Semiconductors
PrivateA genuine low-power-FPGA franchise riding a real AI-server attach cycle off a cyclical trough — but the tape already prices flawless multi-year execution (~20x a $1B run-rate that doesn't exist yet, ~76x fwd non-GAAP EPS), and the $1.65B AMI deal converts a pristine net-cash balance sheet into ~$950M of fresh debt right at the top. Great business, dangerous entry price. WATCHING, not chasing.
Research
The verdict
A genuine low-power-FPGA franchise riding a real AI-server attach cycle off a cyclical trough — but the tape already prices flawless multi-year execution (~20x a $1B run-rate that doesn't exist yet, ~76x fwd non-GAAP EPS), and the $1.65B AMI deal converts a pristine net-cash balance sheet into ~$950M of fresh debt right at the top. Great business, dangerous entry price. WATCHING, not chasing.
Primary sources
Source documents — open to read in full
Lattice is the last pure-play, low-power, small-and-mid-range FPGA company of scale. Headquartered in Hillsboro, Oregon, it designs field-programmable gate arrays that customers configure in software, and monetises them through chip sales plus IP licensing and design services. It is fabless — TSMC (16nm, for Avant + Nexus 2), Samsung (28nm FD-SOI, for Nexus), UMC/USJC (legacy 40–130nm), and Epson (180–500nm legacy) make the wafers; ASE (primary) and Amkor do assembly/test. 1,174 employees as of Jan 3, 2026.
The thesis in one line: Lattice does NOT compete with the CPU/GPU/XPU — it sits beside it. Its FPGAs do secure boot, power sequencing, platform management, hardware root-of-trust, I/O aggregation and sensor bridging — the "first-on, last-off" control plane of a server, a base station, a robot, a car. CEO Ford Tamer's framing on the Q1-FY26 call: "the everywhere companion chip," doing the auxiliary functions around the main processor rather than the compute itself.
Products. Two platform generations carry the franchise: small FPGAs on Nexus (28nm FD-SOI) and Nexus 2 (introduced 2024) — the CrossLink (video/USB), Certus (general-purpose), MachXO5 (secure control, including the industry-first CNSA-2.0 PQC part) families; and mid-range on Avant (16nm FinFET, introduced 2022) — Avant-E/G/X with up to 25G SERDES and PCIe 4.0. Legacy MachXO/CrossLink/ECP/iCE families have run >20 years and still sell. Software (Radiant, Propel) and solution stacks (sensAI, Sentry, ORAN, Automate, mVision, Drive) sit on top to shorten customer design cycles.
Customers & channel. Sells almost entirely through distributors — 84% of FY2025 revenue, and a striking 94% in Q1 FY2026. End customers are OEMs in three (now reorganised to two) end-markets. Foreign sales were 83% of FY2025 revenue; all billing is USD.
Contract structure. No take-or-pay. Backlog is weak as a predictor — POs cancellable up to 60 days out, plus meaningful same-quarter "turns" business. That said, Tamer says bookings now "extend well into 2027," a notable shift from the normal short-cycle pattern.
The transformational fact (see Lens 5/12 in depth): on May 4, 2026 Lattice agreed to buy AMI (American Megatrends International) — the world's largest BIOS firmware vendor and a leading BMC (baseboard-management-controller) software/hardware supplier (MegaRAC, Aptio UEFI) — for $1.65B ($1.0B cash + ~$650M stock), closing ~Q3 2026. This pushes Lattice from a chip company into a system-level secure server-management platform.
Upstream → Lattice → end customer, every named link:
Chokepoints / single-source dependencies:
Where the moat is real:
Bargaining power. Mixed. Against customers: moderate — design-in lock-in is real, but the 94%-distributor channel means Lattice is one step removed from end-demand and exposed to distributor destocking. Against suppliers: weak-to-moderate — it is a ~$0.7B buyer competing with the whole industry for TSMC/Samsung/ASE capacity; in a tight backend it pays up and waits.
Where the moat is thin: at the low end, Gowin, Efinix and QuickLogic are real and cheaper, and the FPGA-market literature explicitly cites "price erosion in the mid-range and low-end... accelerating commoditization" with Lattice named as one of the aggressors. The moat is the architecture + the install base, not pricing power on a commodity LUT.
