tae kim@firstadopter
Here's what's going to happen from here:
-Nvidia will report blowout numbers the next several quarters on the NVL72 product supercycle (a step function up in capability with 72 GPUs in one AI server versus 8 GPUs). It will become one of the largest cycles in technology history, akin to the iPhone versus Blackberry. The clear signs are there in the latest quarter with Nvidia posting its first accelerating revenue growth in two years and triple-digit networking segment growth.
-Google TPUs will be less than 10% of the market for the next few years as the major hyperscaler buyers don't want to support a cloud rival (outside of Meta), while Nvidia's software CUDA/developer ecosystem, TSMC allocation and performance advantages remain strong.
-Most search queries will transition to AI chatbots over the next few years.
-ChatGPT will release a much better model trained on NVL72 Blackwell clusters at Microsoft. Sentiment will shift back to ChatGPT.
-ChatGPT will add digital advertising in its consumer product. The first iteration will not be great. The second version will improve. The third version will work great.
-A significant portion of the digital ad market will move to AI chatbots and AI consumer hardware, away from search ads.
-Google will go from 95% search monopolist to a number 2 or number 3 player in the AI chatbot market, which will dramatically lower its margins over the cycle. Serving the search index was a gold mine. That era will end. Timing is difficult. It may take a while, but it will happen.
-All the talk about AGI and superintelligence is a distraction and a side show. AI adoption and AI progress will accelerate through 2026 as Gemini and Claude Opus proved scaling laws are intact.
Enterprises will unleash massive productivity gains using current technology. Cursor will be the precursor (get it?) of the future. It eliminates tedious work with autocomplete, bug fixing, leading to rapid iteration of new ideas for coding. It enables 40% more productivity.
There will be a Cursor for every vertical. Knowledge workers will become vastly more productive as AI models build upon intuitive understanding of what helps them with proprietary custom data and models.
-But don't they lose money now? Compute performance continues to improve and costs will come down. This is inevitable based on history. Today's loss-making features will become enormously profitable in due time.
Curious to hear your thoughts.
