两位传奇创始人：Travis Kalanick 与 Michael Dell 在得州奥斯汀现场对谈
Travis Kalanick 正将 City Storage Systems 更名为 Atoms，打造覆盖食品、采矿和机器人移动的物理自动化平台。 其架构把制造、房地产和物流分别视为CPU、存储和网络在物理世界中的对应物：「把原子当作比特来处理」。这一投资逻辑远不止云厨房：公司希望用专用机器改造各个行业。
眼下物理AI最现实的机会是“有偿工作的机器人”，而不是人形机器人的炫技。 Atoms 即将完成对旧金山 Pronto 的收购，以自动化采矿设备；这可能提高现有矿山的生产率，并凭借更小的用工和安全成本，让偏远、恶劣环境中的矿藏具备开发价值。Kalanick 认为 Waymo 是自动驾驶领头羊，但受制于制造、规模、紧迫感和竞争强度；Tesla 则在没有其他传感器的情况下，仅靠视觉挑战“困难模式乘以100”。
物理AI仍在等待自己的ChatGPT时刻，也需要能源效率实现跃升。 Kalanick 提问「视觉领域的ChatGPT时刻何时到来」，同时表示，人类运行所需功率约为100瓦，而 Waymo 机器开车消耗的能量约为人的100倍。驾驶、感知和安全代理之间类似语言的通信，可能成为强大的压缩层，但他将其描述为一种可能性，而非已经确定的解决方案。
只有当市场结构把融资变成核心竞争能力时，资本才真正重要。 在 Uber 时代，Kalanick 表示，竞争对手可以从 Masa 手中拿到10亿美元，第二天就夺走20%的市场份额，资本因此成为真正的战略武器。他不愿对冲突后的中东融资作出自信推测，但表示自己的中东业务原定于1月上市，随后沙特股市在两个月内下跌约20%。
Michael Dell 的AI基础设施业务正从约20亿美元扩张至100亿美元、250亿美元，今年预计达到500亿美元。 Dell 仍然认为市场「需求远超供给」，需求来自超大规模云厂商、主权AI，以及4,000多家 Dell AI Factory 企业客户。得州则凭借充足的土地、电力和建设能力，进一步放大这一机会。
企业AI在部分场景可以带来20%或更高的生产率提升，但前提是自上而下重构业务。 Dell 表示，大型企业中或许只有10%-15%真正摸清了方法，其余企业仍在「摸索」，因为采用瓶颈不在技术，而在「文化、领导力和勇气」。Dell 的基础设施业务上季度增长73%，公司预计下一季度增速接近100%。
推理正在向数据所在位置扩散，而不再完全集中于云端。 Dell 认为，「成本最低的token」将在设备端生成；边缘推理和 Gemma、OpenAI 的开放模型、NVIDIA Nemotron 等本地模型也在快速增长。自主代理扩大了机会，同时也让身份、认证、验证和控制机制变得不可或缺。
Invest America 试图让每一个符合条件的孩子都看见“拥有资产”以及资本主义上行空间。 Michael 和 Susan Dell 宣布捐赠62.5亿美元，为符合条件邮编区域内2-10岁的2,500万名儿童各提供250美元。Brad Gerstner 介绍称，18岁以下儿童都可以申领账户；从2027年1月1日起在美国出生的儿童则会自动获得账户。Michael Dell 估计，该项目最终可能在15年内向家庭转移5万亿美元。比种子资金更重要的是其行为逻辑：「我在场内，我有机会」（“I’m in the game. I have a shot.”）。
1. Atoms 将工业基础设施重新定义为一台计算机
Kalanick 的核心框架始于「数字化物理世界」。CPU处理比特，存储保存比特，网络传输比特；在以原子为基础的计算机中，制造处理物质，房地产存储物质，运输或物流移动物质。
City Storage Systems 有意保持「无聊到极致」，但它的第一个产品本质上是一台食品计算机：围绕「为更好的食物提供基础设施」这一使命，将制造、房地产和物流组合起来。新宣布的 Atoms 品牌则把这套架构扩展到其他行业。
公司的食品目标，是以足够高的效率备餐和配送，让成本接近去超市买菜——「对厨房做 Uber 对汽车做的事」。但与 Uber 不同，所需产能并未闲置在那里：Kalanick 表示，餐厅大约只有20%的产能可用，配送应用也无法为高产能食品生产搭建亚马逊式履约骨干。
公司在隐身运营期间的业务足够分散。数千名员工——包括招聘人员和销售人员——甚至无法在 LinkedIn 上说出公司名称；其厨房业务则以 Cloud Kitchens、Kitchen Valley、Namah 和 Cocinas Ocultas 等品牌，分布在30个国家。
2. 采矿自动化扩大经济上可触达的地图
Atoms 的下一台“计算机”瞄准采矿，使命是打造「更高产的矿山，为地球上的产业提供动力」。其对旧金山 Pronto 的待完成收购已「距离交割仅一步之遥」，将加入自动化采矿设备的技术。
