Salience Labs

Phase A — Understand the business

Lens 1 · Company Overview

What it is. Salience Labs is a UK silicon-photonics company spun out of the University of Oxford and the University of Münster in 2021, commercialising ~a decade of academic research into chip-scale photonics for AI infrastructure. HQ is Oxford, with sites in Heidelberg (Germany) and San Jose / Silicon Valley.

The business model has changed once, materially. This is the single most important fact about the company and frames the entire dossier:

• 2021–2023 (v1 — optical compute): the original pitch was an "ultra-high-speed multi-chip processor" packaging a photonic compute chip on top of SRAM — a proprietary "on-memory compute" amplitude-based architecture, "stacking up to 64 vectors into a beam of light," clocking "10s of GHz," targeting exascale AI compute. This was a photonic matrix-multiply / AI-accelerator play, competing conceptually with Lightmatter's and Lightelligence's optical compute cores.
• 2024–2026 (v2 — optical switching): the company pivoted to all-optical circuit switching (OCS) — photonic switches that connect GPUs across racks in an AI datacenter, replacing electrical packet switches and the optical-electrical-optical (OEO) conversions they require. It now describes itself as "a leader in photonic solutions targeting connectivity for AI datacenter infrastructure".

Why the pivot (management's own account). CEO Vaysh Kewada frames it as application-pull, not technology-failure: "what changed wasn't the technology, but the urgency and clarity of the application — AI made inefficient networking impossible to ignore … performance scaling was becoming limited by data movement rather than compute". Read skeptically (Lens 13), a pivot from "compute" to "interconnect/switching" four years in is also what a company does when the original compute thesis did not win design sockets — Lightmatter and Celestial AI both out-raised and out-scaled the optical-compute lane, and Salience does not appear on any hyperscaler compute roadmap. The pivot is, on balance, into a better and more defensible adjacent market (see Lens 3), which is the bullish reading.

Product today. The first shipping product is a 32-port all-optical circuit switch, launched 10–11 March 2026, billed as "industry's highest performing 32-port all-optical silicon photonic switch," fitting in a fraction of 1RU, compatible with existing transceivers. Roadmap: 64-port and 128-port variants to follow.

Customers / contract structure. customers.csv is empty ``. No named customer or design win is public as of 2026-06-24 — the disclosed relationships are a manufacturing partner (Tower) and a test-equipment partner (Keysight), not buyers (Lens 2). Revenue model is presumed hardware unit sales into hyperscale / neocloud datacenter operators, but no pricing, no take-or-pay, no backlog is disclosed — n/a — private, not disclosed.

Lens 2 · Supply Chain

Map the OCS value chain, naming actual stakeholders along it (names or it didn't happen):

Upstream (inputs → Salience):

• Wafer foundry / PIC fab — Tower Semiconductor (TSEM). The load-bearing dependency. Feb 2026: Tower + Salience announced a partnership to manufacture PIC-based OCS, explicitly "moving from development into pre-production," on Tower's silicon-photonics platforms PH18DA (with integrated III-V lasers) and TPS45PH (low-loss nitride waveguides). This is the single-source manufacturing chokepoint — a fab-less startup is entirely dependent on Tower's process yield and capacity.
• Test & validation — Keysight Technologies (KEYS). March 2026 collaboration to build "the first optical circuit switch testing environment," demoed at OFC 2026. Important because OCS has no mature test methodology — a real gap Salience and Keysight are filling, but also a sign the category is immature.
• Core photonic building blocks (per technical press): chip-based optical frequency combs / soliton microcombs at microwave line rates, ultra-low-loss silicon-nitride waveguides, on-chip high-speed detectors and modulators, and integrated III-V lasers (via Tower's PH18DA). III-V laser integration is historically the hardest, lowest-yield step in silicon photonics — a named chokepoint.
• Academic IP base: Oxford (Prof. Harish Bhaskaran's Advanced Nanoscale Engineering group) + Münster (Prof. Wolfram Pernice) — the foundational photonic-compute / phase-change-photonics research.

