Lattice Semiconductor delivered a quarter that beat on every headline line: revenue of $201.1 million against a $185.5 million consensus, non-GAAP EPS of $0.53 against $0.44, and a third-quarter revenue guide of $255 million at the midpoint that sat roughly 32% above what the street had modelled. Revenue grew 62.2% year over year and 17.7% sequentially. Non-GAAP operating margin reached 38.3% against 27.5% a year ago. Free cash flow margin was 40.4%. On any conventional reading this is a clean beat-and-raise from a company executing a genuine inflection.
The number worth sitting with is 69.5%. That is the non-GAAP gross margin Lattice guided to for the third quarter, against 71.7% delivered in the second. The same press release states that the AMI acquisition, closed on 27 July for $1.65 billion, is expected to be accretive to gross margin, and that AMI carries a non-GAAP gross margin expected to exceed 75% by the end of 2026. Those two statements are difficult to reconcile with a 220 basis point sequential decline in the very first quarter that includes AMI revenue.
The arithmetic is straightforward. Lattice guided FPGA revenue of $220 million at the midpoint and total revenue of $255 million, implying roughly $35 million from about two months of AMI. If the FPGA business holds its second-quarter gross margin of 71.7%, it produces about $157.7 million of gross profit. Total gross profit at the guided 69.5% is about $177.2 million. That leaves roughly $19.5 million on $35 million of AMI revenue, a gross margin near 56%. If FPGA margin improves at all with volume, the implied AMI margin falls further below that. Either purchase accounting is taking a deferred revenue and inventory step-up bite that Lattice is not adding back in its non-GAAP presentation, or the acquired business is running considerably further from 75% today than the framing suggests. Neither possibility is fatal, but the first quarter of a $1.65 billion deal is exactly when a management team is most motivated to present the acquired asset favourably, and this presentation goes the other way.
The earnings guide points at the same thing from a different angle. Third-quarter non-GAAP EPS is guided to $0.54 to $0.58, against $0.53 delivered. Yet the implied non-GAAP operating income moves up roughly 18% sequentially on the guided revenue, margin and opex range. Operating income rising by something close to a fifth while earnings per share rises by around 6% is the signature of two things happening at once: the 5.2 million shares issued as the equity portion of the AMI consideration, and the carrying cost of approximately $1 billion of cash consideration funded through the bridge facility whose unamortised costs Lattice wrote off in the quarter. The deal is accretive to revenue immediately and to earnings much more slowly. At $1.65 billion for a business expected to do a little over $200 million of revenue at around 40% adjusted EBITDA margins, Lattice paid roughly 8 times revenue and around 20 times EBITDA, which is not cheap for an asset it now has to defend on neutrality grounds.
That neutrality commitment is the strategic knot at the centre of this transaction, and it is more interesting than the multiple. Lattice’s durable advantage has never really been programmable logic performance. AMD and the reconstituted Altera compete for the high end and have shown little interest in coming back down; Lattice’s small and mid-range low-power position is defended by design-in inertia, a software stack customers have already built against, and the specific accident that platform security and control is a socket nobody wants to re-qualify. The company’s companion chip strategy is an attempt to convert that socket into content per server, and it is working: Compute and Communications revenue reached $126.0 million, up 83% year over year, and contributed roughly three quarters of the entire revenue increase. Industrial and Embedded at $75.1 million grew 36%, which is recovery off a genuinely depressed base rather than a new demand cycle. AMI extends the position upward into the platform firmware layer that runs on essentially every OEM and ODM server, which is the layer above the chip Lattice already sells. Owning both the root of trust silicon and the management firmware is a real moat if hyperscalers permit it. The reason they permit it is precisely that AMI is silicon-agnostic, which means the attach synergy Lattice would most like to harvest is the one it has publicly promised not to pursue. The asset is valuable because it is neutral, and it stops being valuable the moment it stops being neutral.
Two quieter items in the supplemental data deserve attention. Distribution accounted for 95% of revenue against 84% a year ago, which shifts an increasing share of reported revenue to sell-in rather than sell-through, at a moment when management is describing accelerating backlog. Inventory days at 153 are much healthier than the 218 of a year ago, and days sales outstanding fell to 54 from 63, so the working capital picture is not flashing anything. But a channel that has absorbed eleven points more of the mix in twelve months is where a demand air pocket would show up last. Separately, stock-based compensation and related payroll taxes ran $44.9 million in the quarter, about 22% of revenue, which is why GAAP EPS was $0.14 against non-GAAP $0.53. Lattice historically offset that with buybacks, but repurchases fell to $15 million in the first half from $70.9 million a year earlier as cash was conserved for AMI, and it has just issued 5.2 million shares. The share count support that made the non-GAAP number look cleaner is gone for the foreseeable future.
The stock closed at $138 ahead of the print, giving a market capitalisation near $18.9 billion before the newly issued shares, against a 52-week range of $53.06 to $157.01. Adding the acquisition debt, enterprise value sits around $20 billion against a third-quarter annualised revenue run rate of roughly $1.02 billion, or about 20 times sales, and roughly 60 times an annualised non-GAAP earnings run rate of about $2.24. TD Cowen and KeyCorp both carry $165 targets; Stifel, Deutsche Bank and RBC sat in the $140 to $150 band before this quarter and will move up; Bank of America has held the standing Underperform seat throughout the run and has been wrong for four consecutive quarters.
Base case: FPGA revenue reaches roughly $250 million quarterly through 2027 as AI server attach rates and content per box keep rising, AMI contributes a full $55 million or so per quarter, and 2027 non-GAAP EPS lands around $2.80 to $3.00. At 50 times that is $140 to $150, which is approximately where the stock already trades. Bull case: AMI gross margin converges toward the promised 75%, physical AI and robotics design wins convert on the timeline management implies, 2027 EPS reaches $3.40 to $3.60, and the multiple holds near 60 times for $200 to $215. Bear case is a cohort derating rather than a company-specific stumble. LSCC now trades as an AI infrastructure name rather than a programmable logic name, and its beta of 1.81 reflects that. A compression of AI-adjacent multiples from 60 times to 35 times puts the stock at $95 to $105 on unchanged fundamentals, and the gross margin question would give any such move a specific story to attach itself to.
The line to watch is the fourth quarter gross margin guide, which will be the first to include a full quarter of AMI. If it does not recover toward 71%, the accretion thesis on which this $1.65 billion was spent is not what was advertised, and the market will have to decide whether it is paying 20 times sales for a chip company or for a firmware company with a lower ceiling.