AI价值链：Google Gemini 3 Pro、Claude Opus 4.5、Grok 4.1与DeepSeek 3.2……谁是真正的领跑者，它意味着什么？

AI Value Chain: Google Gemini 3 Pro, Claude Opus 4.5, Grok 4.1and DeepSeek 3.2... who is the real leader and what does it mean? With Grok 4.1 on 11/17, Gemini 3 Pro on 11/18, Claude Opus 4.5 on the 24th, and nowDeepSeek v3.2 on 12/1 we've had 4 models come out in the last three weeks that each claimto be the leading frontier model. In this note, we scrutinize these claims and their implications. Stacy A. Rasgon, Ph.D.+1 213 559 5917stacy.rasgon@bernsteinsg.com Daniel Zhu+1 917 344 8309daniel.zhu@bernsteinsg.com We see Gemini 3 Pro and Claude Opus 4.5 as neck and neck, and view DeepSeek’sclaim to leadership with more skepticism.All three model builders have publishedbenchmarks showing themselves as the leader, in accordance with the long historyof model builders cherry-picking favorable-looking benchmarks. While third-partyBenchmarking is still in progress, we believe early results show Gemini and Claude as neckand neck - and Grok 4.1 ahead of GPT-5. Mark Shmulik+1 917 344 8508mark.shmulik@bernsteinsg.com Arpad von Nemes+1 917 344 8461arpad.vonnemes@bernsteinsg.com Zooming out, the bigger and more important takeaway is that the scaling laws areclearly not dead, for both pre-training and post-training. This in turn could renewconfidence for the AI labs and their financial backers to build out AI infrastructureas fast as they are able.That said, the recipe to scale frontier from here likely requires notjust continued access to more and better compute infrastructure, but also tinkering withbest-in-class algos, and potentially also having unique and valuable data to train against.Perhaps the difference in access to data, aside from compute infrastructure, explains whyscaling laws appear clearest at Google, followed by Grok and Meta. Alrick Shaw+1 917 344 8454alrick.shaw@bernsteinsg.com Lavnik Balyan+1 917 344 8563lavnik.balyan@bernsteinsg.com DeepSeek's continued ability to fast follow could renew concerns that if/whenmodel improvement decelerates, fast followers building smaller models couldquickly erode frontier model economics.However, given that model improvement stillappears strong, we believe it is too early to worry about what might happen when modelimprovement decelerates. Wenhuan Chang+1 917 344 8546wenhuan.chang@bernsteinsg.com That said, while industry-wide scaling laws appear healthy, OpenAI appears to befalling behind, amid disappointing GPT-5 performance, three failed pre-trainingruns, a host of high profile departures, and management now sounding a ‘red alarm'.It is hard to be sure on exactly what the issue is, between execution / engineering braindrain, the data gap, or perhaps it is a function of competitors having access to more andbetter compute. Either way, we believe this is a situation worth monitoring. None of the four models were trained on Blackwell and only GPT-5.1 and Grok4.1 were trained on GPUs, which has driven renewed concerns about the threatfrom ASIC/TPUs to Nvidia's moat.However, we believe the defining theme is computescarcity, and that both GPU and ASIC should thrive (we view NVDA’s recent stagnation asparticularly overdone). While there are still a number of highly anticipated model launches on the horizon (withOpenAI and Meta in particular expected to launch new models soon),we believe thatlonger-term, the paradigm may shift away from model benchmarking towardsproduct adoption and monetization- and Gemini being unable to displace ChatGPT atthe top of the app stores despite being a better performing model being a case in point. BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS NVDA (OP, $275 PT):The datacenter opportunity is enormous, and still early, with material upside still possible. AVGO (OP, $400 PT):A strong 2025 AI trajectory seems set to accelerate into 2026, bolstered by software, cash deployment,and superb margins & FCF AMD (MP, $200 PT):AI expectations remain high, but a new deal with OpenAI has the prospect to drive further (possiblysubstantial) growth. GOOGL (MP, $305 PT):Gemini 3’s release and the TPU narrative have capped a strong run for Google. AMZN (OP, $300 PT):We continue to expect acceleration at AWS with large amounts of capacity being brought online, Tranium3 rollout, and a few interesting product-related announcements at Re:Invent focused on simplifying and customizing AI adoptionat the enterprise level. META (OP, $870 PT):A lot is riding on Meta being able to balance earnings growth with AI model frontier progress. Wecontinue to believe Meta is among the best-placed to monetize AI. DETAILS With Grok 4.1 on 11/17, Gemini 3 Pro on 11/18, Claude Opus 4.5 on the 24th, and now DeepSeek v3.2 on 12/1 we've had 4models come out in the last three weeks that each claim to be the leading frontier model. In this note, we scrutinize these claimsand their implications for stocks on our coverage. We see Gemini 3 Pro and Claude Opus 4.5 as neck and neck, and view DeepSeek’s claim to leadership with moreskepticism.All three model builders have p