野村上调创意电子评级，称Google Axion CPU 2027年机会被低估

Global Unichip Corp: Upside well underappreciated; upgrade to Buy
Google CPU alone could be USD3.8bn business opportunity in 2027, nearly as big as 2027 consensus estimate for the whole company, not to mention other big CSP projects
Upgrade to Buy from Neutral with a higher TP of TWD5,800, implying 52% upside
We recommended a trading buy call on GUC from Nov 2025 given our view that the change of business model (from GPM sensitive to maximizing business opportunities in project selection) would drive significant sales upside (but meaningful GPM downtrend), particularly when GUC got the Google CPU and US EV maker’s AI business opportunities. With the share price correcting by 34% from its May 2026 peak, vs. TAIEX up 3% (following a 263% rally from Nov-2025), we notice that overall business visibility has improved, while the valuation has turned favorable again, in our view – both of which are important factors behind our upgrade to Buy. Our view is simple: the Google CPUs “revenue opportunities” for GUC in 2027 is significantly underappreciated by consensus, not to mention that GUC also has the US EV maker’s AI business opportunities (the business upside from which, as per Elon Musk’s many comments, could be significant) from 2H26. At the company’s 31 July earnings meeting, the new Chairlady (first time attending the earnings meeting) did not deny the company’s opportunity in Google’s TPU (which from our industry survey might be TPU v.10, though that won't necessarily conflict with MTK’s [2454 TT, Buy] TPU ASIC upside, in our view). Thus, we upgrade GUC to Buy with a new TP of TWD5,800, on 40x 2027F EPS (vs. 50x earlier). We believe 40x P/E is not demanding, at GUC’s mid-cycle valuation (Fig. 7 currently trading at 26.1x). Our 2026-28F EPS forecasts are 12%/23%/47% above Bloomberg consensus estimates.
A deep-dive into Google CPU’s potential contribution to GUC in 2027F
Google has indicated that it would double the number of physical CPU hosts per server, moving to its custom Axion Arm-based CPUs in late-April 2026 (press , report ). Its CPU vs TPU ratio has doubled to 1:2 from 1:4 from TPU v.8 onward, and it has started enabling its own CPU (Axion 2) to pair with TPU v8t and v8i. We estimate 9.6m units of v8 series TPU (see our June 2026 AI Semi Anchor ), translating into 4.8mn units of Axion 2 demand in 2026-27F. On our estimate, GUC could ship roughly 1mn units in 2026 (translating into USD1bn in sales, or nearly half of its 2026 sales), suggesting another 3.8mn to ship in 2027 — which would mean that Google CPU contribution to GUC could approach USD3.8bn in 2027 (or higher, if the Axion 3 volume ramp-up starts kicking in from 2H27 with ASP much higher than for the prior-gen given our view that Axion 3 is a two-die package vs Axion 2’s one-die, report ). Further upside should come from the adoption by Google’s general servers, but a downside risk to this simulation is TSMC’s (2330 TT, Buy) capacity sufficiency. To compare, Bloomberg consensus for GUC’s total 2027 sales is ”only” double y-y to USD3.5-4bn. Fig. 2 briefs our Google CPU impact simulation.