Salience (the company): fabless PIC design + system integration into a 1RU-class optical switch chassis.

Downstream (Salience → end customer):

• End buyers: hyperscale / AI-datacenter operators — the named category buyers in the OCS market are Google (by far the largest, "tens of thousands of OCS ports" deployed internally via its Apollo/Palomar system), and an expanding set of "operators beyond Google". None are disclosed as Salience customers.
• Channel: direct-to-operator hardware sales presumed; no distributor or OEM reseller named — n/a — not disclosed.

Chokepoints / single-source dependencies: (1) Tower as sole fab; (2) III-V laser integration yield; (3) a customer base that is a handful of hyperscalers, most of whom (Google especially) build their own OCS — the demand side is both concentrated and partially captive to in-house solutions.

Lens 3 · Competitive Advantages (moats)

The structural bull case for the lane (not yet the company). OCS replaces the energy- and latency-cost of converting optical signals to electrical and back (OEO) every time a packet is switched. Salience's claimed deltas vs. incumbent OEO switching: up to 8× power savings and up to 80% improvement in Tokens-per-Second-per-User (an AI-inference latency metric), by removing tail latency. If those hold in production, OCS is a genuine secular win — multiple analysts now argue "all AI datacenter interconnects will be optical within 5 years".

The differentiation Salience actually claims vs. other OCS approaches. The incumbent OCS technology is MEMS micro-mirrors — Google's Apollo/Palomar, Lumentum's R300, Calient. MEMS reconfigures on the order of tens of milliseconds because it physically tilts mirrors. Salience's pitch is an all-silicon-photonic, solid-state switch that "rapidly creates and tears down optical paths" with no moving parts — implying far faster reconfiguration and better reliability/lifetime than mechanical MEMS. (LCoS lasts ~10yr vs ~3yr for wear-prone MEMS — a solid-state photonic switch should beat both on lifetime.)

Candidate moats — and how durable each really is:

• IP / process know-how: a decade of Oxford/Münster photonics + a hard-to-replicate III-V-on-silicon switch design. Moderately durable but not unique — every photonics startup claims a process moat, and the foundational techniques are increasingly available on merchant fab platforms (Tower, GlobalFoundries Fotonix).
• Speed-of-switching: if silicon-photonic OCS genuinely switches orders of magnitude faster than MEMS, that is a real product advantage — but it is a feature, not a moat: Nvidia, Lumentum and well-funded peers can pursue the same physics.
• Switching costs / network effects: ~none yet — no installed base.
• Scale: negative — Salience is the smallest credible player in the lane (Lens 7).

Bargaining power. Weak on both sides: it needs Tower's fab more than Tower needs it, and it needs hyperscaler design wins far more than hyperscalers (who have MEMS OCS today and Nvidia's photonic switches arriving) need a 35-person startup. Bargaining power is the company's weakest attribute.

Verdict on the moat: the market is moaty (physics + hyperscaler capex); Salience's position in it is, today, a promising product without a defensible economic moat. The moat has to be earned via a design win and a manufacturing ramp it has not yet demonstrated.

Lens 4 · Segments

segments.csv is empty ``. As a single-product, pre-scale private company there is no segment or geographic revenue breakout to source. Functionally one segment: all-optical circuit switches for AI datacenter networking, with a stated port-count roadmap (32 → 64 → 128) rather than a product-line split. Geographic footprint (Oxford / Heidelberg / San Jose) is an R&D-location split, not a revenue split. n/a — private, not disclosed.

---

Phase B — Measure performance

+private overlay applied. Lens 5 → Funding & valuation trajectory; Lens 7 → Cap table & secondary marks; plus a Traction & unit-economics add. There is no earnings print, no P&L, no consensus.

