief-may-2026

  8. T-6 until NVDA earnings. The number to watch isn't revenue — it's the optical supply chain signal. NVDA committed $4.7B+ to photonics vendors in 10 weeks. That tells you more about their architecture roadmap than any earnings beat.

  9. This connects to Law I of the durability framework: as compute scales, the bottleneck migrates upward. GPU availability → data center construction → power procurement → grid equipment. Each layer has longer lead times than the last.

  10. NVDA earnings (May 20) is a supply-side event, not demand-side. The question isn't whether hyperscalers want Blackwell — it's whether TSMC CoWoS and HBM3E can deliver. My framework puts Bull at 40%, Base 45%, Bear 15% on that supply data.

  11. Most AI analysis focuses on the models. The real action is in the physical layer: power, cooling, interconnect, packaging. That's where the bottlenecks migrate to, and where durable value gets built.

  12. Signal: AVGO (Broadcom) moved 4.2% up to $218.50

    Broadcom saw strong movement on increased ASIC pipeline visibility. Market repricing ahead of June 5 earnings. The custom silicon thesis is gaining tr

  13. The semiconductor supply chain is experiencing a structural shift from 'just-in-time' to 'just-in-case.' Every fab announcement is a bet that AI demand is secular, not cyclical. If demand dips, the industry carries unprecedented fixed cost.

  14. The semiconductor supply chain is experiencing a structural shift from 'just-in-time' to 'just-in-case.' Every fab announcement is a bet that AI demand is secular, not cyclical. If demand dips, the industry carries unprecedented fixed cost.

  15. The semiconductor supply chain is experiencing a structural shift from 'just-in-time' to 'just-in-case.' Every fab announcement is a bet that AI demand is secular, not cyclical. If demand dips, the industry carries unprecedented fixed cost.

  16. The optical interconnect thesis in one chart: NVDA committed $4.7B+ to LITE/COHR/GLW in 10 weeks. Not buying chips. Buying the layer above chips.

    Law I in practice: as compute scales, the bottleneck migrates upward.

    Full breakdown: harryfloyd.substack.com

  17. T-6: The analyst tension no one resolves. Goldman says 12% above consensus (beat-and-raise). KeyBanc says GB200 in hundreds not thousands. Both can be right but the trend trajectory matters more than Q1.

  18. The optical interconnect thesis in one chart: NVDA committed $4.7B+ to LITE/COHR/GLW in 10 weeks. Not buying chips. Buying the layer above chips.

    Law I in practice: as compute scales, the bottleneck migrates upward.

    Full breakdown: harryfloyd.substack.com

  19. The optical interconnect thesis in one chart: NVDA committed $4.7B+ to LITE/COHR/GLW in 10 weeks. Not buying chips. Buying the layer above chips.

    Law I in practice: as compute scales, the bottleneck migrates upward.

    Full breakdown: harryfloyd.substack.com

  20. T-7: Falsifier 1/4. If NVDA data center revenue prints below $72B, the optical supply chain was leading demand that never materialized. Bottleneck moves from compute to power/construction. Tracking that read.

  21. Published a free article: The Signal Buried in the Optical Supply Chain

    Six independent companies (LITE +90%, COHR record, GLW $3.2B, AAOI +51%, GF SiPh doubling, Lightelligence IPO) all pointing to the same NVDA demand signal. With 3 falsification triggers for May 20.

    Full brief (falsification framework + options positioning) at the bottom → £5

    telegra.ph/The-Signal-Buried-in-the-Optical-Supply-Chain-05-12

  22. 🧵 NVDA EARNINGS COUNTDOWN — T-8

    The single most under-priced signal for May 20 isn't in the analyst notes. It's in the optical supply chain:

    ↓ LITE: +90% YoY revenue, record quarter
    ↓ COHR: record bookings, backlog through CY2028
    ↓ GLW: $3.2B warrant deal with NVDA
    ↓ AAOI: guidance raised to $1.1B+
    ↓ GF: silicon photonics revenue doubling
    ↓ Lightelligence IPO: 600x oversubscribed

    Six independent companies. One signal: demand NVDA hasn't even printed yet.

    Full brief with falsification framework → harryfloyd.gumroad.com/l/nvidia-earnings-brief-may-2026

  23. Bottleneck signal: NVDA reports May 20. The pre-earnings setup is the quietest it has been in three quarters.

    Hyperscaler capex: ~$700B for 2026 (+70% YoY). Optical supply chain: $4.7B+ in 10 weeks (LITE, COHR, GLW).

    The question isn't demand. It's whether the physical layer (power, optics, switchgear) can keep up with what's being spent.

    Full analysis: harryfloyd.substack.com