Northland lifts Semtech rating, sees connectivity as the AI bottleneck

Semtech was upgraded by Northland Capital Markets to Outperform with a $182 price target, citing AI connectivity as a current bottleneck benefiting its copper/optical products. In Q2, Semtech posted non-GAAP EPS of $0.71 on $342M revenue, beating consensus ($0.61, $328M), with gross margin at 54.5% (59.7% ex-business held for sale, up 520 bps sequentially). Data center revenue hit a record $100M (+91% YoY), and Northland expects data center mix to rise from 30% to 44% by end-2027, supporting mid-60% gross margins and FiberEdge gaining to >50% share by FY2027.

Analysis

This is less a one-quarter beat than evidence that AI capex is migrating from pure compute to the plumbing that makes compute usable. That is a favorable setup for SMTC because interconnect content can grow faster than rack count, and once a design win is embedded, revenue tends to lag but margins can re-rate quickly as mix shifts up. The secondary beneficiaries are the broader networking supply chain and module vendors; the hidden loser is any supplier whose value proposition is still tied to commodity-speed connectivity or legacy server architectures that assume bandwidth is not the constraint.

For NVDA, the read-through is mixed but still constructive: demand is strong, yet the growth bottleneck is increasingly outside the GPU. In the next 1-3 months, that can create a rotation where investors pay up for the picks-and-shovels beneath the accelerator layer, while the market questions whether GPU shipment growth is being throttled by network readiness rather than end-demand. Over 6-18 months, if SMTC really converts design wins into share, the more important implication is that AI system BOMs get re-priced higher, which should expand the addressable spend per cluster and support the entire AI infrastructure stack.

The main risk is that the margin improvement story is being over-attributed to operating momentum when part of it may be mix/accounting driven and therefore less durable. A second risk is vertical integration: hyperscalers and module OEMs can dual-source or internalize interconnect faster than the bull case assumes, capping pricing power. This thesis breaks if next few quarters show revenue growth without sustained gross margin follow-through, or if 1.6T adoption slips and the design-win narrative turns into delayed revenue recognition rather than share gains.

Key Decisions for Investors

• Go long SMTC on post-upgrade weakness for a 4-8 week window; target a 10-15% rerating if the market starts pricing it as an AI interconnect compounder rather than a cyclical components name. Falsify on any guide that fails to sustain mid-50s gross margin ex-divestiture effects.
• Use a long SMTC / short SMH pair for 1-3 months to isolate relative upside from AI networking content growth versus broad semis beta. The trade works if investors rotate from GPU-beta to bottleneck beneficiaries; cover if SMTC underperforms on the next earnings print or guidance implies design-win revenue slippage.
• Treat NVDA as a conditional hold, not the immediate trade, unless the next data point shows networking constraints are easing faster than expected. If GPU shipment lead times shorten while interconnect ramps lag, SMTC should continue to outperform; if not, the relative-value case weakens.
• Set a watch item on the next two SMTC quarters: if data-center mix keeps rising but cash conversion or gross margin stalls, fade the rally. That would imply the market is paying for a structural margin story that is really just a temporary mix shift.