Lens 5 · Funding & Valuation Trajectory (replaces Earnings Result)

Round history, seed → latest (all ``, unaudited):

• Total raised to date: ~$45.3M across 4 rounds. This is small for a hardware/photonics company chasing hyperscaler datacenter sockets (peers have raised $370M–$870M — Lens 7).
• Valuation: ~£137M (≈ $172M at ~1.26 USD/GBP) as of Feb 2025 per Tracxn.
• ⚠️ Data conflict — surfaced, not resolved. A second source (getlatka) reports "$11.6M ARR, $34.8M valuation" for Salience. This contradicts the £137M (~$172M) Tracxn/PitchBook valuation by ~5×, and an $11.6M ARR figure is implausibly precise (and high) for a company that only launched its first product in March 2026 — getlatka figures are frequently algorithmic/estimated and should be treated as low-confidence. I do not pick one; the credible read is "valuation roughly $150–175M post-Series-A, revenue de minimis / pre-material." Anyone underwriting this name must verify the valuation and any revenue directly with the company / a primary cap-table source.

Burn signal: $30M Series A in Feb 2025 to fund a fab partnership (Tower), a test program (Keysight), and a 32→64→128-port roadmap is a thin war chest for that scope. Expect another raise within ~18–24 months of Feb 2025 (i.e., by ~mid/late-2026) to reach volume — a near-term financing-risk flag.

Lens 6 · Founder & Public-Voice Sentiment (replaces Earnings Calls)

No earnings calls exist. Substitute: founder interviews and the cadence/confidence of public messaging.

• Tone has shifted from "compute moonshot" to "infrastructure pragmatism." v1 messaging (2022) was exascale-compute, "stack 64 vectors in a beam of light" — ambitious, science-forward. v2 messaging (2025–26) is disciplined and customer-problem-framed: "moving networks from electronic packet routing to highly predictable, energy-efficient optical connectivity," "fundamental networking bottlenecks," explicit power/latency numbers. This is a healthy maturation — the company sounds like it found a real, fundable problem.
• Kewada's public narrative stresses latency, designing for future AI demand, and (revealingly) "the importance of human relationships and mentorship in navigating obstacles" and that "building … hasn't been without its challenges". Read positively: candor. Read skeptically: a four-year-old company on its second strategy talking about "obstacles overcome."
• Recurring 2026 phrases: "all-optical," "optical circuit switch / OCS," "AI datacenter networking layer," "8× power," "tail latency." Dropped from the lexicon: "on-memory compute," "amplitude-based," "exascale compute," "AI processor" — the v1 vocabulary has been retired entirely, confirming the pivot is complete, not hedged.

Lens 7 · Cap Table & Secondary Marks (replaces Comps)

Syndicate quality (the IPO-proximity tell): mixed-to-good for a Series A, but no crossover funds (no Fidelity / T. Rowe / Coatue), which are the marks that signal IPO proximity. What it does have is strategic semiconductor capital — Applied Ventures (Applied Materials) is a high-signal strategic for a fab-dependent photonics company (fab-equipment maker validating the process), plus ICM HPQC Fund, Braavos, Oxford Sciences Enterprises, Cambridge Innovation Capital, Silicon Catalyst, and angel Jalal Bagherli (ex-Dialog Semiconductor CEO). Strategics > tier-1 financial VCs here — but the absence of crossover money + the small round size says early, not pre-IPO (readiness ≈ 2/5 growth-stage on the private-watch scale; it is not in private-watch.json, this is my ``).

Photonics-peer valuation table — the real "comps." This is where the company's smallness is starkest. All ``, unaudited:

Public OCS-incumbent comp (the demand-side reality check): Lumentum (LITE) is guided to ~$100M/quarter in OCS revenue by end-2026 — i.e., a single public incumbent will out-revenue Salience's entire enterprise value in roughly one to two quarters. Calient, Coherent (COHR), Polatis/Huber+Suhner, Telescent, DiCon, Drut round out the field.

Read: Salience is the smallest credible name in AI-datacenter photonics by ~20–25× on valuation and ~8–19× on capital raised vs. the leaders. That is simultaneously the bear case (under-capitalised for the fight) and the only conceivable bull case for an entry (asymmetric upside if the OCS approach wins and it gets acquired — Celestial→Marvell at $3.25B is the template).

