The video’s core claim: GPT6 Astra marks an AGI break, and the spend required to chase it pushes money down the whole AI stack — NVDA, AMD, TSMC, memory, GPU-clouds, power, cooling, and select software. Astra trained on a little over 100,000 of NVDA’s newest chips; the next wave is framed as 400,000 GPUs, with Microsoft pricing Astra at about 2.5x the prior generation and the top five hyperscalers heading toward $1.3 trillion of capex next year. The final segment is a sponsored pitch on VNRX, a small epigenetics diagnostics name with sepsis and cancer tests, dated catalysts, and high execution risk.
GPT6 Astra is treated as AGI, trained on a little over 100,000 of NVDA’s newest chips; the next wave is 400,000 GPUs.
Microsoft made Astra available inside its cloud the same day it launched, priced at roughly 2.5x the previous generation.
NVDA said on its earnings call that each gigawatt of Grace Blackwell data center build is about $25 billion of revenue opportunity, or about $40 billion per gigawatt with Ver Rubin.
The top five hyperscalers are on track to spend close to $800 billion on capex this year, heading toward $1.3 trillion next year.
NVDA stock is described as rising from 108 at the end of 2022 to just under 858 today; the company is near $5.5 trillion in market cap.
AMD is framed as having the bigger percentage upside from today because compute demand is going parabolic, with the setup described as a “ridiculous, ridiculous 2025.”
TSMC manufactures the chips NVDA and AMD design, and every one of the 400,000 GPUs and competing AMD chips runs through its fabs and advanced packaging lines.
Micron’s HBM and other memory benefits from larger models and longer-running agents, and Asian memory peers like SK Hynix and Samsung gapped up on the Astra launch while SanDisk ripped.
CRWV has roughly $22 billion in contracts with OpenAI specifically, rents GPU capacity by the hour, and jumped 12% on launch day.
NBIS is pitched as the European version of the GPU-cloud model, while APLD is converting Bitcoin-mining power and land into AI data centers with faster deployment than a greenfield build.
CEG is said to run 22 reactors and to be the clearest nuclear-power fit for AGI-scale data centers, including its deal with Microsoft to restart Three Mile Island.
VST is described as owning nuclear and gas plants across Texas and the Midwest, with a deal with Meta for about 2.6 gigawatts of nuclear power.
VRT supplies cooling systems, power distribution, and racks, which are necessary because Grace Blackwell racks run on liquid cooling and every one of those 400,000 chips needs it.
NVTS is positioned as an 800-volt power-architecture beneficiary: its newest board takes 800 volts straight down to 6 volts in one step, and the stock ripped almost 20% the day it announced it was inside Nvidia’s MGX ecosystem.
PLTR is framed as the software layer that lets governments and big companies point AI at private data and operations; a stronger model makes its platform more valuable, not less.
MSFT is said to gain from Astra adoption because companies still need Azure identity, logging, security, and private networking even when agents do the work.
LLY is portrayed as going hardest into AI among drugmakers, with a DGX super pod of over 1,000 Blackwell GPUs, a January deal with NVDA for up to $1 billion over 5 years, and a March $2.75 billion deal with In Silicon Medicine.
TM is said to control one of the world’s largest patient-data libraries and to benefit because every drug company that wants an AI agent to find new targets needs that data.
RXRX is framed as a pure-play AI-first drug company with automated lab experiments feeding its own models, multiple candidates in human trials, and backing from NVDA.
VNRX is the sponsored close: a multinational epigenetics diagnostics company whose NU.QNETS, NU.QCANCER/capture SEQ, NU.Q VET, and NU.Q DISCOVER pillars target sepsis, cancer, companion animals, and research tools.
VNRX says NU.QNETS addresses a roughly $3.8 billion annualized market, capture SEQ detected over 95% of stage 1 and stage 2 cancers at 95% specificity in a blinded cohort, and the company has a $5 million milestone tied to the feline lymphoma manuscript publication.