All-In PodcastJensen Huang: The Doomer Hoax, Superintelligence is Here, and The Future of AI (ft. President Trump)
All-In Podcast
Introducing Jensen Huang 0:00
The episode opens with the hosts introducing Jensen Huang, founder, president, and CEO of Nvidia, treating him as one of only three guests important enough to preempt their regular show, alongside President Trump and, jokingly, Jesus. He arrives to a standing ovation and is greeted warmly before the conversation turns to serious industry matters.
Reacting to Dario's essay 1:30
The hosts ask Jensen to respond to a widely discussed essay, apparently from Dario Amodei, that touched on AI safety, an internal whistleblower situation, and predictions about danger from AI. Jensen says safety matters but insists safety and leadership are not opposing choices, since America can innovate quickly, execute quickly, and still act safely. He takes the whistleblower concern seriously, praising the courage involved, but separates that from what he calls unscientific predictions about the future mixed into the same piece. He suggests the real issue might be labs struggling through the difficult transition from research culture to engineering discipline, and that this transition can be clumsy.
A history of failed predictions 4:31
Jensen pushes back hard on catastrophic forecasts, calling them made up and irresponsible when they come from credentialed researchers. He lists examples that proved false: radiology being fully replaced by AI in five years, when instead more radiologists are needed even as scan-reading became automated; predictions that ninety percent of code would be AI generated within months; and forecasts that fifty percent of entry-level jobs would vanish. The hosts add more failed predictions, including claims that GPT-2 and Llama 3 would be too unsafe to release, and warnings of a coming jobs apocalypse. Jensen argues these repeated wrong calls undermine the case for slowing down and are inconsistent with America winning the AI race.
Why companies stay quiet 7:31
Asked why leading AI labs seem to generate so much public alarm, Jensen says these are consequential companies doing extraordinary engineering, and such companies should be built quietly the way companies used to be built. He explains that at Nvidia, employees are told to keep political and social discourse outside the company, since Nvidia is deliberately apolitical and bipartisan, focused on helping America succeed regardless of who is in government. He credits this stability and consistency, along with care for employees and meaningful work done quietly, for the company's happy culture.
What regulation should target 10:00
On AI regulation, Jensen argues rules should solve actual demonstrated problems, and so far the only real incidents have come from frontier labs, simply because they hold the most computing power and take on the hardest problems. He notes a high school student or ordinary startup could not cause similar harm for lack of compute. He expects labs to root-cause their incidents, build better sandboxes, runtimes, and continuous monitoring, and prevent repeats through normal engineering discipline rather than needing outside rescue.
Debating self-improving AI 13:31
The conversation shifts to a Chinese lab reportedly committing three billion dollars toward recursive self-improvement, AI that helps train the next AI. Jensen calls this a sensible combination of existing techniques like in-context learning, reflection, reinforcement learning, synthetic data, and methods like Lora for improving model weights without retraining the whole base model. He doubts it will spiral out of control, since any product still has to pass evaluation, testing, and regression checks before release, and better lab controls will only improve as labs mature.
Open models versus closed models 16:00
Jensen argues the AI ecosystem needs both closed and open models, comparing closed models to bottled water, useful but not the only option, since raw water, like open models, is free and abundant. He cites four hundred billion dollars in venture funding into AI native companies over six months, eighty percent of which relied on open models to build their businesses. He notes China contributes much of the world's open-source work due to its larger population of engineers, but stresses that once downloaded, any open model becomes the user's own to modify freely.
Who actually wins the race 19:30
Jensen reframes the AI race as being less about who invents the technology and more about who exploits it best, comparing it to the last industrial revolution, whose inventors, like Maxwell, Volta, and Ampere, were mostly European even though the West went on to capitalize on the technology. He contrasts America's doom-laden narrative with China's more pragmatic framing of AI as a tool for economic advancement, suggesting the alarmist tone in the US is not grounded in science and mainly worries people who cannot act on it anyway.
Engineering without typing 21:00
Reflecting on his own career, Jensen recalls starting as an engineer before software made typing central to the job, having to build computers before software existed. He predicts a future where much engineering work no longer centers on typing code, joking that his favorite computer key is backspace because the best software is the smallest software, and telling his own engineers they are "just typing."
A surprise call from Trump 23:03
Midway through the taping, President Trump calls in and is put on speakerphone. He jokes that Jensen can build the world's most advanced computer chip but cannot figure out how to answer a phone call. Trump declares that fears about AI ending the world are "a hoax," arguing that AI and data centers are creating wealth for struggling communities and states, comparing AI's economic importance to oil over the next twenty to twenty five years and calling it bigger than the internet.
