NVDA's scale-up is the core trade: a record quarter near $100B in 3 months, 100% y/y growth, a $13B Hugging Face buy, GPU prices up 15% on memory costs, and banks lifting targets as high as $420 all point to demand still outrunning supply. The bull case extends into open source, inference, and robotics: NVIDIA is backing open models, funding data-center buildouts, and positioning CUDA/cloud plumbing as the moat while Chinese silicon, Huawei, and Z.AI's GLM 5.3 Flash show inference and robotics getting cheaper and more distributed. The rest of the video ties that to Tesla, Whimo, Anthropic/OpenAI usage shifts, Apple hardware timing, and a separate SEC probe around a $400M private bet tied to Leopold Ashenback. The through-line is that compute demand, not just model quality, keeps expanding — and the biggest near-term swing factors are open-source adoption, China hardware progress, and whether Nvidia's pricing power survives higher memory costs.
NVDA bought Hugging Face for $13B, a deal framed as paying 50x Hugging Face's earnings to own the open-source platform.
NVDA posted a record quarter with almost $100B earned in 3 months and 100% y/y growth, despite already being the world's most valuable company at over $5T.
Every bank raised NVDA price targets after earnings; the highest target mentioned was $420/share, and the stock was up 7% early after being down in after-hours trading.
NVDA said 2028 revenue growth would be 70% y/y versus 44% expected, while quarterly revenue was described as exceeding the annual revenue of roughly 480 S&P 500 companies.
NVDA was described as printing just over $1B of cash per day, with the episode day called a day NVDA was printing $1B of revenue.
NVDA raised GPU prices 15% because of soaring memory costs, with the higher chip prices said to be filtering through the whole hardware stack.
Jensen Huang contrasted OpenAI's "jalapeno" chip with NVDA, saying NVDA are "the kings" after 33 years, while OpenAI has been at it for 3 months and should be judged after shipping a chip, putting it in data centers, and scaling it.
NVDA was said to be sold out for as much as it can make, operating at full capacity, and working with five of the world's largest banks to help fund more data-center buildout using NVDA GPUs.
Three reasons were given for NVDA's open-source push: hedge against customers building alternatives, cheaper open weights driving more token volume as frontier margins compress, and moving the moat up the stack via CUDA optimization plus one-click deployment onto NVDA cloud.
Hugging Face was described as hosting open-source AI models and datasets; the notes put about 155,000 GLM models on the platform, with some fine-tuned for trading, research, and niche uses.
A "mystery model" that gave 100red trillion tokens of availability for free for a week was identified the next day as Z.AI's GLM 5.3 Flash, run on Chinese silicon and priced at 15 cents per million tokens.
The Z.AI model was said to understand video, audio, and text, and the benchmark readout was called "opus tier agentic" but not frontier.
U.S. export bans over the last 4 months were said to have left NVDA's quarterly earnings untouched by China sales, while Chinese teams built around the ban by using distillation and their own chips for inference.
Chinese chip companies were said to have gone from 2.5 years behind to running the top stack of one of the top open-source models on Chinese silicon; Huawei was named as the chipmaker, with a cluster announced about a month ago and already at full-scale production.
Whimo's new vehicle, Ohigh, was said to use a new chip architecture that is 20x more powerful than over the last 8 years and 5x cheaper, cutting the sensor-suite cost to about $25K per car from roughly $100K-$125K.
Tesla's Cyber Cab was said to launch at scale on September 3 in Austin, Dallas, Houston, Miami, Orlando, and Tampa, with a $25,000 all-in vehicle price and about eight smartphone-like cameras; FSD was said to work anywhere, unlike Whimo's geofenced setup.
Leopold Ashenback was said to be under SEC investigation, with subpoenas sent to banks tied to his "situational awareness funds" over alleged insider trading, and Ajet Balwit identified as his wife and Anthropic chief of staff.
Financial Times data showed Anthropic model usage shifting: Fable growth accelerated from mid-June to late July then flattened, while OpenAI usage rose, especially codec usage; the read was that cheaper tokens and model preference are driving a larger total pie.
Enterprises were said to want frontier models but also use zero data retention as a compliance constraint; OpenAI was said not to be enforcing ZDR now, possibly explaining faster recent acceleration, while Anthropic may be moving toward it next.
Robotic learning was said to be accelerating via in-context learning, where a robot learns from a short video such as making pancakes; examples included a robot correcting mistakes, handling a dustpan swapped for a banana, and getting paid $15/hour to collect household training data.
Apple's hardware event was dated September 19, with Siri expected to follow shortly after the hardware launch, alongside a new iPhone 18 Pro and a foldable device; the first event under the new CEO was said to have John Turnis running it and Tim Cook offstage.
Sam Alman was said to have ordered seven ultra-rare custom Swiss watches from Van, with one intended for his wrist and a secondary value estimated at $500K to $1M; the watch rumor was framed as a possible peace offering ahead of a rumored IPO and as evidence OpenAI expects AGI internally by year-end.