Type: thesis-shift (subtype: hardens — invisible-bottleneck layer added)

event_id: 2026-05-10-temple8-ndk-timing-layer-hyperlight-tfln-routing

Sources:

• Temple 8 Research 2026-04-12 — "NDK (6779.T) — The Timing Layer of 1.6T and F-35"
• Temple 8 Research 2026-04-12 — "Hyperlight: Leading The TFLN Optical Revolution"

Shift: Routes two month-old Temple8 reads that predate the 2026-05-04 three-source convergence refresh and weren't cited downstream. Both add a new physics-layer to the thesis, not new names within layers we already cover.

(1) NDK (6779.T) — the timing/clock layer. Ultra-low-jitter Oven-Controlled Crystal Oscillators (OCXOs) required for 1.6T transceiver signal integrity at 200G-per-lane. A single picosecond of jitter corrupts the entire data burst — no software workaround. NDK controls ultra-purity N-Grade EX synthetic quartz feedstock; part of the "Elite 5" (Seiko Epson, Murata, Kyocera, NDK, Microchip) controlling ~45% of the $3.2B crystal-oscillator market. ~$350M Tokyo-listed micro-cap. Dual-use angle: F-35 Block 4 avionics also require NDK-spec radiation-hardened oscillators. China quartz-export restrictions (Jiangsu, 2024) add a geopolitical scarcity overlay — analogous to the tungsten / rare-earth posture. This is the layer our 3-agent supply-chain trace systematically missed (acknowledged in the Temple8 article itself); Temple8 finds the layers others forget about. Adding to key_tickers + watchlist.

(2) Hyperlight (PRIVATE, ~$1.3B Series B) — TFLN modulator dominance. Cambridge, MA private company calling itself "the ARM of photonics." Confirmed design-in relationships with Broadcom, Marvell, AWS, Meta, and NVIDIA simultaneously — that's the moat shape. Thesis: silicon photonics can't handle 200G-per-lane at rack temps >85°C, InP hits thermal limits at 1.6T speeds, TFLN is the materials-science solution for the 1.6T → 3.2T transition. NOT directly investable (pre-IPO); flagging as private-IPO-watch per the same Cerebras-shape "private-but-watch" pattern documented in 2026-05-06-ai-counterparty-taxonomy-cerebras-trigger. Investable supply chain (per Temple8): NGK (TFLN wafer substrates — NGKIF already in our watchlist, confirms that add was correct), UMC (foundry partner — broader than optical, not adding to optical-supply-chain), Jabil (JBL packaging partner — already in our watchlist). If Hyperlight IPOs, Temple8 frames it as "the most important photonics IPO since Lumentum (2015)" — file Phase 1 prep work mechanically (S-1 read + cross-reference) per Cerebras playbook when filing surfaces.

(3) LITE EML moat may be time-limited. Temple8 is openly questioning the LITE bull case beyond the 1.6T cycle — TFLN is the next-gen modulator path and Hyperlight (private) + LWLG (public, already key_ticker) are the natural successors. Not yet a re-grade on LITE — Wave 1 still works through the current 1.6T print cycle — but a 2027/2028-horizon watch-for: if hyperscalers signal TFLN as the 3.2T standard, LITE's EML monopoly story compresses. Adds a measurable watch-for to the perspective.

Effect on thesis:

• 6779.T (NDK) added to key_tickers (now 20 names) AND to optical-supply-chain.json watchlist with full provenance. Adds the timing/clock physics-layer. Conviction: medium — micro-cap liquidity + Tokyo-primary listing are the cleanups, but the substantive bottleneck is real and the dual-use F-35 demand is independent of the AI cycle.
• Hyperlight flagged as private-IPO-watch (not investable directly; not added to any watchlist).
• NGKIF reaffirmed — Temple8 read independently corroborates the TFLN-substrate thesis we already hold.
• JBL reaffirmed — Temple8 names Jabil as the Hyperlight packaging partner; already in optical-supply-chain.
• UMC not added — too broad a foundry play to fit optical-supply-chain specifically; defer until a 2nd substantive voice surfaces UMC at the optical layer.
• New watch-for added: if hyperscalers signal TFLN as the 3.2T standard during 2027 capex commentary, LITE EML monopoly story compresses. Track via Coherent + Lumentum earnings transcripts.

Per-ticker: 6779.T / NDK ↑ (new key_ticker; timing-layer chokepoint). NGKIF → (reaffirmed; TFLN substrate). JBL → (reaffirmed; Hyperlight packaging). LITE → (no re-grade now; 2027+ watch-for filed). LWLG → (reaffirmed; polymer-modulator path is the TFLN-adjacent public option). Hyperlight → (private-IPO-watch flag, no ticker).

Methodology note: Two month-old Temple8 reads sat un-cited because they predated the 2026-05-04 convergence refresh and the orphan-mention audit hadn't been run against article-class inputs at the time. Same orphan pattern as CGEH (8 days) and the phynvesting substack — different cadence (5 weeks), same shape. Article-class orphans are likely a separate audit slice from tweet-class orphans; file as methodology task if pattern repeats.