被问及自动化能否释放更难触达的资源时，Kalanick 表示，第一步是提高现有矿山的生产率；随后，通过缩小用工规模、降低安全暴露，也减少说服人们住到「天涯海角」的必要，自动化可以让环境恶劣、监管严格或位置偏远的矿区具备开发价值。
主持人认为，稀土矿物未必真的稀缺，真正稀缺的是允许激进开采、且有人能够在那里工作的土地。Kalanick 补充称，自动化可以开放更多地点；他还提到自动化隧道施工以及 The Boring Company，认为它们可能与更深层资源开发相关。
主持人将物理AI堆栈定义为土地开发、化学、制造、计算和模型，并称 Tesla 是「这个时代的 Google」。Kalanick 参与讨论了更广泛的物理自动化问题，但没有把这一 Tesla 判断作为自己的核心论点。
3. 自动驾驶已有领跑者，但时间表仍未确定
被要求在 Waymo、Tesla 和 Uber 中做选择时，Kalanick 表示，除领先者之外，行业「叫得响，真正有咬合力的没几个」。Waymo 已经领先并提供了存在性证明，但仍面临制造、规模、紧迫感和「竞争强度」等问题。
Tesla 正在「困难模式乘以100」的条件下推进基础科学和工程。真正的不确定性在时间：成功可能明天到来，也可能要等5年，因为决定性问题是「视觉领域的ChatGPT时刻何时到来」——尤其是视觉在没有额外传感器的情况下运行。
视觉-语言-动作系统仍处于研究与落地交织的阶段，既有希望，也有不少有意思的进展。Kalanick 将人们等待的临界点，比作从 ChatGPT 3.5——已经让人觉得货真价实——跨越到 ChatGPT 4 的过程；到了4，「世界突然变了」。
Kalanick 自己优先考虑的是专用设备，而不是人形机器人。人形机器人演示往往从跳舞和格斗开始，而 Atoms 要的是「有偿工作的机器人」：比如打开纸袋但不把它撕破，或执行不需要类人灵巧度的采矿任务。
4. 语言可能成为物理AI的效率层
Kalanick 称语言是人类发展出的「史诗级压缩」机制，而人类运行所需功率约为100瓦。相比之下，他表示 Waymo 机器开车消耗的能量约为人的100倍，说明机器感知距离生物效率仍有多远。
汽车目前会吸收大量无关信息——天空中云的样子对驾驶员并不重要，但机器必须学会这一点。Kalanick 设想代理之间可以传递类似语言的信息：一个系统负责驾驶，另一个系统高效标记少数真正值得关注的危险。
主持人用松鼠作类比，将问题拓展到软件之外：生物会把微小效率层层叠加，而机器人则会在电机、执行器和材料环节层层累积损耗。Kalanick 认同材料科学及相关硬件研究值得大量投入，因为一旦软件开始奏效，操控硬件也必须跟上。
5. 奥斯汀的优势在于允许建设
Kalanick 表示自己在12月搬到得州。Michael Dell 说，自2021年以来他一直在 Lake Austin 拥有住所，每年约有15个周末在此度过。主持人形容奥斯汀以一半价格提供两倍空间，当地社区的主流情绪是：「我们正在建设未来。」
离开加州对 Kalanick 仍然是一场哀悼。他依然怀念旧金山，以及在那里创办 Uber 的经历，但他指出，许多自行车道和公交专用道经常闲置，而修建一英里这样的车道可能花费4亿美元，这在他看来体现了「下意识地要把城市掐死」。
他的政治判断不止于税收：「真相与正义是社会的免疫系统。」当这套免疫系统遭到压制，社会问题就会扩散；因此，在执法和制度问责已经恶化的地方，他支持通过公投倡议和其他干预手段进行修复。
Dell 将得州描述为长期低税、支持增长的环境，而不是一场新近出现的迁徙潮。他表示，如果当前增长持续，未来美国最大的10座城市中将有4座位于得州；他还提到，美国每10名儿童中就有1名出生在得州，纽约证券交易所公司数量超过纽约州及其他任何地方，以及德州大学是创业人才的重要源头。
6. 只有市场结构赋予资本战略意义
Kalanick 不认为积累资本本身就是目的：「资本作为一种独立存在的战略武器，并不是一回事。」但在 Uber 时代，融资变成了世界级能力，因为 Masa 可以向竞争对手投入10亿美元，几乎立即抹去20%的市场份额。
他给出的不是预测，而是一个具体的地区性警示。Kalanick 表示，自己的中东业务原定于1月上市，但随着油价走弱，沙特股市在约两个月内下跌约20%，令局势「大幅降温」。
对于伊朗战争是否会中断海湾资本流动，他明确表示不确定：自己目前没有在融资，相关新闻也才出现约两周，他不会在战争期间打电话向投资者募资。
7. 中国制造业纵深是竞争标杆
Kalanick 对中国的印象来自多次访问：50年前的深圳感觉像「非常潮湿的堪萨斯城」，如今却像是在力压新加坡。包括 Xiaomi 在内的制造业参观，以及大量中小企业的走访，呈现出一个「到处都是能打、野心勃勃创业者」的生态。
他没有把这种进步定义为敌人的失败。中国建设者充满饥渴，正在「凶猛地追求真相和进步」，因此美国的回应应是「提升我们的水平」，同时保持一场「像成年人一样」进行的友好竞争。