Lattice reports as a single operating segment (the core Lattice business), so there is no product-segment P&L — only end-market and geographic disaggregation. In Q1 FY2026 management reorganised end markets from three buckets to two: "Compute and Communications" and "Industrial and Embedded" (Consumer folded into the latter).
By end market:
| End market | FY2023 | FY2024 | FY2025 | Q1 FY2026 (vs Q1 FY25) |
|---|---|---|---|---|
| Comms & Computing | $257.5M (34.9%) | $228.1M (44.8%) | $292.7M (55.9%), +28.3% | $106.6M (62.4%), +86% |
| Industrial & Auto | $433.5M (58.8%) | $236.9M (46.5%) | $194.0M (37.1%), −18.1% | $64.3M (37.6%), +2% |
| Consumer | $46.1M (6.3%) | $44.3M (8.7%) | $36.6M (7.0%), −17.4% | (folded into Industrial & Embedded) |
The story the mix tells: a hard, multi-year inventory correction in Industrial & Automotive (the old growth engine — $433M in FY2023 → $194M in FY2025, a 55% collapse) masked by an explosive data-center/AI-server recovery in Compute & Communications. The recovery flipped the mix: Comms & Computing went from 35% of revenue (FY23) to 62% (Q1 FY26). Industrial bottomed and turned (+2% in Q1 FY26 on industrial/aerospace recovery) — if that re-accelerates on top of the data-center ramp, both engines fire at once.
By geography: Asia 67.6% of FY2025 revenue, rising to 77.6% in Q1 FY26 (+71% YoY, "hyperscaler demand"); Americas 11%; Europe 11.4%. The Asia concentration is a function of where servers/equipment are built, not solely end-demand, but it is a real geopolitical and disclosure-judgement exposure.
The trough-to-inflection arc:
| ($M unless noted) | FY2023 (peak) | FY2024 (trough) | FY2025 | Q1 FY2026 | Q1 FY2025 |
|---|---|---|---|---|---|
| Revenue | 737.2 | 509.4 | 523.3 (+2.7%) | 170.9 (+42.2%) | 120.2 |
| Gross margin (GAAP) | 69.8% | 66.8% | 68.2% | 68.8% | 68.0% |
| Income from operations | 212.3 (28.8%) | 34.5 (6.8%) | 11.2 (2.1%) | 26.1 (15.3%) | 7.0 (5.8%) |
| GAAP net income | 259.1 | 61.1 | 3.1 (0.6%) | 21.8 (12.8%) | 5.0 |
| GAAP diluted EPS | $1.85 | $0.44 | $0.02 | $0.16 | $0.04 |
| Adjusted EBITDA | 324.7 (44.0%) | 162.0 (31.8%) | 183.0 (35.0%) | 67.8 (39.6%) | 40.1 (33.4%) |
| Non-GAAP diluted EPS | $2.01 | — | $1.05 | ~$0.41 | ~$0.22 |
Latest print (Q1 FY2026, ended April 4, 2026): revenue $170.9M, +42% YoY / +17% QoQ — a clean beat; non-GAAP EPS ~$0.41 (+86% YoY). Operating margin nearly tripled YoY (5.8% → 15.3% GAAP) — operating leverage is the story: non-GAAP op margin 34.4%, +370bps, on EPS growing 80% against revenue growing 42%. Driven almost entirely by Compute & Communications +86% (data-center general-purpose + AI servers + wireline networking).
Guidance. Q2 FY2026: revenue $185M midpoint (~+50% YoY), non-GAAP EPS ~$0.44, non-GAAP GM "70% ±1%", opex $64–67M. The midpoint came in ~8.7% above the sell-side estimate — a guide-up, not a maintain.
Balance-sheet flags:
Market reaction / what's priced in: the stock is up ~146% over the trailing year and trades near $143–147 vs a 52-week low of $43.90. The market has fully embraced the inflection — the print confirmed the narrative rather than surprising it.
(No transcripts on disk; sentiment read from the Q1-FY26 call vs the documented arc of CEO changes.)
