Short video / excerpt
100 Data Sources for NVIDIA GPUs. One for TPUs.
Will there be a compute market for non-NVIDIA chips? Carmen Li explains how Silicon Data builds its indices: more than 100 data sources for NVIDIA GPUs, 40 in the calculation, each with less than 3.5% influence. For AMD she has five, and for TPUs one, which is why an index is not automatically tradable. From our conversation at NVIDIA GTC Berlin.
Short video / excerpt
I Went on CNBC to Defend Jensen Huang
Why I signed the Stand with Jensen petition, on CNBC’s Squawk Box on NVIDIA earnings day. Jensen is the lifeblood of NVIDIA, and the Board owes shareholders a rational decision.
Short video / excerpt
NVIDIA Earnings Day: Why Jensen Huang Is the Lifeblood of NVIDIA | CNBC Squawk Box
NVIDIA reported earnings today, and the biggest question over the stock is leadership. On CNBC's Squawk Box this morning, Ben Pouladian explained why he signed the Stand with Jensen petition: Jensen Huang is the lifeblood of NVIDIA, and the Board owes shareholders a rational decision. Jensen has since said he would be honored to return as CEO if the Board decides it is in the best interest of shareholders.
Short video / excerpt
Anything in Life Can Be Turned Into Code
On Fox Business, Ben Pouladian explains why Anthropic spreads its silicon risk across suppliers, why he sees NVIDIA as the best compute for training larger models, and why the opportunity is wider than coding: any business process can be turned into code.
Short video / excerpt
Why Are H100 Rental Prices Still Going Up?
An older-generation GPU that keeps getting more expensive to rent? Carmen Li of Silicon Data compares it to secondhand cars during COVID, and shares what Jensen Huang told her when she asked about residual value: when the newest chips are sold out, you take the next chip that fits your use case. From our conversation at NVIDIA GTC Berlin.
Short video / excerpt
Why Compute Could Become the World's Largest Commodity
Carmen Li started as an energy options market maker, where she saw why an airline hedges its fuel. She explains how AI startups complaining about on-demand GPU prices led her to leave Bloomberg and start Silicon Data. From our conversation at NVIDIA GTC Berlin.
Short video / excerpt
You Can't Be a Product Manager and Not Use Your Product
NVIDIA's Adel El Hallak on one of his operating principles: product managers now have to use their own products, and AI agents make that possible even with deep technical tools. Ben Pouladian adds how he builds software with AI. From our conversation at NVIDIA GTC Berlin.
From the channel
NVIDIA After Jensen: What Actually Changes?
Jensen Huang stepped down as NVIDIA CEO. I went live with Nick Dorsey to talk through what the board owes shareholders, what a leadership change means for the business and the stock, and what I’d look for in the next CEO. This is the edited cut of our emergency livestream.
Short video / excerpt
Why Elon Musk Says He’s Signing the Jensen Huang Petition
Last year, Elon Musk talked about who sticks around when the chips are down. Now he says he’s signing the petition asking NVIDIA to bring Jensen Huang back.
From the channel
AI Agents Need Seatbelts | NVIDIA's Adel El Hallak at GTC Berlin
How do you give AI agents useful autonomy while keeping enterprise data and systems under control?
Short video / excerpt
Open vs Closed AI Models? NVIDIA Uses Both
NVIDIA's Adel El Hallak on why every agentic system combines different models: a frontier model like Opus orchestrating, and an open model like Nemotron researching, planning and executing. He argues open models are about accuracy, not only cost. From our conversation at NVIDIA GTC Berlin.
Short video / excerpt
Is AI Circular Financing? Why GPUs Earn Revenue
On Fox Business, Ben Pouladian pushes back on the circular-financing argument. He points to Apollo's lending against large GPU assets and its reported $40 billion deal with SpaceX, and argues that GPUs are revenue-generating assets that produce tokens.
Short video / excerpt
Cost Per Completed Task: How NVIDIA Measures AI Agents
Ben Pouladian argues the next metric for enterprise AI is cost per completed task. NVIDIA's Adel El Hallak explains why every agent has to be instrumented, and how NVIDIA's unified telemetry collects traces from the agents deployed inside the company. From our conversation at NVIDIA GTC Berlin.
Short video / excerpt
What Happens When You Give an AI Agent the Keys?
