Optical Computing
PrivateReal structural node in the scarce InP-laser chokepoint, but a ~$33M-revenue, going-concern-flagged, restatement-tainted micro-cap is now priced at a ~SEK 29bn / ~31x P/S momentum bubble — the technology is investable, the stock at this price is not. BEARISH on the equity, WATCH the business.
Research
The verdict
Real structural node in the scarce InP-laser chokepoint, but a ~$33M-revenue, going-concern-flagged, restatement-tainted micro-cap is now priced at a ~SEK 29bn / ~31x P/S momentum bubble — the technology is investable, the stock at this price is not. BEARISH on the equity, WATCH the business.
Sivers Semiconductors AB is a Swedish dual-division compound-semiconductor and photonics company — not an "optical-computing" pure-play (the coverage bucket is a misnomer). It is a picks-and-shovels supplier of the light sources and RF front-ends that other people's AI, autonomy and connectivity systems need. Two business areas ``:
Plain-terms business model. Sivers sells (a) physical InP laser chips/arrays it fabricates, and (b) RF chip designs it has a foundry build — increasingly via NRE / development programs that (the bulls hope) convert to volume production in 2027–2028. Today the revenue is small and lumpy: ~SEK 304M / ~$33M FY2025, +25% y/y , of which *product* revenue was only **~SEK 85.7M** (+13% FX-neutral) — the rest is development/NRE/services .
Customers / partners (named). CPO: Ayar Labs (laser arrays for its "SuperNova" light source), POET Technologies (optical-interposer co-development), O-Net Technologies + Enablence (ELSFP module ODM), Jabil (1.6T LRO transceiver using Sivers DFBs), GlobalFoundries (June 2026 SCALE CPO platform reference designs). LiDAR: a "strategic customer" widely suspected to be Aeva Technologies (NVIDIA DRIVE Hyperion). SATCOM: ALL.SPACE (~$9k Sivers content/terminal), Intelsat. Wireless: a Tier-1 telecom OEM (likely Nokia, unconfirmed), plus defense names (Northrop Grumman, Raytheon, BAE, Ericsson) ``.
Contract structure. No take-or-pay, no recurring SaaS. Mostly project/design-win economics: NRE up front, then per-unit product revenue if and when the customer's own product ships. The headline "opportunity pipeline" ($453M end-2025 → $799M YTD Q1 2026, +77%) is explicitly a 5-year, non-binding funnel, not backlog — contract liabilities on the balance sheet were only ~SEK 11.6M ``. That distinction is the whole game (see Lens 13).
Sivers sits in the middle of two distinct chains. Names, not generics:
Photonics (in-house fab — the differentiator):
InP wafers (IQE plc, Cardiff/Taiwan) → MOCVD epitaxy (Aixtron tools) → Sivers Photonics fab, Glasgow — etched-facet, wafer-scale InP laser/SOA fabrication & test → optional co-fab / second-source qual (WIN Semiconductors, Taiwan) → module integrators (O-Net ELSFP, POET interposer, Jabil transceivers, Ayar Labs light source) → hyperscaler / switch OEM end demand (CPO for AI clusters) ``.
Wireless (fabless):
RF/mmWave design (Sivers) → GlobalFoundries 22FDX (Dresden/Malta) → module/antenna-array assembly → 5G infra OEM / SATCOM terminal maker (ALL.SPACE, Intelsat) / defense prime. Chokepoint here is foundry allocation at GF, shared with the entire RF industry. CHIPS Act money (initial ~$11M award, late 2024) part-funds US wireless work ``.
Verdict on the chain: the in-house InP fab is the one genuinely scarce, hard-to-replicate asset on the map — Lumentum and Coherent are the only other Western names with comparable scaled InP laser capability (Lens 3/7). But Sivers' capacity is a rounding error next to theirs, so it is a structural node, not a structural supplier.
What is genuinely defensible:
Where the moat is thin / overstated:
Net: a narrow, real process moat (one of three Western InP laser houses) wrapped around a balance sheet too weak to fully exploit it. The technology has a moat; the company's ability to monetize it at scale is the open question.
segments.csv is empty (web-only), so this is / throughout — no research-layer segment numbers exist to cite.