Lens 8 · Funding & Product Catalysts (replaces Stock-Price Catalysts)

No tradeable stock, so "what moves the story" = the funding/product events that re-rate a private:

• May 2022 — $11.5M seed; v1 compute launch (validates team).
• Apr 2023 — hired ex-Imagination Technologies CPO to lead productization (commercial intent).
• 7 Feb 2025 — $30M Series A + the pivot to OCS becomes public (the re-rating event; £137M valuation).
• 25 Feb 2026 — Tower Semiconductor manufacturing partnership → "pre-production" (de-risks fab).
• 10–11 Mar 2026 — 32-port OCS product launch (first real product).
• 12 Mar 2026 / OFC 2026 — Keysight test-environment collaboration (ecosystem).

Pattern: the value inflections track funding rounds and ecosystem/manufacturing partnerships, not yet revenue or design wins. The next catalyst that matters is binary and not yet on the board: a named hyperscaler/neocloud design win or a first production order. Until that prints, every catalyst to date is input-side validation, not demand proof.

---

Phase C — Judge people & books

Lens 9 · Management

• Vaysh Kewada — Co-Founder & CEO. MSc Physics (Imperial College London) → McKinsey (business analyst, 2017) → Entrepreneur-in-Residence at Oxford Sciences Enterprises (~18 months incubating the spinout with the academic founders) → founded Salience 2021. Forbes 30 Under 30 (2023). Archetype: commercial/operator founder (physics-literate but not the inventor) — a good fit for a deep-tech company that must cross from lab to fab to customer, which is exactly the v1→v2 transition she has led. The pivot itself, executed cleanly with a top-tier strategic round closed into it, is the strongest single data point on management quality.
• Prof. Harish Bhaskaran — academic co-founder (Oxford). Pioneer of phase-change photonic computing; FREng; led Oxford's Advanced Nanoscale Engineering group. ⚠️ Red flag (attention/skin-in-the-game): Bhaskaran is reported to be "currently on leave from Oxford and serving as a Director at Apple Inc.". If accurate, the scientific founder's day-job is now Apple, not Salience — a material dilution of founder scientific leadership at a company whose entire edge is photonics IP. Verify directly — but if true it is a genuine governance/continuity concern.
• Prof. Wolfram Pernice — academic co-founder (Münster). Integrated-photonics / photonic computing researcher; co-author of the foundational Nature work with Bhaskaran. Academic, not operational.
• Dr. Johannes Feldmann — CTO / co-founder. Cited as technical co-founder from the Münster side. Operational technical leadership.
• Ex-Imagination Technologies CPO — hired Apr 2023 to lead productization; brought semiconductor product discipline.
• Jalal Bagherli (ex-Dialog Semiconductor CEO) — investor/advisor; credible semiconductor operator on the cap table.

Capital-allocation history: with ~$45M raised and one product shipped, the team has been capital-efficient (Lightmatter/Ayar spent ~$850M+ to reach scale; Salience reached a launched product on ~5% of that). Whether that is admirable discipline or under-investment depends entirely on whether $45M can fund a hyperscaler-grade manufacturing ramp — it almost certainly cannot, so another raise is the implicit next allocation decision.

Founder vs. professional manager: founder-led (Kewada is CEO + Director), with founders still central — appropriate for the stage. Headcount ≈ small ("7 executives" cited; total headcount not disclosed but consistent with a ~30–60-person deep-tech startup). n/a — exact headcount not disclosed.

