Baker 的视角戳破了 AI 投资中的估值幻觉，做 AI 硬件投资或关注产业链的读者，看完会重新审视 NVIDIA 与基础设施公司的相对价值。
Atreides 管理合伙人 Gavin Baker 指出，AI 相关公司估值存在严重不一致：内存制造商 PE 仅 3-5 倍，而 NVIDIA 的 PE 也很低。他认为，市场将 NVIDIA、内存、定制芯片、光网络、电力设备、冷却系统和数据中心建设者都视为同一支出热潮的赢家，但每个板块定价的是不同的繁荣版本。这种横截面低效意味着市场未能一致地对相关 AI 公司进行排名。利润池会流向最难替代、最难延迟、最难被客户压价的部分：如果稀缺的是电力和物理容量，基础设施供应商应获得溢价；如果稀缺的是加速计算，NVIDIA 和内存可能被低估。ASIC 叙事进一步增加了迷雾，Broadcom 在定制芯片上快速增长，但未必能夺取 NVIDIA 最有价值的增长部分。
"If you look at the valuations for all these AI na…
"If you look at the valuations for all these AI names, they just can't all be accurate. You have memory makers at 3-5X PE. You have NVIDIA at a really low PE."
~ @GavinSBaker Managing Partner & CIO of Atreides ($11B AUM).
Nvidia, memory, custom chips, optical networking, power equipment, cooling systems and data center builders are all being treated as winners of the same spending boom.
But each bucket is pricing a different version of the boom.
There is Cross-sectionally inefficiency, i.e. the market is not ranking related AI companies consistently, because one group implies massive data center growth while another group implies slower or riskier growth.
A boom does not automatically make every supplier a compounder, because the profit pool moves toward whatever is hardest to substitute, hardest to delay, and hardest for customers to bargain down.
If the scarce thing is electricity, cooling, and physical capacity, then infrastructure suppliers should command the premium.
If the scarce thing is still accelerated compute, then Nvidia and memory are probably too cheap relative to the rest of the chain.
The ASIC narrative adds another layer of fog.
Broadcom can be growing fast in custom chips and still not be taking the most valuable parts of Nvidia’s growth.
----
From " All-In Podcast" YouTube channel, (link in comment)