更好的关系也会恢复现实中的人员流动。Kalanick 提到，一位长期在中国工作的员工有美国籍配偶，他希望将对方带到美国工作；他认为移民事务完全可以妥善处理，而历史上「好的迁移」曾不断带来追逐自己美国梦的创新者。
8. 企业AI需要重构，而不是打一个AI勾选框
Dell 在 ChatGPT 发布前几周推出了第一台 H100 服务器。相关AI基础设施业务随后从约20亿美元增长至100亿美元、250亿美元，今年预计达到500亿美元，驱动力来自云端、主权AI和企业客户对token的需求。
Dell 表示，Dell AI Factory 产品已经覆盖4,000多家企业。在 Dell 自身业务中，他看到了大量可以带来20%或更高投资回报，或生产率与效率提升的场景，但「不是按一下按钮就能获得20%」。
机制是彻底重构业务：先定义目标结果，再标准化流程、整合工具和数据，最后自上而下应用AI。孤岛不会「自发地自我改善」。Kalanick 补充称，当一个人的奖金取决于不把事情搞砸时，艰难的变革尤其令人不适。
Dell 曾警告员工，设想中的竞争对手原本会在5年后出现，如今却只剩2年；它会更快、更便宜、更具创新力。Dell 不得不让自己的公司成为这样的竞争对手。他表示，基础设施业务上季度增长73%，下一季度预计增长约100%。
9. AI原生公司压缩时间，推理向外扩散
Dell 表示，一个季度如今相当于过去的一年，转型速度或许比早期变化快5倍。「技术采用的障碍不是技术，而是文化、领导力和勇气。」他估计，真正摸清这套方法的大型企业只有10%-15%。
Dell 称，Stripe 的 Collison 兄弟认为，2025年成立的公司，增长速度约为2018年成立公司的4倍。传统企业仍然拥有品牌、资产负债表和客户关系，但这些都是「价值会到期的资产」，除非它们足够快地抵达另一边。
Dell 不接受未来只是更少的人做同样的工作。更好的工具应该带来「多得多的事物」：更多科学发现、更好的医疗和教育、新发明，以及解决过去无法解决的问题的新方案。
因此，基础设施将是混合式的。「成本最低的token」会在靠近数据的地方生成——手机、PC、工业设备或嵌入式系统上；而当账单真正到来时，公有云的经济性会变得不那么有吸引力。Dell 表示，公司已有10,000家客户将其产品嵌入自己的产品中。
10. 更好的模型和自主代理提高控制要求
Dell 强调 Google 的 Gemma 模型、OpenAI 的开放模型、NVIDIA Nemotron、Hugging Face 和 OpenClaw，认为它们体现了蓬勃发展的本地模型生态。现场观众中约20%表示已经搭建了 OpenClaw。
在企业内部，这些代理需要身份认证，需要验证它们是谁、正在做什么，也需要配套的控制机制。
对AGI的诚实回答，Dell 是「我确实不知道」。他转而提到 Gemini 3.1、Opus 4.6 和 OpenAI 5.4，并表示团队如今只需1天或2周，就能完成过去需要几个月或9个月的工作。
主持人指出，私有「暗数据」是另一个前沿：组织可以在保留数据私密性的同时，用专有历史数据重新训练标准模型，从而建立只使用公开数据的竞争对手无法获得的优势。
11. Invest America 将所有权变成社会契约干预
Michael 和 Susan Dell 宣布，为居住在家庭收入中位数不超过15万美元邮编区域内、年龄2-10岁的2,500万名儿童各提供250美元，合计捐赠62.5亿美元。Dell 将账户体系视为一个直接分配平台，用于承接其基金会长期关注的城市贫困工作。
Gerstner 表示，所有18岁以下儿童都可以申领账户；从2027年1月1日起在美国出生的每个孩子都会自动获得账户，并将其附在社会安全卡上。账户结构永久有效，而1,000美元的注资必须每4年重新授权。
Michael Dell 表示，每天有超过100,000名儿童注册。Gerstner 随后称，已有450万名儿童申领账户，并按当时的增长速度预计到7月4日将接近1,000万名。
这项计划的野心，是打造一场由慈善家、各州、父母、企业、IPO股票捐赠和普通人共同资助的运动。Gerstner 预计每年会新增370万名新生儿账户；Michael Dell 估计，15年内可能有5万亿美元流入家庭。
该应用会把每个孩子持有的标普500指数基金拆解为 NVIDIA、Walmart 和 Dell 等熟悉的公司，家庭则通过日常仪式为账户添加小额赠款。Gerstner 的行为逻辑是：看得见的复利会告诉孩子，「我在场内，我有机会」。
I don't know if some of you knew that I was an angel investor in some companies. On the count of 3, what's my favorite angel investment of all time? 1, 2, 3? Thank you. Give it up for Travis Kalanick.
