"I can deeply empathize with Jensen Huang's situation!" Sam Altman made a rare statement: It is unimaginable for NVIDIA to exist without him.
From the fate of Jensen Huang, to computing power investment in the AI industry, and the question of whether superintelligence will make humans lose control, Sam Altman talked about a series of key issues facing the AI industry in his latest interview.
He expressed strong respect for Jensen Huang, deeply empathized with Jensen's current situation, and stated that it is hard to imagine an AI revolution without Jensen. Regarding competitor Anthropic's listing this week, he said he is not worried about its early listing, and believes there is still huge capital and commercial space for AI.
He also reminded that chip manufacturers may have overlapping expectations for demand, and the DRAM shortage has exposed the problem of lagging supply chain expansion. In the face of public resistance to AI and data centers, he believes the industry must regain trust and make a clear commitment that humans will always hold control. Even if AI begins to accelerate its own R&D, OpenAI will not abandon human supervision and safety boundaries.
From data privacy to AI scientific research, Altman's attitude is equally clear: users should own their own data, and the most anticipated long-term value of AI is to drive scientific discoveries, engineering innovation and productivity leaps.
TL;DR Version
Q: Do you think Jensen Huang should return to NVIDIA?
Sam: Jensen has always been someone I admire greatly. I can particularly empathize with Jensen's current situation, because I remember exactly how I felt back then. It is hard to imagine the AI revolution without Jensen.
Q: Will Anthropic going public first affect investor demand for OpenAI?
Sam: I'm not worried. I'm not worried at all about a lack of sufficient investor demand in the market. In fact, looking back over the past year, there was far more investor demand than available investment opportunities and stock supply. The entire industry is creating incredible value.
Q: Is there duplicate construction and overestimation of demand in the AI chip industry?
Sam: Right now many companies are producing chips, and quite a few of them assume that most of the chips they make will be sold to us, Anthropic and other companies. But our demand, the scale of data center construction, and other related demands all have a certain upper limit. Market demand and data center construction itself are solid, but it is impossible and unnecessary for us to buy the full production capacity from every chip supplier.
Q: Why is the public increasingly resentful of AI and data centers?
Sam: If we promise the public to cure cancer and lower commodity prices, but in return make them lose their autonomy, control and say over the future, that is a terrible way of promotion. I don't want to live in such a world, and I don't think most people want that future. We should not hand over the future to AI and let it make decisions for humans. AI should be a tool for people to use.
Q: Has OpenAI entered the RSI stage?
Sam: Are we using our own technology to develop new models faster than we could without it? Yes. Can we already press a button, tell the model "please design the next generation of models", and then quickly achieve self-improvement without any human supervision? Absolutely not.
Q: Can users export Dots' memories to avoid being locked in by the product?
Sam: Dots' memory function is more complex than that of ChatGPT. I don't have a ready-made solution right now, but I agree that users should own their own data and be able to use it as they wish.
Q: Do you oppose ordinary people installing large-capacity memory at home and running local models?
Sam: If there is a shortage of memory supply, the ideal way to use resources is to make memory shareable and keep it working 100% of the time, instead of being placed in someone's edge device and only used occasionally. Of course, the highest resource utilization rate is not necessarily the only reason that should be considered when making decisions.
Q: Will data center construction be restricted by debt and public opposition?
Sam: I think data centers will eventually be built. The specific locations, construction methods, and how to make them align with the interests of local communities will all change. Every gold rush comes with some bubbles, and there will be some foolish investments in the market. The pace of data center construction may outpace the speed at which the entire economy can absorb this capacity, but I have not seen such a level of blind expansion yet.
Q: How much impact will AI have on scientific research?
Sam: Some scientific discoveries will quickly transform society and the economy, while some fundamental breakthroughs may take decades to be translated into practical applications. Right now, mathematicians have used AI to solve problems that were completely unsolvable in the past, and physicists, chemists and materials scientists are also starting to make similar progress. I think the ability of the next few generations of models to drive scientific discoveries will really exceed many people's expectations.
The full conversation below has been organized and abridged on the premise of not changing the original meaning, for the benefit of readers.
1 "I can particularly empathize with Jensen's current situation"
Host: First I want to ask, did you watch Saturday Night Live this weekend?
Sam: I just watched it.
Host: How did you feel about it?
Sam: It was a little funny. I quite like Saturday Night Live.