Lens 10 · Forensic Red Flags

No financial statements exist — Salience is private with no audited filings, so classic forensic accounting (revenue recognition, receivables vs. revenue, SBC dilution, goodwill) is not assessable — n/a — private, unaudited. The red flags that do apply to a private deep-tech name:

1. Strategy pivot (the big one). A four-year-old company changed its core product from optical compute to optical switching. Cleanly explained and arguably into a better market — but pivots this fundamental are inherently a yellow flag: prior investor capital (the 2022 seed) funded a thesis that did not reach product. (Lens 13 presses this.)
2. Scientific-founder attention. Bhaskaran reportedly on leave from Oxford / Director at Apple — see Lens 9. Founder-IP continuity risk.
3. Single-source manufacturing. Total dependence on Tower's fab and III-V yield (Lens 2) — concentration risk with no disclosed second source.
4. No disclosed customers/revenue. Marketing claims (8× power, 80% TPS/user) are vendor-stated, un-third-party-validated; the Keysight collaboration exists precisely because OCS test methodology is immature, so the performance numbers should carry a "manufacturer's claim" discount until independently benchmarked.
5. Valuation/revenue data conflict (Tracxn ~£137M vs. getlatka $34.8M/$11.6M ARR) — unreconciled public data (Lens 5); underwrite only on primary-sourced figures.
6. Under-capitalisation vs. lane — a financing-risk flag, not an integrity flag.

Regulatory findings (required sub-section). Per regulatory/regulatory-findings.md (fetched 2026-06-24): total_sec_findings: 0 — Salience has no CIK and is not required to file with the SEC, so no EDGAR LR/AAER search is possible ``. Non-SEC web search — "Salience Labs" (FTC OR DOJ OR FDA OR CFPB OR "consent decree" OR settlement OR fine OR penalty) enforcement and "Salience Labs" lawsuit/litigation/patent dispute — returned no material regulatory, enforcement, litigation, IP-dispute, or layoff findings as of 2026-06-24. There is no 10-K Item 3 to quote (private). Conclusion: no material regulatory or legal findings — verified via SEC EDGAR EFTS (no CIK → none possible) and web search as of 2026-06-24; all "clean" findings are limited by the absence of audited/public disclosure (unaudited per public sources).

---

Phase D — Project & stress-test

Lens 11 · IPO-Readiness & Path-to-Tradeable (replaces Forward Projection)

Salience is not in research/private-watch.json, so there is no pre-seeded stage/ipo_readiness/catalyst row to lift. The following is my ``, derived from the web record. (No forecast.ts Brier line is logged — --watchlist rule, and there is no scoreable EPS/binary-readout metric for a private switch startup at this stage.)

Stage / readiness (``): ~2/5 (growth-stage) on the private-watch scale (1=seed … 5=S-1/IPO-imminent). It has a Series A, a launched product, and manufacturing/test partnerships — past pure seed — but it is years and at least one large growth round away from S-1 readiness. No crossover-fund marks, no disclosed revenue scale, single product just launched.

Milestones that must print to unlock a tradeable path (in order of importance):

1. A named hyperscaler / neocloud design win or first volume production order for the 32-port OCS — the binary that converts "interesting science" into "fundable business." Not yet achieved; this is the gating catalyst.
2. Independent third-party validation of the 8× power / 80% TPS/user claims (the Keysight test environment is the enabler).
3. Tower ramp from "pre-production" to qualified volume at acceptable yield (esp. III-V laser integration).
4. 64/128-port products shipped — proving the roadmap scales to hyperscaler radix.
5. A growth round (Series B) with crossover participation — the IPO-proximity tell that is currently absent. Likely needed by ~mid/late-2026 given the thin $30M Series A.

Most probable "tradeable" outcome (``): acquisition, not IPO. The template is Celestial AI → Marvell (~$3.25B, late 2025) and the pattern of semiconductor majors buying photonics IP (AMD/Intel/Nvidia all on Ayar's cap table; Applied Materials' VC arm already inside Salience). A standalone OCS-switch IPO is unlikely at this scale; an acqui-hire/strategic acquisition by a networking/optics major (Marvell, Broadcom, Coherent, Lumentum, Nvidia, or Applied) is the realistic liquidity path if the technology proves out — and an acqui-talent down-case if it does not.

Write-back note: Salience is absent from private-watch.json; I am not editing that file in this --watchlist wave (wave boundary — no research/*.json edits). Recommend a follow-up conversational step to add salience-labs to the optical-computing private-watch ledger with readiness: 2, catalyst: "first hyperscaler OCS design win / Series B", and this dossier path, so privates.ts shows it dossier-warm.

