您的浏览器禁用了JavaScript(一种计算机语言,用以实现您与网页的交互),请解除该禁用,或者联系我们。 [美国银行证券]:半导体IP:聚焦版税业务;买入力旺,将M31下调至跑输大市 - 发现报告

半导体IP:聚焦版税业务;买入力旺,将M31下调至跑输大市

2026-03-31 美国银行证券 周剑
报告封面

Semi IP: Highlighting the royalty business;Buy eMemory, M31down to UnderperformRating Change Prefer eMemory over M31 given steadier growthSemi IP firms generate two kinds of revenue: licensing and royalty. Licensing is one-off and it is recognized when signing the contracts with foundry/chip-design customers,while royalty comes along each piece of wafer shipment and it is based on certainpercentage of wafer price. Thus, royalty revenue could represent a more sustainableincome and CF. In Taiwan semi IP industry, eMemory has stronger earnings profile givenmuch higher mix from royalty, which could translate into steadier growth profile and asustainably higher ROE/OpM. We downgrade M31 to Underperform (from Neutral) as itremains in a transition period before heavy investment (for advanced nodes) pays off. eMemory:returning to a high-growth modeeMemory’s revenue growth was hampered by unfavorable FX and wafer price in 2025, but we expect both factors to become tailwinds in 2026, making 20% revenue YoYgrowth a low bar to achieve along with rising contribution from AI and PUF. Lookingbeyond 2026, we expect the firm’s growth to become more structural, thanks to thegrowing traction to datacenter viaAspeed’s BMC(see report), and CXL/SSD controllersstarting from Vera Rubin platform. Accordingly, we raise 2028E net income by 5%, andarrive at a new PO of NT$3,650 (earlier NT$3,100), still based on DCF. Reiterate Buy. M31: still in the doldrums M31’s opex has increased sharply since 2021 given the investment in advanced node IP,but we note it’s unlikely to pay off until’27 at the earliest, gauging from the cycle of IPdevelopment and lead time from licensing to royalty. We also note that the firm hasbeen more focused on R&D of advanced node IPs (Exhibit 9) over the past few years,which could help it fend off competition from global peers. Yet we think it takes moretime for such a strategy to come to fruition, given a longer lead time from licensing toroyalty in advanced nodes. Thus, we see a limited rise in royalty revenue mix (%) into2027-28, which in turn implies a choppier revenue trend, and milder growth momentumand sluggish OpM recovery in our forecast period. In fact, unsatisfactory execution bynon-TSMC foundries in advanced nodes (where M31 has invested heavily) has been anoverhang to M31’s growth. In the latest earnings call, M31’s CEO mentioned uncertaintyfrom a US foundry’s execution, timing of revenue contribution, and yield rateimprovement, and this makes us turn more conservative on its foundry licensing androyalty revenue outlook. As such, we cut 2026/27E EPS by 52%/26%. Our revised PO ofNT$370 (was NT$450) is now based on 50x (was 53x) P/E, given the softer earningsgrowth outlook, with the roll-over of valuation period to 2H26-1H27E (from 2026E). Weflag downside risk to M31’s valuation—higher P/E but lower ROE when compared withpeers’average. Semiconductor IP: Highlighting the royaltybusiness; M31 down to Underperform Prefer eMemory over M31 given steadier growthSemiconductor IP (intellectual property) companies generate revenue from two different sources: licensing and royalty. The former refers to the one-off revenue recognized whensigning the contracts with foundry/chip-design customers, while the latter comes alongeach piece of wafer production/shipment and it is based on certain percentage of waferprice. In other words, royalty revenue is recurring and could represent a more sustainableincome and cash flow. In Taiwan semiconductor IP industry, we compare the key financial metrics of eMemoryand M31 (Exhibit 2), and we note that eMemory has been recording 1) a steadier growthin revenue, net income, and operating margin; and 2) a sustainably higher ROE. We couldattribute such differences to a much higher revenue mix from royalty business.Elsewhere, we do not compare the key ratios relevant to balance sheet or cash flow,given that both companies operate in asset-light business model and have been profit-making (thus no cash flow or working capital issue). Looking ahead, we prefer eMemory in the IP industry in view of its steadier growthprofile and sustainably higher ROE/OpM (operating margin), while we downgrade M31 toUnderperform (from Neutral), given 1) a bigger uncertainty associated with the nature oflicensing business; and 2) a less satisfactory margin profile. For eMemory, we also flag positives from growing traction to AI (artificial intelligence)and datacenter via BMC (baseboard management controller), DDR5 (double data rate 5)and CXL (compute express link, which in turn will underpin a more structural revenuegrowth. As for M31, we see a much stronger net income growth in 2026/27, which, inour view, can be mostly attributed to an easier comparison base (following a sharpdecline in net income in 2024-25). In terms of valuation for eMemory, the implied 2H26-1H27E P/E multiple by our PO(NT$3,650) is 83x, which is substantially higher vs the multiple (50x) we apply on M31 toderive its valu