Four filings pulled from SEC EDGAR and read directly (KLIC 10-Q filed 2026-05-07; AXT 10-K filed 2026-03-17; COHR 10-K filed 2025-08-15 + 10-Q filed 2026-05-06; LITE 10-Q filed 2026-05-06). Three standing rows resolve; two finds are thesis-relevant NEGATIVES.

1. KLIC — the AP mix SHRANK; this is a legacy wire-bond cycle (thesis-relevant negative)

At the only grain KLIC discloses (four reportable segments; "thermocompression" and die-attach are bundled into one "Advanced Solutions" line; "hybrid bonding" appears nowhere in the 10-Q):

Six-month basis: Advanced Solutions fell 14.0% → 9.4% of revenue while Ball Bonding jumped 38.4% → 61.2%. The end-market disaggregation says why: General Semiconductor $273.5M (vs $156.2M) and Memory $47.5M (vs $13.5M) — the current KLIC boom is the legacy ball-bond memory/general-semi upcycle, not the advanced-packaging mix-shift. The AP-growth-engine leg of the KLIC thesis is UNPROVEN at the disclosed grain and mix-wise moved the wrong way; the cyclical-earnings-turn leg (which drove the June verdict) is what the numbers support. No AP-specific backlog/capacity language exists. Prior-year AS operating income carried a one-time $71.1M project-cancellation reimbursement (technology unnamed — do not infer TC/HB).

2. AXT — permits are lane-by-lane and unpredictable; China now 62% of revenue

• Concentration: no customer ≥10% of revenue in any of FY2023–25; top-5 basket 29% (FY25) / 30% / 25% — a basket risk, not a single-name overhang. One customer >10% of receivables.
• Geography tightened the real risk: China 62% of FY25 revenue (53% in FY23) while North America fell 77.5% — the filing states directly the NA collapse is "primarily due to export permit requirements on InP established by China on February 4, 2025."
• Permit cadence (verbatim substance): InP added to China's export-control list 2025-02-04; portal opened March 2025; Europe/Japan permits granted 2025-06-11; U.S.-bound InP AND GaAs permits still unapproved as of the filing date (U.S. GaAs customers are "dual use"). "The timing for receiving permits remains uncertain, unclear and beyond our control." Each destination lane effectively needs its own permit — there is no fixed renewal schedule to track; the MOFCOM tracker should watch permit GRANTS by lane, not renewal dates.
• Sumitomo: named only as a general substrate competitor; no head-to-head 6-inch InP comparison in this filing (AXT's "we can be the dominant supplier in the emerging 6-inch market" is its own forward claim). The Sumitomo catch-up timeline needs non-filing sources.

3. COHR/LITE — the committed-dollar floor is NOT footnote-verified (thesis-relevant negative)

• COHR 10-Q has NO commitments footnote at all (note index skips it; zero hits on "purchase obligation"/"unconditional"/"noncancelable"). The NVIDIA arrangement appears only as MD&A narrative: "multi-billion-dollar purchase commitment" with no dollar figure, no time buckets, and "no material liability was recorded... solely as a result of entering into the agreement."
• LITE's only quantified commitments run the WRONG DIRECTION — $1,795.6M of purchase obligations are LITE's own to suppliers ($1,629.9M <1yr). On the customer side its 10-Q states: "the majority of our customers do not have contractual purchase commitments." Customer deposits are immaterial ($7.3M current, mostly an acquisition artifact). Zero hits on "hyperscaler."
• The one hard, dollar-verified lock is a SHAREHOLDER lockup, not a supply floor: NVIDIA's $2B private placement (7,788,161 shares at $256.80, closed 2026-03-02) carries a six-month transfer restriction — lapsing ~September 2026, a date the lane should carry as an overhang watch.
• Read: the "customers contractually locked into optical supply" framing is currently a narrative claim from COHR's MD&A, not a footnote-verified obligation. Treat the $2B as an equity raise with a capacity-agreement side letter until a take-or-pay number appears in a filing.
• COHR FY25 capex by segment (pre-realignment grain, the finest disclosed): Networking $263.4M (vs $91.0M FY24 — the AI build is visible here), Materials $137.4M, Lasers $40.0M. The segment structure changed 2025-07-01 (Datacenter & Communications / Industrial), so this table won't carry forward.

Verdict + reasoning

Resolved: all three standing filing questions answered from primary documents. Two negatives now on the record — KLIC's AP mix is shrinking (the boom is legacy), and the optical committed-floor is unverified narrative — plus one new dated watch (NVIDIA's COHR lockup lapse ~September 2026). None of these falsify the optical-supercycle demand thesis on their own; they narrow which LEGS the desk may cite as verified.

Sources

• research/market-engine/data/stocks/KLIC/edgar/10Q-2026-05-07.html (Note 15 + MD&A Net Revenue)
• research/market-engine/data/stocks/AXTI/edgar/10K-2026-03-17.html (concentration note, geographic split, export-control disclosure)
• research/market-engine/data/stocks/COHR/edgar/10K-2025-08-15.html (Note 21 segment capex) + 10Q-2026-05-06.html (MD&A NVIDIA agreements; Note 12 equity)
• research/market-engine/data/stocks/LITE/edgar/10Q-2026-05-06.html (Note 14 purchase obligations; risk factors)