eMemory Q2 2026 slides: strong growth, AI security push

eMemory’s Q2 2026 results and AI infrastructure security push drove a major stock reaction: shares rose 22.7% to $2,700 after reporting NT$1,096.6M revenue (+17.1% YoY) and NT$7.77 EPS (+45.0% YoY), with operating margin expanding 240 bps to 60.7%. Licensing grew sharply (NT$429.6M, +35.1% YoY) and operating income rose 22.0% YoY (NT$666.1M), while management guided for H2 royalty acceleration. The company highlighted 150+ advanced-node tape-out licenses (70 leading-edge, 88 mainstream) supporting a stronger AI/advanced-node royalty pipeline into 2027.

Analysis

This is less a one-quarter earnings story than an early signal that AI infrastructure is starting to pay a recurring tax for trust, identity, and attestability. That matters because the economic value accrues not to the smallest IP layer, but to the platforms that can standardize and certify it at scale; in practice, that favors TSM more than the niche IP vendor itself. If chiplet and CXL security becomes table stakes, the foundry with the deepest advanced-node footprint captures the stickiest volume while competitors without comparable ecosystem control see lower design-win conversion and weaker pricing power.

The second-order effect is that security IP should increase switching costs and raise the bar for qualification on 3nm/2nm and advanced packaging. Over the next 1-3 months, the key question is whether this is a narrative premium or the start of a real royalty ramp; the stock reaction already prices in some 2027 upside, so near-term upside likely depends on proof that tape-outs are turning into wafer starts. Over 6-18 months, a successful standardization of root-of-trust across AI servers and chiplets would reinforce TSM’s advanced-node mix and make the ecosystem more closed, which is bullish for pricing discipline.

Contrarianly, the market may be overestimating how quickly security becomes mandatory versus optional. If OEMs or foundries internalize the trust anchor, or if open standards commoditize the block, the royalty pool could be smaller than bulls expect and the current multiple expansion would be vulnerable. Falsifiers are straightforward: no step-up in royalty growth in 2H26, no visible conversion from advanced-node licenses into production, or any evidence that AI infrastructure customers are deferring security content to save BOM cost.

Key Decisions for Investors

• Long TSM on any 2-4 week pullback; thesis is relative outperformance from higher advanced-node content and tighter ecosystem control. Risk/reward: modest downside if AI capex pauses, but 8-12% upside if the market starts pricing security content into 2027 wafer mix.
• Pair trade: long TSM / short SOXX for 1-3 months to isolate the foundry-specific moat expansion versus broad semis beta. Falsifier: broad semiconductor multiple expansion on macro easing would swamp the spread.
• Set an alert on TSM commentary for 2nm/3nm qualification and advanced packaging attach rates; if security-related silicon content is explicitly embedded in next-gen ramps, add to longs. If not, fade the move as narrative premium.
• Avoid chasing pure-play security IP here; the better risk-adjusted expression is the manufacturing tollbooth, not the small licensing name. Use the current reaction as a watchlist event unless royalty conversion is independently confirmed.