Lens 12 · Bull vs Bear

Bull case. The OCS lane is one of the clearest secular wins in AI infrastructure — removing OEO conversions saves power (8×-class) and latency at exactly the moment hyperscaler networks are power- and latency-bound, and credible analysts call all AI-datacenter interconnect going optical within five years. Salience has (a) a real, shipped 32-port product with a coherent 64/128 roadmap; (b) a genuine technical angle vs. incumbent MEMS — solid-state silicon-photonic switching with no moving parts, implying faster reconfiguration and longer lifetime than Google/Lumentum/Calient MEMS; (c) high-signal strategic capital (Applied Materials' VC arm) and a de-risking fab partner (Tower) + test partner (Keysight); (d) capital-efficient, founder-led execution that navigated a clean pivot into a better market. The asymmetric prize: at ~$172M post-money it is ~20-25× cheaper than Lightmatter/Ayar — a single hyperscaler design win or a Celestial-style strategic acquisition could re-rate it multiples higher.

Bear case (2–3 things that permanently impair it).

1. Nvidia eats the lane. Nvidia announced Spectrum-X and Quantum-X silicon-photonics co-packaged-optics switches (Quantum-X InfiniBand GA early 2026, Spectrum-X Ethernet 2H-2026) — 115–400 Tb/s class, with TSMC/Coherent/Corning/Lumentum/Foxconn as partners — explicitly to "scale AI factories to millions of GPUs". When the company that owns the GPU fabric ships its own integrated photonic switching, a 35-person merchant OCS startup's addressable socket shrinks toward the non-Nvidia / disaggregated fringe. This is the existential bear.
2. Incumbent MEMS is "good enough" and already deployed at scale. Google has tens of thousands of OCS ports live; Lumentum is heading to ~$100M/qtr OCS revenue. Salience must prove its silicon-photonic switch is enough better to displace a working, shipping, lower-risk MEMS incumbent — a high bar for a first-gen product with no production reference.
3. Under-capitalisation + single-source fab. ~$45M raised vs. peers' $370–870M; one fab (Tower); III-V yield risk; a near-term forced raise. Any stumble in the ramp, or a down-round in a cooling photonics-funding tape, could starve it before the design win lands.

Pre-mortem (18 months out, thesis broke — what happened?): Nvidia's photonic switches (and Marvell-via-Celestial, Broadcom CPO) captured the hyperscaler OCS sockets; Salience won zero named volume customers; the 32-port's vendor performance claims didn't survive independent benchmarking or Tower yields stayed sub-economic; the company was forced into a flat/down Series B or a soft acqui-hire by a larger optics player at a fraction of a strategic price.

Are the multiples too high? On a standalone basis ~£137M for a pre-revenue, single-product company is rich; on a strategic-optionality basis (acquisition comps at $3.25–4.4B) it is cheap conditional on a design win. The entire valuation is an option on one binary.

Contrarian view (what the market may be missing): the consensus frame is "photonics startups = optical compute (Lightmatter/Lightelligence) or co-packaged interconnect (Ayar/Celestial)." OCS switching is a distinct, less-crowded, arguably more-defensible sub-lane with a clear incumbent (MEMS) ripe for solid-state disruption and a proven exit (Celestial→Marvell). Salience pivoting into that specific gap, early, capital-efficiently, with Applied Materials' strategic backing, is an under-appreciated position — if it can survive Nvidia's shadow long enough to land one anchor customer.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case:

