Market cap: ≈$30-40M USD (subject to confirmation of post-2025 raise diluted share count) Coverage: No US sell-side. One Nordic broker historically.

Setup

Napatech is a Danish FPGA SmartNIC company on the Oslo Bors. Q1 2026 revenue $5.7M, +69% year-over-year. FY2026 guide unchanged at $32-38M (vs. $22.4M FY2025). Pre-breakeven, ≈6-7 quarters of cash runway. The May 7 earnings call surfaced two structural facts that don't appear to be in the price: Napatech is the FPGA hardware inside d-Matrix's JetStream NIC, and the Tier 1 server OEM relationship first disclosed in April 2024 has progressed to "commercial stage" — the third of three alignment phases. At ≈1x forward revenue, the market is treating Napatech as a flat-line component supplier.

What the filing says

Q1 2026 revenue $5.7M (+69% YoY). Gross margin 70%. EBITDA -DKK 18M, improved from -DKK 29M in Q1 2025. Cash >DKK 120M (≈$17M). Did not draw on credit facility. FY2026 guidance unchanged: $32-38M revenue, GM 60-70%, staff/external costs DKK 170-180M.

CEO framed AI infrastructure as transitioning from compute-limited to efficiency-limited: "AI infrastructure is no longer a compute problem. It is an efficiency problem, which immediately translates to 2 components of the data center: networking and memory."

The d-Matrix architectural confirmation comes from d-Matrix CEO Sid Sheth, on the public Napatech press release announcing the d-Matrix scale-out NIC partnership: "Partnering with Napatech and Altera on JetStream enabled us to deliver efficient, ultra-low latency AI NIC." JetStream is a PCIe Gen5 400G card with sub-2-microsecond latency. d-Matrix designs and brands; Napatech supplies the FPGA hardware. The SquadRack reference architecture, announced at OCP Global Summit October 2025, lists JetStream as the scale-out NIC alongside d-Matrix Corsair compute, Broadcom PCIe switches, and Arista Leaf Ethernet. Distributed via Supermicro starting Q1 2026.

Two pieces of guidance language matter. First, on the conservatism: CEO disclosed FY2026 incorporates only 20% of d-Matrix's own forecast: "we have only taken 20% of that in." Second, on the OEM: the three-phase alignment model is executive / product+engineering / commercial. The CEO confirmed all three groups are now engaged. Use case is data reduction and deduplication for LLM loading via Napatech's Altera Agilex 7-based N3070X NIC. Identity undisclosed.

Core business is recovering. Five new design wins in Q1 across four verticals. A fintech tier-1 bank multiyear deal lands in Q2 ($1.5M+, described as repeatable). FY2025 closed at $22.4M (+33% YoY) on 27 design wins.

What the market thinks

At ≈1x forward revenue on $35M FY26 midpoint, the stock trades at a discount to the closest peer, Silicom (SILC), which trades around 2x forward revenue with no commercial AI traction yet — SILC's CEO said on the April 30 call that significant AI inference revenue is "more 2027 than 2026." The peer comp set is thin (effectively two public FPGA NIC pure-plays); take the multiple anchor with that caveat.

We don't have liquid options on Oslo Bors to extract risk-neutral probabilities. The "implied" probabilities below are estimated, not derived — they reflect what a 1x revenue multiple suggests the market is currently underwriting versus our reading:

The wedges concentrate in optionality: Tier 1 OEM identity reveal, d-Matrix hyperscaler announcement, and forward visibility into 2027.

Probability-weighted view:

EV-weighted forward return ~+68% over 12-18 months. Single-name Sharpe modeled at ≈0.7 unhedged.

Why the gap exists

The structural reasons the price doesn't reflect what the call surfaced:

• No US sell-side coverage. Oslo Bors primary, OTC ADR. No US screen catches it.
• d-Matrix is private. Commercial traction is unverifiable through public channels. d-Matrix was absent from MLPerf Inference v6.0 (April 2026), which had 24 organizations submit including NVIDIA, AMD, Google, Dell, Cisco, HPE, and Intel. That absence is itself information — a non-NVIDIA inference contender with hyperscaler ambitions normally shows up to the benchmark fight.
• The Tier 1 OEM reads as stale. First disclosed April 2024. To anyone scanning headlines, "2-year-old engagement, no production yet" — the Q1 2026 progression to commercial stage is new but easy to miss.
• The architectural lock-in requires synthesis. It surfaces only when you cross-reference Napatech's transcript with the d-Matrix CEO statement, the d-Matrix product page, and the SquadRack OCP announcement.

What caps the upside

The TAM is bounded. FPGA SmartNIC for AI inference is a niche tied to non-NVIDIA AI compute deployment — d-Matrix, Cerebras, Etched, Groq, Tenstorrent, SambaNova. The standard hyperscaler AI server stack is NVIDIA ConnectX-7/8 + BlueField + Astera Labs Scorpio + Marvell switches; that path is closed by NVIDIA bundling. We estimate combined revenue ceiling for FPGA NIC vendors at $200-500M by 2027+. There is no $20 path on this stock — the bull case caps at roughly $5.50-9.50 and requires the non-NVIDIA AI inference ecosystem to actually scale.

Risks (ranked)

1. AI orders don't materialize H2 2026. ≈$6-7M of the FY26 guidance is AI-dependent and entirely H2-weighted. The CEO was explicit: "if the AI orders don't materialize, then that's what the impact is going to be." A 15-20% guidance miss compresses runway materially.
2. d-Matrix commercial stall. Funding crisis, customer loss, or vertical integration of NIC hardware in-house. The April 2026 GigaIO acquisition expands d-Matrix's rack-scale capability and raises a 2-3 year vertical-integration risk that is not present today.
3. Tier 1 OEM remains commercial-stage indefinitely. "Commercial" does not mean signed production order. CEO Q1 2026: "when that converts into something meaningful remains to be seen."
4. Gross margin compression. Guidance band 60-70%. Q1 2026 was 70% on favorable mix; AI custom work has historically been lower margin. Trajectory toward 60% damages operating leverage even on revenue beats.
5. FPGA inventory pre-buy write-down. The company is pre-buying components at extending lead times. If AI orders miss, inventory sits on the balance sheet through H1 2027 with write-down risk and a working capital drag.

Catalysts

What would change our mind

Strengthens the read:

• A SMCI 8-K or Q3 FY26 transcript references Napatech, FPGA SmartNIC, or Altera in an AI server context — would resolve the OEM identity to the highest-probability candidate
• d-Matrix announces a named hyperscaler (AWS, GOOGL, MSFT, META, ORCL)
• d-Matrix submits MLPerf v7.0 with credible numbers (within 2x of NVIDIA on key benchmarks)
• Norwegian PDMR notifications show insiders buying NAPA.OL at current levels

Kills the read:

• d-Matrix down round at <$2B valuation, or a public funding gap
• Tier 1 OEM engagement officially stalled or chooses a competitor (Marvell, Nvidia)
• FY2026 guidance cut at the Q2 print, or Q2 revenue <$6.5M
• Equity raise announcement at <$1.80
• d-Matrix vertically integrates NIC hardware using GigaIO IP

Evidence

Open items not addressed in this memo: NOK 200M raise dilution math (diluted share count uncertainty widens the per-share target ranges), credit facility covenants, exact 2025 quarterly revenue cadence for cleaner H2 seasonality modeling.