Host: Just a little funny? Alright. This episode is about the Jensen Huang controversy. Almost 3 years ago, you went through a similar thing: you were removed by the board of directors at that time, and later a joint letter in support of you was launched. A recent circulating letter in support of Jensen, I think, is almost a copy of the letter drafted by your supporters back then. So I want to ask, do you think Jensen should come back?
Sam: I don't have enough information to answer that question. Jensen has always been someone I admire greatly. Of course, heroes have flaws too. I am very grateful for the company he founded. I can particularly empathize with Jensen's current situation, because I remember exactly how I felt back then, even though it feels like it was a long time ago. It is hard to imagine the AI revolution without Jensen. The company he founded chose to support us very early on, and we also chose to support him very early on. I am extremely, extremely grateful for that.
Host: NVIDIA also provided a guarantee for your data center in Ohio, I remember the amount is 105 billion US dollars. They are like a bank, the financial supporter behind the entire industry. Do you think the weight of this guarantee today is still the same as it was a few weeks ago?
Sam: NVIDIA is an extremely large company, and I think people still have great trust in its guarantees. But yes, what would NVIDIA look like without Jensen? Jensen and NVIDIA are obviously very closely linked. I think the ambition and scale he brought to this company are very important to the entire industry.
Host: So you think this incident will have a ripple effect and impact the entire industry?
Sam: I think the current market demand is very strong, AI is indeed working, and people really want to use it. Related revenue and demand are growing rapidly, so I think the entire industry will remain healthy. But Jensen does play a very special, very valuable role, pushing the entire industry forward with great ambition.
2 Anthropic's listing will not affect investor demand for OpenAI
Host: Anthropic is going public this week, but investors seem to have a lot of doubts about the market now. Do you really want OpenAI to go public yourself? If you could avoid it, would you choose not to go public?
Sam: I think it's mostly journalists who are skeptical of this industry, and investors still seem quite optimistic about AI. Of course, the market will have ups and downs, prices will rise and fall, the industry will go through phases of overbuilding and underbuilding, things will go wrong, and of course there will be times when things go exceptionally smoothly. But the overall trend is that enormous value is being created, while a lot of existing value is being disrupted. I think investors have realized that. We also felt that when we heard about the scale of capital investment backstage just now.
As for whether I want OpenAI to go public, we will eventually go public at some point. I think going public does have some obvious downsides. Right now the industry is very volatile, involving issues of security and how society will coexist with this technology, many decisions are very important and far-reaching. As a private company, some things are undoubtedly easier to handle.
On the other hand, it is not just investors who happen to have access to the private market that want to invest in this technology, the whole society wants to participate in it. In that sense, I think it is a good thing to give more people the opportunity to invest.
Host: Then if Anthropic goes public first, will there still be enough investor demand later? Will you become Lyft while the other party is Uber? Are you worried about that scenario?
Sam: I'm not worried. I understand you need to write every day, you always have to create some topics, like this competition, that competition, who won, who lost, what happened again. But I focus on the broader trends and the value that is being created.
I'm not worried at all about a lack of sufficient investor demand in the market. In fact, looking back over the past year, there was far more investor demand than available investment opportunities and stock supply. But I think we can continue to build, and the entire industry is creating incredible value.
3 "The entire supply chain is already very large and highly complex"
Host: You previously spoke to Samsung executives at a closed-door meeting that there is a certain degree of double-counting in the AI industry. What exactly did you mean by that, can you explain it?
Sam: Of course. For a very long time, there was basically only one chip supplier in this industry. Therefore, when counting wafer demand, supply chain companies could ask companies like us about our demand expectations, to roughly estimate how much end-user demand there would be for the chips they produce.
That's different now, there are many companies producing chips, and quite a few of them assume that most of the chips they make will be sold to us, Anthropic and other companies. But our demand, the scale of data center construction, and other related demands all have a certain upper limit. I think many chip manufacturers are communicating with the entire semiconductor supply chain right now, and all of them are assuming that we will buy all the chips they produce.
I want to emphasize that market demand and data center construction itself are solid, but it is impossible and unnecessary for us to buy the full production capacity from every chip supplier.
Host: What does that mean for the entire industry? Does it mean the industry is overheated and many things are overvalued?