Trump on winning the AI race 25:30
Trump insists robots will not take over the world and that alarmist narratives serve political opponents or China, adding a personal note that his uncle was a longtime top professor at MIT, which he jokes gives him some genetic insight into AI. He states that whoever wins AI wins overall, calls it bigger than the internet, and vows the administration will act prudently but will not let fear stop the industry, promising that every state, company, and industry in America will benefit from the AI race.
A Surprise Call From The Oval Office 27:32
During the event, President Trump calls in to speak to the crowd, telling them the country has never done better and citing 20 trillion dollars of incoming investment compared to under 1 trillion during the Biden administration. Jensen Huang admits he thought the call was staged at first, then realizes it is real, recalling an earlier moment when he was asleep and the President had him woken up and put on the phone during a dinner planning session, even though Huang had postponed a five-year-delayed vacation to be reachable.
Why Trump Rejects The AI Panic 29:02
Huang is asked why the President seems to see through what he calls the AI doomer hoax, even though public opinion polls at minus 80 would normally push a politician toward the popular position of shutting down data centers and AI. Huang says he isn't fully sure why, since so many people are falling for the fear narrative. He notes the story first rested on national security concerns, which have since been disproven, and has now shifted to safety. His answer is that AI labs should be tested and held to extraordinary standards, with multiple independent third-party evaluators similar to financial auditors, so no single company becomes overly influenced or captured.
Jobs, Energy, And Reindustrialization 31:32
Huang explains that Trump's core priority has always been creating American jobs and re-industrializing the country, including building the energy capacity needed for what he calls the next industrial revolution. He points to 400 billion dollars of venture financing flowing into the AI industry in just six months, which is generating software jobs and massive demand for compute and data centers. He mentions a conversation with Texas Governor Abbott about being sensitive to small communities affected by data center construction.
Nvidia As Financier Of The AI Buildout 32:30
Huang describes Nvidia's role as effectively becoming the bank of the AI industry, financing land, power, and infrastructure deals with partners like Cloverleaf, BlackRock, and Goldman. He frames AI as a new industrial revolution requiring manufacturing at every layer, from chips to data centers to electricity generation, and says he watches the entire supply chain for bottlenecks, much as Nvidia long ago worked with suppliers like Corning, Lumentum, TSMC, and memory makers to prepare for growth before it happened.
Running Every Model, Staying Uncompetitive 35:30
Huang says Nvidia now runs nearly every major AI model, a shift from a year and a half ago when OpenAI was the only one on its platform; now Meta, Grok, Gemini, and Anthropic's models all run there too. His stated strategy is to build technology as far up the stack as needed but as low as possible, so smaller labs and companies can build on top rather than being squeezed out, pointing to Nvidia's creation of foundational tools like cuDNN and Megatron Core as enabling the entire ecosystem. He says he would welcome more hyperscaler-level competition, noting that Neoclouds have become important because hyperscalers plan only once a year while market dynamics move too fast for that, and that countries increasingly treat compute as strategic, with Nvidia partnering with regional players in places like Australia and Southeast Asia.
Building Frontier Models Out Of Necessity 40:01
Huang confirms Nvidia is the frontier model in five domains, driven by necessity rather than a wish to win outright. He cites Alpamo, described as the world's first thinking self-driving car that reasons about situations rather than needing billions of hours of training data, built because most car and truck companies lack the scale to build the whole stack themselves. He also lists biology-focused models Nvidia created, including ESM2, ESMFold, OpenFold, AlphaFold2, and Proteina Complexa, saying companies like Lilly and Merck need this work and Nvidia fills the gap because it can.
Elon Musk And China's Chip Progress 42:30
Asked about Elon Musk's announced 100 million square foot Terrab facility, Huang says if anyone can pull it off, Musk can, recalling a flight they shared where they discussed it at length. On China's progress toward advanced lithography systems, Huang predicts they will get there by 2030, adding that China is very good at high volume production and that in the meantime it barely matters since two or three years is a short span when thinking in decades.
Already Living In Superintelligence 44:31
Huang argues that within narrow domains, superintelligence has already arrived, pointing to self-driving cars with one-tenth the accident rate of humans and AI systems doing protein synthesis and virtual screening. He says he is having fun being on this frontier and hopes the drama around AI can be toned down, while urging that all of America come along for what he calls a future where humanity succeeds together.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