• The pivot is the tell. A company does not abandon its founding product (optical compute) four years and one funding round in because the market got more exciting — it does so because it was losing. Lightmatter ($4.4B) and Lightelligence out-raised and out-shipped the compute lane; Salience quietly exited it and re-badged as a switching company. The charitable "application-pull" narrative papers over a thesis that did not reach a paying customer. What is the evidence v2 won't meet the same fate?
• Revenue is concentrated in zero. There are no disclosed customers. Every performance number (8× power, 80% TPS/user) is vendor-marketing, un-benchmarked by an independent third party — and the company is co-developing the test methodology itself with Keysight, which means there was no neutral way to verify these claims when they were made.
• The most dangerous competitor bulls underestimate is the one that owns the rack: Nvidia. Bulls treat OCS as a green-field startup lane. It is not — Nvidia is shipping co-packaged photonic switches into its own fabric in 2026, Lumentum is at $100M/qtr, Google builds its own, Marvell bought Celestial for $3.25B. Salience is the least-capitalised entrant in a lane being paved by the four best-resourced players in datacenter silicon.
• Capital allocation / governance: the scientific founder (Bhaskaran) appears to have decamped to Apple — at a company whose only asset is photonics IP. And $30M is not a "win a hyperscaler socket" war chest; it is a "get to a demo and raise again" war chest, which hands negotiating leverage to whoever leads the next (possibly down) round.
• What must hold for the price: that silicon-photonic OCS decisively beats MEMS in production and that at least one hyperscaler chooses a tiny merchant startup over Nvidia/Lumentum/in-house and that Tower yields are economic and that a Series B closes on good terms. If growth disappoints 20–30% (read: the first product slips or wins no anchor customer), the £137M is unsupported and a down-round or distress sale follows.
• The single scenario that permanently impairs it: Nvidia's 2026 photonic switches + Marvell/Celestial + Broadcom CPO lock up the hyperscaler OCS sockets before Salience lands one — and a 35-person startup with ~$45M cannot outspend its way back in. Plausibility: moderate-to-high — the timing of Nvidia's roadmap is the live threat.

Lens 14 · Management Questions (ordered by information value)

1. Do you have any named, paying hyperscaler or neocloud design wins or production orders for the 32-port OCS, and what is committed volume/revenue for 2026–27? (The whole thesis.)
2. How does your silicon-photonic OCS win sockets against Nvidia's Spectrum-X / Quantum-X co-packaged photonic switches arriving in 2026 — where specifically do you fit, and is it inside or outside Nvidia-fabric accounts?
3. Have the 8× power and 80% Tokens-per-Second-per-User figures been independently third-party validated (e.g., via the Keysight environment), and can you share the benchmark methodology and against which incumbent (MEMS R300? Apollo?)?
4. What is current cash runway from the $30M Series A, and when/how large is the next raise — and are crossover or strategic investors already engaged?
5. Why did you pivot from on-memory optical compute to optical switching — candidly, what did the compute product fail to achieve commercially, and what de-risks the switching bet from repeating that?
6. Is Tower your sole foundry? What is the current III-V-laser-integration yield, where are you on the "pre-production → qualified volume" path, and what is the second-source plan?
7. What is the reconfiguration time of your switch vs. MEMS (tens of ms) — quantify the latency/switching-speed advantage and why it matters to a real workload.
8. What is Prof. Bhaskaran's current role and time commitment to Salience given his reported Apple directorship — who now owns photonics IP leadership day-to-day?
9. What is the realistic liquidity path — strategic acquisition (à la Celestial→Marvell) vs. standalone scale — and which acquirers' roadmaps does your IP most complement?
10. Reconcile the public valuation/revenue data: is the ~£137M post-Series-A valuation correct, and what was 2025 revenue (the $11.6M-ARR figure circulating appears wrong)?
11. What is your defensible moat in 24 months once Tower's platform (PH18DA/TPS45PH) is available to any fabless competitor — IP, process, or system integration?
12. At what port radix (64? 128?) does the product become hyperscaler-relevant, and what is the shipping timeline for each?
13. What is the BOM/unit economics and target gross margin at volume vs. an incumbent MEMS OCS?
14. What is current total headcount, the hiring plan, and your most critical talent gap (PIC design? III-V? customer/field engineering)?
15. What does an OCS deployment require operators to change in their network architecture/software, and how does that switching cost help or hurt adoption?

---