The case is for owning AI infrastructure, not chasing app-layer hype: NVDA, AVGO, AMD, ARM, ASML, MU, ANET, CDNS, LRCX, TSM, AMZN, MSFT, GOOGL, VRT, GEV, CIEN, BE, CEG, DLR, PLTR, MDB, CRWD, and TSLA are framed as the 24-stock setup for the next AI super cycle. The core claim is that agents, not chatbots, drive a much larger 24/7 demand shock across compute, memory, networking, power, cooling, orchestration, and robotics, with a potential 500x demand expansion if AI spreads from developers into Excel, Word, PowerPoint, Outlook, and general knowledge work. Multiples are treated as secondary to growth and moat: NVDA and PLTR are cited as examples where expensive stocks compounded through revenue acceleration, while current leaders show strong revenue, income, and cash-flow growth. The closing portfolio advice is defensive and process-heavy: keep a 6-month emergency fund, hold at least half in the S&P, use DCA/DCA X2 and trimming rules, and avoid selling on up moves, down moves, or macro fear.
NVDA is the template: it was called expensive at 20x sales and 22x sales, yet revenue rose from 61 to 131 to 216 and the stock is up 5x from late 2023 to now.
PLTR was called expensive at 17x sales in 2023, then revenue doubled and the stock rose 20x; it now trades at 58x sales.
Roic Academy’s October 2020 top-stocks list included NVDA and PLTR and is up 1,300% and 1,200% since, and it also included ANET.
Agentic AI is treated as a much bigger shift than chatbots because agents run 24/7, build context layer upon layer, and can require 10x more tokens than chatbots.
If agents move beyond developers into Excel, Word, PowerPoint, Outlook, and informational work, the TAM expands about 50x and token usage about 10x, implying 500x more AI demand than today.
Global AI capacity is already on track to be 24x by 2030, and cheaper AI can trigger a Jevons-paradox flywheel that lowers cost and raises demand again.
NVDA is the compute-layer engine of the stack: every developer, lab, and data center is built on it, moving away takes years and billions, and there are at least five or six capex cycles ahead before growth slows.
AVGO is the completion to NVDA: custom AI chips for data centers plus the network connecting GPUs and infrastructure; its AI semiconductor revenue is up 143% y/y, operating income 89% y/y, and net income 292% y/y.
AMD has a double thesis in GPUs and CPUs as second-sourcing plus renewed CPU demand from AI context and memory; net income is up 164% y/y, operating income 94% y/y, and free cash flow 180% y/y.
ARM is a toll-booth on the AI industry because every chip maker pays for its blueprint; it was blocked from being bought by NVDA, and genetic AI adds more royalty income.
ASML is described as a literal monopoly in lithography, with no alternative and an order book that keeps it at full capacity for the next 5 years.
MU has turned from a commodity business into a high-bandwidth-memory moat; building a competing facility would take 5 years and billions, the stock still trades at 5x forward earnings, operating income is up 650% in the past year, and free cash flow is up 1,300%.
ANET is the GPU-networking gear inside data centers and becomes more important with agents; it is up about 340% since being added to the top-stocks list, with revenue growth at 30% and income growth at 23%.
CDNS is the EDA layer that simulates chip production before billions are spent on tooling and manufacturing; agents create more chip demand ahead, and CDNS shows 42% free cash flow growth, 22% operating income growth, and a 31% operating margin.
LRCX owns 50% of chip-etching equipment, integrates with ASML’s patterning tools, and is set up as a major winner of the agentic AI phase; it has 24% revenue growth, 40% operating income growth, and 40% net income growth y/y.
TSM is sold out into infinity, is called unavoidable for agentic AI, and trades below 20x forward P/E with 48% operating income growth.
AMZN is framed as AWS being the toll booth of AI: 31% net income growth, 3.6x sales, and 25x forward P/E.
MSFT owns the cloud, business software, data, and tools for AI agents inside organizations; it trades at 20x earnings, 9x sales, has a 47% operating margin, and its stock is down recently.
GOOGL’s TPUs reduce GPU dependence and improve margins; it is the fastest-growing cloud at 63% annual growth, with 32% net income growth and 23x forward earnings, versus MSFT cloud growth at 40% and AMZN at 28%.
VRT is tied to AI cooling and thermal management as density rises; it has 170% net income growth, 67% free cash flow growth, and 38% operating income growth over the past year.
GEV is the grid-upgrade play for data centers that cannot expand without major power infrastructure; it has 76% operating income growth, 214% net income growth, and 62% operating profit margin growth.
CIEN is the optical network between data centers, and agent traffic multiplies both inside facilities and between them; it has 47% income growth, 63% operating income growth, and 76% free cash flow growth.
BE is positioned as onsite power when the grid is strained, with almost 40% revenue growth y/y, 217% operating income growth y/y, and 72% free cash flow growth y/y.
CEG is slower growth but is included because nuclear is needed for 24/7 reliable power for AI and data centers.
DLR is the landlord for data-center real estate; it trades at 3x sales, has 125% free cash flow growth y/y, 117% net income growth, 520% free cash flow growth, and a 61% EBITDA margin.
PLTR is the orchestration play: agents need software that governs, watches, monitors, and secures them, and the claim that chatbots will displace it is dismissed as ridiculous; its metrics are 85% revenue growth, 145 rule of 40, and 360% operating income growth y/y.
MDB is the context/database bottleneck in orchestration, with 314% free cash flow growth last year and 23% revenue growth.
CRWD is the cybersecurity layer for agents that will access information, code, and other sensitive assets; it has 120% net dollar retention, 22% revenue growth, and 23% free cash flow growth.
2030 to 2035 is called the robotics decade, embodied AI is projected to become the largest AI-economy layer, and US working hours already could be replaced by technology equal to 13% of current labor time.
TSLA is already leading robotics because FSD is treated as humanoid robotics on wheels; its ingredients include electric motors, vision and sensors, batteries, supply chain, manufacturing, data, and 10 million real miles.
TSLA’s camera-only choice is tied to building robotics early; humanoid robots cannot be manufactured with lidar on the head, so cameras were the right choice.
TSLA still has solid fundamentals while pivoting: 74% free cash flow growth over the past year, $44 billion in cash, and $95 billion of annual revenue from car sales alone.
The investing framework is infrastructure over applications, moats over hype, and orchestration over shiny wrappers, with five layers named as compute, cloud and power, orchestration, embodied AI, and consumption.
An emergency fund for 6 months is required before investing, at least half of a portfolio should be in the S&P, and the DCA X2 system deploys 50%, waits with 50%, then doubles down as the stock comes down while always buying something.
The trim rules are to trim at 50%, 100%, 150%, and 200% gains and recycle into DCA; the standing rules are never sell because a stock went up, down, or because of macro.