NVIDIA's Adel El Hallak on "unbounded autonomy": handing a new agent the keys to the data center, the CEO's office and the wallet. He explains how OpenShell starts from deny by default, and how a policy prover checks what agents are allowed to do together. From our conversation at NVIDIA GTC Berlin.
Short video / excerpt
Why I'm Buying the NVIDIA Selloff
I’ve been an NVIDIA shareholder for nearly a decade, and I’ve seen plenty of reasons to sell along the way.
Short video / excerpt
Why Investors Should Watch Jensen's GTC Keynotes
On Fox Business, Ben Pouladian explains why he tells investors to watch Jensen Huang's GTC keynotes, and how hardware in the loop could connect lab instruments and robotic arms to AI models in drug discovery.
Short video / excerpt
NVIDIA's Chip Design Agents: From Weeks to Hours
NVIDIA's Adel El Hallak on how teams of agents are being applied to chip design with Cadence. He describes a test, debug and regression loop that he says used to take weeks and now runs in hours. From our conversation at NVIDIA GTC Berlin.
Short video / excerpt
Jensen Huang: A Standard of Responsible Optimism #Shorts
Jensen Huang's standard: responsible optimism. At GTC Berlin, he argued that adults should hold themselves to that standard.
Short video / excerpt
Jensen Huang: AI Safety Requires Engineering #Shorts
Jensen Huang on developing and deploying autonomous AI systems safely. He argues that safety requires technology and engineering.
Short video / excerpt
Jensen Huang: Why He Expects More Optics #Shorts
Larger data centers. More GPUs. What does that mean for networking? At GTC Berlin, Jensen Huang explained why he expects demand for switches and optics to grow.
From the channel
AI Bubble or Buildout? Nvidia, CoreWeave & the AI Trade | Ben Pouladian on Fox Business
Is the AI trade running out of steam, or is the infrastructure buildout still getting started?
Short video / excerpt
How Long Can a GPU Stay Valuable?
How long do GPUs keep their economic value? Thomas Galbraith of Barkr says resale data challenges common assumptions about useful life. In this clip from NVIDIA GTC Berlin, he explains why he thinks better market understanding could show up in yields. Full interview tomorrow at 5 AM PST.
Short video / excerpt
Jensen Huang: No AI Without Energy #Shorts
At GTC Berlin, Jensen Huang described AI as a manufacturing industry and argued that its buildout can help fund the future of nuclear energy.
Short video / excerpt
Jensen Huang: Partner With Nokia, Don't Disrupt It #Shorts
At GTC Berlin, Jensen Huang described NVIDIA's approach to working with established industries. His Nokia example: build technology together and help partners grow.
Short video / excerpt
Jensen Huang on NVIDIA's Underdog Mindset #Shorts
Jensen Huang discusses NVIDIA's future, staying an underdog and avoiding complacency at GTC Berlin.
Short video / excerpt
Can Airplanes Explain GPU Value?
An AI chip and an airplane might seem like an odd comparison. Thomas Galbraith of Barkr explains how both have an asset value and income from leasing, while making clear that the analogy has limits. From our conversation at NVIDIA GTC Berlin. Full interview Friday at 5 AM PST.
Short video / excerpt
What Is an AI Chip REALLY Worth?
New chip launches get the headlines. What happens when those GPUs are resold? Thomas Galbraith of Barkr shares a real-market example from our conversation at NVIDIA GTC Berlin. Full interview Friday at 5 AM PST.
Short video / excerpt
Jensen Huang: AI Jobs Go Beyond Software #Shorts
Where does Jensen Huang see jobs from the AI buildout? At GTC Berlin, he pointed to the energy sector, construction workers, plumbers, electricians and networking engineers.
Short video / excerpt
Jensen Huang: Why Better Technology Makes Us Busier #Shorts
Does better technology give us more free time, or bigger ambitions? At GTC Berlin, Jensen Huang argues that productivity gains expand what people take on.
Short video / excerpt
Jensen Huang: AI Could Explore 1,000 Car Designs 🚗 #Shorts
What if AI could explore 1,000 car designs? 🚗
Short video / excerpt
How NVIDIA Cools AI Chips | Dion Harris at GTC Berlin 2026 #Shorts
NVIDIA’s Dion Harris explains closed-loop liquid cooling in this excerpt from my GTC Berlin 2026 interview. He compares recirculating coolant with evaporative cooling in the systems we discussed.
