The video argues the next AI trade is not NVDA, but the CPU names that benefit when AI shifts from chatbots to agents and robotics, with the biggest setup centered on AMD. Chatbots and LLMs are framed as early, expensive, ask-and-answer tools; agents and robotics are framed as the second tidal wave, still barely adopted, but set to drive far more CPU demand as memory and context requirements rise. The case says the market is late, AI deployment is already profitable, and the real opportunity will come when falling costs trigger a Jevons-paradox-style demand surge. AMD is presented as the cleanest way to play that setup: cheap relative to its history, improving fast, and positioned to take rack-level share and some of NVDA's pie.
The pitch is for a once-every-few-decades investing opportunity that most investors miss, with prior calls cited as PLTR in 2020, NVDA in 2020 for 1,500%, FTNT for 400%, ANET for 400%, and MU at the start of this year for 200%.
The current AI cycle is framed as real and disruptive, not a bubble, despite complaints that capex is too high.
The next wave is shifted away from chatbots and LLMs to agents and robotics: chatbots/LLMs are ask-and-answer systems, while agents/robots are autonomous systems that complete tasks on their own.
Agents and robotics are described as still at super-early-adopter stage, with about 10% of the market using new tech early and roughly 90% not yet using agents, much less robotics.
The main blockers to adoption are complexity and cost: training workers on agents is too expensive, tokens are too expensive, and an office worker in Excel is not economic at current pricing.
Agentic AI is said to require much more context and memory, pushing CPU demand from roughly 1 CPU per 12 GPUs in the classic LLM setup toward 1-2 CPUs per GPU and possibly 1:1.
When agents become a household item, CPU demand is expected to enter a new realm that is not yet priced in.
The inflection point is defined as the moment when use becomes $1 cheaper than the benefit it provides; after that, demand explodes and the market usually arrives only after the move is already priced in.
Jevons paradox is used as the mechanism: lower cost and higher efficiency can increase total demand, not reduce it; the 1865 steam-engine example is cited, where cheaper engines led to more coal use instead of less.
OpenAI is cited as having dropped prices for developers a few days ago, pushing AI costs lower and moving the inflection point closer.
AI deployment is claimed to already be profitable: on $100 billion of data-center capex, the numbers given are about $37 billion of revenue over 5 years, minus about $5 billion a year of operating costs, leaving about $60 billion over 5 years, or roughly $12 billion a year.
The next question is framed as who benefits from the hyper cycle, but infrastructure software is explicitly set aside for a later video.
In CPUs, the names to watch are ARM, INTC, and AMD; AMD is down 21% from June, ARM down 40%, and INTC down almost 40%, even though all three remain sharply up over the past year after a spike and correction.
AMD is preferred over INTC and ARM because it has the best story, fundamentals, setup, and the most misunderstanding, with a CEO described as one of the best in the business.
AMD is presented as cheap at under 34x P/E and 17x sales, the cheapest since 2022 and the cheapest in 4 years; fundamentals cited include about 50% annual growth now, 1,600% operating-income growth, 160% net-income growth, cash up 123%, and debt tripled.
Over 4 years, AMD operating income is said to have risen from $1.3 billion to $3.7 billion, free cash flow from $3.1 billion to $6.8 billion, and cash from $5.9 billion to $10.5 billion.
Stock MVP is pitched as a tool to analyze any stock in about 30 seconds, show whether a company is real, and provide a 5-year forecaster; its AMD model is given as 158% upside in the bearish case, 534% in the medium case, and 1,500% in the bull case over 5 years.
AMD is compared to the scale of upside seen in PLTR and NVDA, while retail is said to be late again: excited about PLTR at 200, rejecting it at 120, then wanting it at 180 a month later; similarly, everybody wanted AMD at 600 and now nobody wants it.
The buy rule is to add when price falls and business quality rises, and the suggested implementation is a double-down DCA plan: add a little, wait for more pullbacks, double down, and build over the next 12 months.
AMD is said to be building networking effects, selling its whole rack in one piece, with Venice expected to be sold alongside the AMD rack, letting AMD monetize not only CPUs but also take some of NVDA's pie because NVDA was never meant to supply 100% of the chain.