Tone trajectory: the franchise has moved through three distinct postures in ~24 months — (i) 2023 confidence under Jim Anderson (record revenue, "growing revenue and profitability"); (ii) a 2024 air-pocket — Anderson abruptly left for Coherent in June 2024 (stock −15% on the day), an inventory correction gutted Industrial, and interim leadership steadied the ship; (iii) a sharp turn to offence under Ford Tamer (CEO since Sept 2024).
What management is focused on now (recurring Q1-FY26 themes ):
Confidence signals: "early innings of a multiyear growth cycle"; bookings "extend well into 2027"; CFO "growth this year will be... stronger than we originally thought." Caution signals: backend supply "straining"; cost pressure "to continue and increase in the second half"; channel inventory still normalising ("close to two months," heading "under two"). The honesty about supply strain and 2H cost pressure reads as credible rather than promotional. The "stopped saying": far less about Industrial recovery as the driver; the narrative has fully pivoted to data-center/AI and system-level.
Lattice trades at a structural premium to every FPGA and analog/MCU peer, on every metric. Peer multiples are ``; the FPGA pure-play set is tiny (the duopoly is buried inside AMD/Intel; the small players are private).
| Company | Ticker | Mkt cap | EV/Sales | EV/EBITDA | P/E (fwd) | Notes |
|---|---|---|---|---|---|---|
| Lattice | LSCC | ~$19.9–20.1B | ~35x (LTM); ~20x on the targeted ~$1.0B combined run-rate | ~285x LTM GAAP; ~52x NTM | ~76x | Pure-play low-power FPGA |
| AMD (Xilinx owner) | AMD | ~$880B | n/a — not broken out | ~115x | ~60x | FPGA is a sliver of AMD |
| Intel (Altera) | INTC | ~$658B | n/a | ~57x | ~179x | Sold 51% of Altera to Silver Lake, Apr 2025 |
| Microchip (PolarFire) | MCHP | ~$40B (Jan 2026) | n/a | n/a | n/a | Broad analog/MCU + FPGA |
| ON Semiconductor | ON | n/a | n/a | ~17.8x | ~35x | Analog/power comp, not FPGA |
| QuickLogic | QUIK | small-cap | n/a | n/a | n/a | eFPGA IP, niche |
| Efinix | private | private; ~$55M raised | n/a — private | n/a | n/a | Low-power FPGA challenger |
5-year-avg ROE: n/a as a clean 5y average. Spot ROE is depressed (~2.75% ) because FY2025 GAAP net income was crushed to $3.1M by SBC; on a normalised/non-GAAP basis prior-cycle ROE ran far higher (FY2023 net income $259M on ~$692M equity ≈ ~37% ).
Read: even on the generous ~20x EV / combined-$1.0B-run-rate and ~52x NTM EV/EBITDA, Lattice is priced at a clear premium to the semis group (industry ~26x EV/EBITDA trailing ) and to its own 5-year-median P/E of ~65x. The premium is the growth + margin + AI-attach story. There is no multiple cushion.
What moves LSCC >5%, from the documented record:
Pattern: the market reacts hardest to (1) the cyclical inflection of revenue + margin together, (2) AI-server attach data points, (3) key-person/management events, and (4) channel-inventory direction. It is a high-beta cyclical wearing a secular-growth multiple.
CEO — Ford Tamer (since Sept 2024). This is the central management fact. Tamer ran Inphi for ~9 years, building it into the electro-optics leader for cloud/telecom before its ~$10B+ sale to Marvell — a genuinely elite value-creation track record. Prior: CEO of Telegent; SVP/GM of Broadcom's Infrastructure Networking Group (grew to $1.2B revenue in 5 years); CEO of Agere. He is a proven serial operator and dealmaker in exactly Lattice's adjacency (data-center connectivity/infrastructure silicon).
Other: CFO consistent and credible on the call (margin discipline, de-levering plan). Interim CEO Esam Elashmawi (Chief Marketing/Strategy Officer) bridged the 2024 gap.
The single biggest forensic issue: stock-based compensation has detonated, and it is the entire gap between "great non-GAAP" and "near-zero GAAP."
Other items, labelled:
Regulatory findings (required sub-section):
No forecast.ts forecast logged (watchlist rule). EPS built bottom-up; every input labelled; output ``. Two regimes: standalone Lattice vs pro-forma with AMI (closing ~Q3 2026).