Full interview
Inside NVIDIA’s AI Factory Strategy: DSX POD, Power & Allocation | Dion Harris | GTC Berlin 2026
The AI race runs through power lines, cooling loops and construction sites.
Short video / excerpt
TPUs ≠ GPUs. Here’s why CUDA matters 🧠 #Shorts
A TPU and a GPU can both do AI math, but the hardware and software ecosystem are different. Here’s why NVIDIA’s CUDA stack matters.
Short video / excerpt
Jensen Huang: Who Teaches the Robots? 🤖 #Shorts
At GTC Berlin, Jensen Huang compares teaching specialized skills to a robot with training an apprentice.
Full interview
Google's Chip Gambit: TPUs, NVIDIA and the AI Compute Race | Ben Pouladian on TITV
Google's TPUs are moving beyond its own cloud. What does that mean for NVIDIA, AI infrastructure and the companies building on it?
Short video / excerpt
Why AI Agents Want Faster Simulations | NVIDIA's John Linford #Shorts
I asked NVIDIA's John Linford how AI agents will use engineering tools. His view: they'll reach for faster options.
Short video / excerpt
Jensen Huang: Why AI Data Centers Make Your Brain Explode 🤯 #Shorts
Jensen Huang explains why building AI data centers is so complex—and why the pressure to bring them online is rising.
Full interview
NVIDIA cuDSS: Faster Simulation for Chips and Cars | John Linford
Ben Pouladian speaks with NVIDIA’s John Linford about cuDSS 1.0, sparse solvers, and accelerating engineering simulation—from chip design to virtual wind tunnels for cars.
Research briefing
NVIDIA’s CPU Bet: Where AI Agents Run Their Code
https://newsletter.bepresearch.com/p/nvidia-wants-the-computers-behind
Short video / excerpt
How AI Data Moves: From GPU Racks to Optical Networks | BEP Research
AI infrastructure gets a lot more interesting when you follow the data.
Short video / excerpt
NVIDIA GPUs: The Fifth Avenue of Compute | Ben Pouladian
I compare NVIDIA GPUs to Fifth Avenue real estate: when one tenant leaves, who wants the next lease?
Short video / excerpt
The Next Inflection Is the Lab | BEP Research AI-Science Basket
AI is moving from answering questions to running experiments.
Full interview
Why AMD Is the Most Overhyped Chip Stock | Ben Pouladian on Meta’s Muse, Anthropic IPO, & Semis
Ben Pouladian joins Jack Farley on Monetary Matters to discuss Meta’s Muse, the economics of agentic AI, NVIDIA versus AMD, data center power constraints, Anthropic’s IPO filing, and AI’s move into biology.
Research briefing
Agent Wars: Grok Bot vs Muse vs Dots — Bigger AI Bill?
https://newsletter.bepresearch.com/p/devday-cheaper-ai-faster-agents-nvidia
Short video / excerpt
Anthropic passed OpenAI. Did OpenAI just take it back? #OpenAI #Anthropic #AI
Anthropic went from about $9B to $65B in annualized revenue run-rate between the end of 2025 and July 2026, passing OpenAI in the spring. Axios reports OpenAI is now approaching $70B.
Full interview
NVIDIA, OpenAI and the AI Rent-or-Own Decision
Ben Pouladian joins ETN’s Luke Knight and Ronan Chambers to discuss NVIDIA, open models, cost per task, and AI’s next inflection in the lab.
Research briefing
Free Isn't Cheap Enough: What GPU vs TPU Headlines Leave Out | BEP Research
Read the full article: https://bepresearch.substack.com/p/free-isnt-cheap-enough
From the channel
NVIDIA: The Chip Is Dead, Long Live the Factory
Two days after the print, and the argument Ben Pouladian keeps making is that people are still
pricing NVIDIA as a company that sells chips. What it sells is a building that turns electricity
into tokens, and almost every competitive question in AI right now falls out of that one
distinction.
From the channel
NVIDIA's Q2: Supply Is the Cap, Not the Demand
NVIDIA guided into a ceiling it cannot lift fast enough, and the market spent the next morning
calling the ceiling a slowdown. Ben Pouladian's read the day after the print: the quarter never
had a demand problem, and almost nobody is pricing what is arriving behind it.