| Segment | What it is | FY2025 read | Trend |
|---|---|---|---|
| Photonics | In-house InP lasers/SOAs — CPO, LiDAR, SATCOM, sensing | The growth + narrative engine. Achieved its "first profitable full year" at ~SEK +1.5M adjusted EBITDA `` | Accelerating — all the 2026 catalysts (GF, O-Net, Jabil, Ayar, LiDAR ramp) live here |
| Wireless | Fabless mmWave/RF for 5G, SATCOM, defense | NRE-heavy cash ballast; hit by US-government-shutdown/defense-budget delays into Q1 2026 | Lumpy / flattish — repeatable product sales still thin vs. NRE |
. Historical 2016–2025 CAGR cited at **~36.7%** off a tiny base .| Metric | Q1 2026 | Comp | Source |
|---|---|---|---|
| Net sales | SEK 61.9M | −22% y/y | `` |
| EBIT | SEK −41.5M | loss | `` |
| Net result | SEK −42.7M | loss | `` |
| Operating cash flow | SEK −49.2M | cash burn | `` |
| Opportunity pipeline | $799M | +77% YTD (non-binding 5-yr funnel) | `` |
FY2025 context ``:
Balance-sheet flags (this is where it gets serious):
— against ~SEK 50–66M annual cash burn, i.e. **well under a year of runway without raising.** The company has raised serially to stay alive: **SEK 108M (Jan 2025), SEK 95M (Sep 2025), ~$17M (Feb 2026: $5M term loan + $12M convertible), SEK 125M (Apr/May 2026 directed issue at SEK 14.50)**.Market reaction / what's priced in: the stock did not fall on the −22% Q1 print — it went up into a parabola, because the tape is being driven by the GlobalFoundries CPO deal (Jun 2026), index inclusion flows and a short squeeze, not by the income statement. The fundamentals deteriorated sequentially while the multiple expanded ~20x. That is the definition of a narrative-driven melt-up. Flag: unusual vs. the company's own history in every respect.
No transcripts on the shelf (transcripts/ empty); this is from report commentary.
Peer table — InP-laser / silicon-photonics / CPO names.
| Company | Ticker | Mkt cap (USD) | EV/Sales | P/E | Notes | Source |
|---|---|---|---|---|---|---|
| Sivers Semiconductors | SIVE.ST | ~$2.7–3.0bn (SEK ~29bn) | ~31x P/S (``); ~4x NTM was the pre-parabola read | n/a (loss-making) | ~$33M rev, going-concern, restated accounts | `` |
| Lumentum | LITE | ~$69.5bn | n/a (EV/EBITDA ~90x cited) | ~114x (Q3 FY26) | Only supplier shipping 200G/lane EMLs at volume; NVIDIA ~$2B investment | `` |
| Coherent | COHR | n/a | n/a | n/a | FY25 rev $5.81bn (+23%); ~25% transceiver share; NVIDIA Spectrum-X collaborator; ~$2B NVIDIA investment; ramping 6-inch InP across 4 sites incl. Järfälla SE | `` |
| POET Technologies | POET | n/a | n/a | n/a (loss-making) | Optical-interposer platform; ~$5.6M Q3'25 production orders; >$300M cash; Foxconn partner | `` |
Read: on EV/Sales ~31x, Sivers trades at ~7.6x the European semiconductor peer-group average (~4.1x) and ~1.8x the closest-peer average (~17.5x) — and it does so as the *smallest, loss-making, going-concern* name in the set. Lumentum and Coherent are expensive *and* shipping at volume with NVIDIA capital behind them; Sivers is expensive on *future optionality* with a sub-year cash runway. The comp set says the technology theme is richly valued everywhere — but Sivers is the most extreme expression of it with the weakest balance sheet. MACOM has also been cited as "the closest analog — a micro MACOM"; MACOM multiples n/a.
Mostly ``. The 2026 tape is a near-vertical sequence of partnership-announcement spikes:
Pattern the market reacts to: named partnerships with Tier-1/AI-adjacent counterparties (GF, NVIDIA-ecosystem via Aeva/Ayar, Jabil) move this stock far more than the income statement. Accounting red flags and a declining revenue print were ignored in 2026 — a hallmark of a momentum/retail/index-flow regime where good fundamentals aren't required and bad ones don't bite. That asymmetry cuts both ways: the same insensitivity that ignored the going-concern flag on the way up will amplify the unwind when the narrative cadence breaks.
insider-transactions.csv absent) — n/a. The serial directed share issues (to institutions, not insiders) and a convertible suggest the cap table is being broadened for the US listing, not founder-concentrated.This is the richest lens in the file. Sivers carries an unusually dense cluster of accounting/governance flags for a company at this valuation.
Income statement / revenue recognition:
Cash vs. earnings / balance sheet:
Restatement (material): the company restated/"audit-uplifted" its 2023, 2024 and 2025 consolidated accounts to align with US PCAOB standards ahead of a potential Nasdaq New York dual-listing, and delayed the 2025 annual report (Apr 27 → May 15) to do so ``. A multi-year restatement is, by itself, a reason to widen the risk discount — prior reported figures were not US-GAAP/PCAOB grade.
Going concern (material): auditors issued a going-concern qualification in the 2025 annual report, explicitly doubting the company's ability to continue without external support ``. For a company the market values at ~SEK 29bn, an auditor going-concern flag is a glaring dissonance.
Regulatory findings (required sub-section):
Summary: the combination of going-concern + multi-year restatement + an active criminal leak probe + serial dilution + a non-binding "pipeline" headline + a declining latest revenue print is a forensic constellation, not a single flag. None of it is necessarily fraud — the restatement is plausibly benign PCAOB-uplift, the probe is about a leak not the accounts — but the density of flags at a ~31x P/S valuation is the most important risk signal in this dossier.