Anchors:
Revenue path:
**Non-GAAP EPS path ** (anchored on FY2025 non-GAAP $1.05 and Q1+Q2 FY26 run-rate ~$0.41+$0.44):
GAAP EPS will remain far lower while SBC runs at 15–22% of revenue and AMI intangible amortisation hits the P&L — likely $0.40–1.00 FY2026 GAAP even as non-GAAP prints ~$1.90. The GAAP/non-GAAP chasm is the crux of the valuation debate.
What the price implies: at ~$145 the stock is ~76x base FY2026 non-GAAP EPS and ~57x base FY2027 — i.e. it already discounts the successful FY2027 pro-forma. To merely hold the multiple, Lattice must hit ~$1.9 FY26 / ~$2.55 FY27 non-GAAP and keep the market paying ~55–75x. Any combination of AI-attach deceleration, AMI integration slippage, or a multiple de-rate toward the group (~30–40x) implies material downside.
Bull case. Lattice is the only scaled pure-play on the secular thesis that AI servers, robots, and electrified everything need more low-power control/security/connectivity silicon next to every processor. The numbers are inflecting violently the right way (revenue +42%, op margin +900bps YoY, +50% Q2 guide), management says "early innings of a multiyear cycle" with bookings into 2027, and the AMI deal is a masterstroke if it works: it converts a chip vendor into a system-level secure-management platform, ~doubles SAM to ~$12B, lands a high-GM/high-EBITDA, immediately-accretive business with no synergy dependence, and slots Lattice's control FPGA + AMI's BIOS/BMC into the exact same server socket (including NVIDIA GB200-class racks). Operating leverage is structural — 70% GM, opex growing slower than revenue. A serial value-creator (Tamer, ex-Inphi→Marvell) is running it. If the AI build-out keeps compounding and Industrial re-accelerates on top, $1.0B exit run-rate is just the start.
Bear case (2–3 ways it permanently impairs / de-rates):
Pre-mortem (18 months out, thesis broke): AI-server FPGA attach decelerated as hyperscalers optimised content or pushed pricing; the Industrial recovery stalled again; AMI integration distracted management and grew slower than the "high-teens" promise while $50–60M of annual interest expense bit; channel partners (94% of revenue) over-ordered into the ramp and then destocked; the multiple compressed from ~75x to ~35x. Net: revenue ~$950M instead of ~$1.15B, non-GAAP EPS ~$2.10 instead of ~$2.55, and a sub-$80 stock — a >45% drawdown with the business still "fine."
Are multiples too high? Yes, on any historical or peer frame. The bull retort is that secular AI-attach + the platform pivot justify a premium. Both can be true; it just means the entry price carries the risk, not the business.
Contrarian view (what the market refuses to see): the market is treating LSCC as a clean secular AI-infrastructure compounder and under-weighting that (a) GAAP earnings are ~nil because SBC is 15–22% of revenue, and (b) it just levered up at the top to buy its way into a new category. The contrarian read isn't that the business is bad — it's excellent — it's that a great business at ~35x sales with fresh leverage and SBC-flattered earnings is a mediocre investment at this price. The asymmetry favours patience.
Dismantling the bull case:
A 3-engine specialty-hardware roll-up wearing an "AI factory" costume — the AI-systems story (Advanced Computing) is the lowest-margin, most lumpy, most hyperscaler-concentrated leg, and the actual FY26 EPS beat is being driven by a cyclical memory (DRAM/Flash) price spike that the bulls are extrapolating as if it were the AI thesis; own the re-rating only if you trust the Shaikh-led non-hyperscaler pivot to convert before the memory cycle rolls.
The pure-play AOI/metrology pick on the HBM-and-chiplet inspection supercycle — >40% HBM-inspection share and 50% of revenue now AI-driven — but a ~50x forward multiple already prices the boom while 49% China revenue sits under a tightening export-control gun.
A best-in-class analog compounder mid-way through a violent cyclical recovery — the business is pristine, the cycle is real, but at ~35x forward / ~65x trailing the tape has already paid for the upturn; the edge is in the next destock, not at today's price.