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。[伯恩斯坦]:高通(QCOM):罗德尼·丹杰菲尔德迎来人工智能转型? - 发现报告

高通(QCOM):罗德尼·丹杰菲尔德迎来人工智能转型?

2025-10-27伯恩斯坦阿***
AI智能总结
查看更多
高通(QCOM):罗德尼·丹杰菲尔德迎来人工智能转型?

Qualcomm now an AI play?Earlier today Qualcomm announced new chip and rack-scaleAI datacenter products for inference (the AI200 and AI250), as well as a a partnership withHUMAIN in Saudi Arabia to deploy 200 MW of these racks starting in 2026. Stacy A. Rasgon, Ph.D.+1 213 559 5917stacy.rasgon@bernsteinsg.com Sometimes it takes a few repetitions...Qualcomm has actually sold AI acceleratorsfor years, but many investors seem unaware. The company’s first-generation parts (theAI100) were actually announced way back in 2019, with commercial shipments in 2021,and even announced an updated version (the Cloud AI 100 Ultra) for generative Ai at theend of 2023). These new parts appear to be next-gen, and are the first to bring a rack-scalearchitecture to the company’s offering. For whatever reason the company’s earlier effortsdrew little to no attention but today’s announcement appears to be garnering more, andthe stock is (as of this writing) up ~15% (and was up almost 20% at one point), significantlyoutperforming even on a strong market day. Alrick Shaw+1 917 344 8454alrick.shaw@bernsteinsg.com Arpad von Nemes+1 917 344 8461arpad.vonnemes@bernsteinsg.com Lower cost, lower power?It is hard to concretely outline a value proposition for a newproduct solely from a press release, but the company suggests low TCO given 768 GBof LPDDR (no HBM?) for higher memory capacity and lower cost, as well as a (supposed)“innovative memory architecture based on near-memory computing” delivering “greater than10x higher effective memory bandwidth and much lower power consumption.” Racks willbe liquid cooled, use PCIe for scale-up, and Ethernet for scale out, with 160 kW rack-levelpower consumption. We don’t know how many AI200 cards (or QCOM CPUs?) might be in arack though. How big an opportunity?Again, hard to say. However, the 200 MW HUMAIN installationwould (at 160 kW/rack) translate to ~1,250 racks. We don’t know how much a rack wouldcost though; at ~$1.5M per (about half that of an NVDA rack?) that would be close to ~$2Bto HUMAIN alone? And perhaps there will be others at some point. We note that 200MWis dwarfed by many of the other deals we have heard tossed around by OpenAI and others,but for now we’ll take it (and we feel sure Cristiano would take a call from Altman if theopportunity were to arise!). We have liked Qualcomm as an out of favor play(in fact in June we pitched it as theRodney Dangerfield of semiconductor stocks as a best idea). We have understood all thecomplaints (AAPL rolling off, numbers possibly too high as a result, smartphone exposureetc). That being said, we have viewed the AAPL situation as entirely understood, haveviewed the diversification story as very real, and have believed the name carries with ita number of potentially lucrative options. We note company targets calling for ~$50B in2029 revenues (with no AAPL) and ~$14-$15 in EPS, targets that did not even includeany datacenter opportunities (of which we now have both AI server CPUs, and AI serveraccelerators). We still don’t know how much traction Qualcomm’s AI products will garner,but they are zero in our model, and (at least until yesterday) offered zero contribution to themultiple. Plenty of other names are benefiting from the “small slice of a big pie” thesis;perhaps it is not unreasonable to start giving Qualcomm a little bit of benefit from the same?We think so. We rate Qualcomm Outpeform. BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS QCOM (OP, PT $185):AAPL headwinds are now here but are known, the product portfolio is stronger than ever, option valueexists, and the shares are very inexpensive NVDA (OP, $225):The datacenter opportunity is enormous, and still early, with material upside still possible. AMD (MP, $200):AI expectations remain high, but a new deal with OpenAI has the prospect to drive further (possiblysubstantial) growth AVGO (OP, $400):A strong 2025 AI trajectory seems set to accelerate into 2026, bolstered by software, cash deployment, andsuperb margins & FCF. INTC (MP, $35):Intel's problems have broken through to the forefront. I. REQUIRED DISCLOSURES References to "Bernstein" or the “Firm” in these disclosures relate to the following entities: Bernstein Institutional Services LLC(April 1, 2024 onwards), Sanford C. Bernstein & Co., LLC (pre April 1, 2024), Bernstein Autonomous LLP, BSG France S.A. (April 1,2024 onwards), Sanford C. Bernstein (Hong Kong) Limited盛博香港有限公司,Sanford C. Bernstein (Canada) Limited, SanfordC. Bernstein (India) Private Limited (SEBI registration no. INH000006378), Sanford C. Bernstein (Singapore) Private Limited andSanford C. Bernstein Japan KK(サンフォード・C・バーンスタイン株式会社). On April 1, 2024, Société Générale (SG) and AllianceBernstein, L.P. (AB) completed a transaction that created a new joint venturein which their respective cash equities and research businesses operate in a new business combination. Although their respectiveownership percentages in the joint venture differ between North Ameri