From the channel
He Led People Blindly Into the Slaughterhouse: The Nebius Short
Michael Burry told his subscribers he was short Nebius. Four sessions later the company printed
454% revenue growth and the stock closed up 34%. Ben Pouladian's read on air: "He led a lot of
people blindly into the slaughterhouse."
From the channel
Google Has Become the Elder Statesman IBM of the Hyperscalers
Hours after Demis Hassabis stepped back from the DeepMind CEO seat and Jeff Dean left to found
Discovery Loop, Ben Pouladian joined the European Technology Network with a four-month-old
receipt: DeepMind had one talk and no booth at Google Cloud Next, and he wrote about it at the
time.
From the channel
A Generation Behind: AMD's Big Helios Ramp Is Q1 2027 The Information TV
AMD grew 50% in Q2 and the stock sold off anyway. Ben Pouladian's read is that the market stopped
looking at the print and started reading the ramp schedule.
From the channel
I Couldn't Find the Chip": An Open Letter to AMD's Lisa Su Ben Pouladian
AMD says its new Venice CPU beats NVIDIA's Vera by about 20% on single-core performance. Ben
Pouladian of BEP Research read all 132 pages of AMD's own Advancing AI deck looking for the
part behind that score and could not find it: no model number, no wattage, no price, and
nothing matching it on AMD's published price list. He published an open letter to Lisa Su at
6am the morning of this interview. AMD reports Q2 the next day.
From the channel
Ben Pouladian on the Memory Crash, Bloom Energy and Open Models
Memory stocks gave back a third of a parabolic run in a month and a lot of people decided the
AI trade was finished. Ben Pouladian of BEP Research reads it as leverage coming out rather
than a thesis breaking, then makes the harder argument: memory was never the binding
constraint anyway.
From the channel
Why Etched's Inference Chip Has a Depreciation Problem GPUs Don't
Etched came out of stealth on June 30 with $800M raised and over a billion in customer contracts, and X spent the day calling it the next NVIDIA killer. Ben Pouladian (BEP Research) joined Jack Farley and Max Wiethe on MTS Live to explain the part the hype skipped: a chip with the transformer architecture etched into fixed-function silicon is frozen at tape-out, while a GPU keeps getting faster in software over the same life. The bet Etched is making is that attention never gets displaced.
From the channel
Nebius vs CoreWeave: Debt, Equity, and Who Wins the Compute Crunch | Ben Pouladian
Ben Pouladian (BEP Research) breaks down the real difference between Nebius and CoreWeave: how Nebius rolls with equity and shorter-term contracts to capture spot pricing, while CoreWeave's debt-laden, long-term-contract model trades pricing power for stability. Plus Nebius's Yandex roots, its custom-everything build approach, the Microsoft prepay, and why the Bitcoin-miner-to-HPC crowd faces a consolidation reckoning when margin compression hits.
From Ben's conversation with Jack Farley and Max
From the channel
The AI Compute Shortage Is Worse Than You Think | Ben Pouladian
Ben Pouladian (BEP Research) joins Jack Farley and Max Wiethe to break down where we really are in the AI build-out: whether token demand is peaking, why Google reportedly paid double the market price to rent compute, how Nebius and CoreWeave differ on debt vs equity and pricing power, the coming neocloud consolidation, the photonics timing debate out of COMPUTEX, and the bear case on high bandwidth memory (and why he's still long NVIDIA).
Ben is the founder of BEP Research, an institutional-gra
Short video / excerpt
NVIDIA's VP Tells Me Most People Don't Understand This Ben Pouladian | GTC 2026
An interview from BEP Research’s GTC 2026 coverage, hosted by Ben Pouladian.
From the channel
The Orchestration Layer Is the New Platform War — Ben Pouladian | GTC 2026 #nvidia #software
BEP commentary from GTC 2026 on the software layer coordinating AI infrastructure.
From the channel
GTC 2026: The Future of AI Infrastructure — NVIDIA, Groq & the Inference Revolution | Ben Pouladian
Ben Pouladian’s GTC 2026 overview of NVIDIA, Groq, and the inference stack. A starting point for BEP’s system-level perspective.