Web-only; no financials.csv/guidance.csv to ground a model — all lines /, arithmetic shown. Per SKILL.md --watchlist rules, no forecast.ts create is logged from this unattended run. Sivers reports in SEK; revenue base SEK 304M FY2025 ($33M).
Revenue path (the company is the input, not EPS — it's loss-making):
(FY28E rev SEK 646M, EBIT SEK +58.9M, net income SEK +47.9M — i.e. *first* full-company profit). cross-check: SEK 304M × (1.30)^3 ≈ SEK 668M — consistent with the base.EPS: n/a / not meaningful. The company is loss-making with a moving (rising) share count; no credible 3-year EPS series exists in sources. Forcing one would be fabrication.
The number that actually matters (runway, not EPS): cash ~SEK 43.5M end-2025 + ~$17M (Feb) + SEK 125M (May) ≈ ~SEK 350–400M post-May-raise against ~SEK 50M/quarter burn ⇒ ~6–8 quarters of runway to the first real CPO/LiDAR volume in 2027–2028 ``. The thesis lives or dies on whether product revenue inflects before the cash (and the narrative) runs out — and whether the next raise is at SEK 80 or SEK 15.
Brier forecast (logged for tracking, not created here per watchlist rule): the scoreable binary would be "Sivers Photonics CPO/ELS product revenue (not NRE) exceeds SEK 100M in any single fiscal year by FY2028" — base probability ~35–45% ``: the technology and partners are real, but converting design wins to >SEK 100M of recognized product revenue against capacity, qualification and balance-sheet constraints is genuinely uncertain.
Bull case. Sivers is one of only ~three Western houses that can fabricate InP lasers at scale, sitting in a structurally supply-constrained chokepoint (sub-30% yields, demand >> supply) exactly as the AI buildout forces a copper→optics transition and CPO goes mainstream ``. It has four independent revenue vectors — CPO/ELS (Ayar, POET, O-Net, Jabil, GlobalFoundries SCALE), FMCW LiDAR (Aeva → NVIDIA DRIVE Hyperion, $53–138M lifecycle, ramps Q4 2026), SATCOM (ALL.SPACE ~$9k/terminal), and mmWave 5G/defense — any one of which could rival today's entire revenue. Photonics just turned EBITDA-positive for the first time, the pipeline grew to $799M, and a Nasdaq New York dual-listing would re-rate it onto US photonics comps (Lumentum/Coherent at 90–114x earnings, NVIDIA-funded). The contrarian bull: the market still under-appreciates that the light source, not the switch, is the gating component, and Sivers is a pure-play on it.
Bear case (2–3 permanent-impairment risks).
Pre-mortem (18 months out, thesis broke — what happened?): CPO product revenue stayed in "sampling/qualification" through 2027, the LiDAR customer pushed its SOP from Q4 2026 into 2027/28, and the AI-photonics momentum trade rotated. With cash low, Sivers raised at a fraction of the SEK 80–92 peak; the going-concern flag recurred; the Nasdaq dual-listing stalled (audit/probe overhang). The stock round-tripped most of the 2026 move — the business was fine, the 2026 price was a bubble.
Are multiples too high? Yes — emphatically — at ~31x P/S vs ~4.1x European-semi-peer average ``, for a loss-making, going-concern, just-restated micro-cap whose latest revenue print declined. Multiple DCFs cluster at SEK 9.8–14.5 fair value vs ~SEK 78–92 price — an ~80–85% implied overvaluation on cash-flow fundamentals. The valuation prices the bull-2030 scenario as the base case.
Contrarian view (what the market refuses to see): the consensus momentum read is "AI-photonics pure-play, infinite TAM, must-own." What the tape is ignoring is that the most important recent data point — the Q1 2026 −22% revenue print — went the wrong way, and that an auditor going-concern qualification at a ~SEK 29bn valuation is one of the loudest contrarian signals available. The market has decided accounting and current fundamentals don't matter for this name; that decision is itself the risk.
Dismantling the bull case:
Short-seller's verdict: the company is a legitimate technology asset; the stock is a momentum/index-flow vehicle priced ~5–8x its fundamental value with an auditor going-concern flag and an open criminal probe attached. The business may well be worth owning at SEK 15. It is a short, not a buy, at SEK 80.
A re-rated telecom-equipment turnaround wearing an AI-infrastructure mask — the optical/AI-RAN engine is real and accelerating, but the stock already prices the re-rating (P/E ~90, +109% in 12m) while ~92% of revenue is still slow-growth telecom/IP/patents. Right business, wrong entry.
The unglamorous AI-optics toll-collector — a 12%-gross-margin Thai contract manufacturer the market is now paying a 45× fabless multiple for; the business is genuinely inflecting (DCI +90%, 1.6T capacity doubling) but the price has front-run two years of execution AND ignores that co-packaged optics is a 2027+ structural axe over the whole pluggable-transceiver thesis.
A $2.5B market cap on $682K of FY25 revenue — QUBT is a $1.5B treasury wrapped in a photonics R&D lab, sold as a quantum-computing story; the balance sheet is real, the revenue is not, and a securities-fraud class action over the exact gap between the two is unresolved.