Sam: I am sure the market will have ups and downs. In the long run, the market's demand for massive computing power, and the value that computing power can unlock, are very clear. Right now, people are already using this technology on a large scale and creating incredible results. Just like other technology revolutions, the long-term trend is clear, but it is much harder to accurately predict whether supply and demand will perfectly match every quarter or every year.
Host: You just said that in the past chip suppliers expected you to buy all the chips. I originally thought you did want to buy all the chips. Now that NVIDIA no longer supplies goods to China, does that mean you have a chance to get more supply?
Sam: The supply chain is very complex. At different stages, we are constrained by different factors. Sometimes it's memory, sometimes it's rack assembly, sometimes it's data center capacity, and sometimes it's even a completely unexpected component. For example, it seems incredible: we cannot provide more services because of the global shortage of PCB copper clad laminate supply. Right now the entire supply chain is already very large, and the level of complexity is also very high.
4 Why there is always anti-AI sentiment among voters
Host: Before Jensen's incident, another news that got attention was that the Democratic Party gained an electoral advantage. I saw that except for one data center-related referendum that failed, all others passed. In other words, a lot of people don't like data centers. Anti-AI sentiment has emerged among voters from both parties. I'm curious, what exactly did the AI industry do to frame this technology so poorly that the public is so resentful?
Sam: If we market this technology to the public by saying: "We will provide you with a cure for cancer, and make goods cheaper." But the tradeoff is: "You will lose your autonomy, control and say over the future. You will live like a Labrador retriever. The machine may try hard to make you happy, but you will not be able to decide for yourself how to be happy. We will rush to hand over control of the future to AI, and then make AI very friendly so it takes good care of you."
That is a terrible way of promotion. I don't want to live in such a world, and I don't think most people want that future.
Of course, there are also many economic issues at play here. But I think if you talk deeply to people who really dislike AI, and get past their initial stated concerns, you will find that data centers are just an easy target to point to, because they are the physical embodiment of a larger overarching problem in the real world. Employment is a very immediate, strongly felt concern for people right now.
But there is also a long-term question: What role will people play in society in the future? What will the power relationship between humans and AI look like? What kind of power gap will form between people who own a lot of AI and computing power, and those who do not have these resources?
I think the AI industry must take a very clear stance on this issue: we should firmly support and commit to ensuring that humans always hold control over the future. We should not gradually or suddenly hand over the future to AI, and let AI decide the future for humanity. The common term used in Silicon Valley right now is "handover", and that should never happen. AI should be a tool for people to use.
I think the public can perceive that not everyone in the AI industry holds this stance, so their fears are not unfounded. Coupled with economic uncertainty and various cybersecurity issues caused by AI, it is natural for people to oppose data centers. But from my experience talking to them, what they are really opposed to is a sense of uncertainty: whether the people building AI are truly on humanity's side.
Host: You once said in July 2025 that you were "politically homeless", and that the Democratic Party had lost its culture of innovation. Do you still think that way now?
Sam: By and large, I still do.
Host: How does that affect your view of the recent election results? Obviously, you want to work with any elected leader.
Sam: Yes, I want to work with any democratically elected leader. I still think the United States needs a political party that is more aligned with the ideas I identify with, and closer to the longstanding innovative spirit of America. It would be great if such a party could emerge. I hope both parties move in that direction, and I think there are some positive signs on both sides.
Host: You have contacted government officials in the past. You can call David Sacks, you can reach Susie Wiles. What about Stephen Miller, one of Trump's most core and longest-serving confidants? Can you get in touch with him?
Sam: I don't know, we can try.
Host: When was the last time you talked to Trump?
Sam: At a state banquet in China.
Host: What did you talk about back then?
Sam: We didn't get a chance to have a very in-depth, very detailed conversation, there were a lot of topics.
Host: Someone once secretly installed a microphone in your office. Have these things changed the way you interact with people? You have a lot of investors from the Gulf region, will you look at them differently because of this?
Sam: In fact, our experience in this regard may be richer than the outside world imagines. We adapted very early to the reality that AI is an extremely important thing, and many people are trying every means to gain a competitive advantage. Someone might install a camera in our conference room, someone else might ask our engineers out for a few drinks and try to extract confidential information about pre-training technology, and so on.
So we have actually been living in this environment for some time, it's just that other people are only starting to publicly and suddenly realize this reality now. We get tricked sometimes, our operational security and overall security measures are definitely not perfect, but we have had time to gradually adapt