I appreciate you.
All right. Wow. On a big news day, Travis is here—on a very big news day. You spent, wow, I guess, 7 years just in the lab building. Every year I ask you, “Hey, do you want to come to the All-In Summit?” He said, “Nah, I’m just going to chill. I’m building.” The next year, I said, “Hey, you know, I’m always available to you.” He said, “You don’t understand. I’m stealth.”
He’s stealth. I’m stealth. Nobody knows where I am or what I’m doing. The employees aren’t allowed to put the name of the company on their LinkedIn—thousands of employees weren’t allowed to put the company name on LinkedIn. Incredible.
And then he’s like, “By the way, Jason, you can invest. You can’t announce it, and you have to sign an NDA. You can’t mention that you’re an investor.” So I’m like, “Okay, no problem. I’m just happy to be on the cap table.” Is he secretly saying what he wasn’t supposed to say? Now he’s talking about the company. What just happened? You did it. Well, you came out—Chamath, you’re out now. Okay, let’s go. You’re out. It’s out. You came out of stealth today. That’s so funny. Great, you came out of stealth. You came to the All-In Summit last year. Is that fair? Are you saying you’re coming out of stealth today? Is that right?
Their parents thought they worked for the CIA.
It’s kind of fun, too, right? It was like, “What is this? Why is there this massive density of stealth startup people in Los Angeles? What is happening over there?”
Well, look, let’s start with what that meant for our employees. Imagine if you’re at a multithousand-person company and every single employee has “stealth” on their LinkedIn, including salespeople and recruiters. They were living life on hard mode.
Also, technically, the name of the company in different countries was very generic. Everything was designed to be stealth.
Right.
We operate in 30 countries. In the U.S., the Kitchens product is known as Cloud Kitchens. In Korea, it’s Kitchen Valley. In the Middle East, it’s Namah. In parts of Latin America, it’s Cocinas Ocultas. I mean, you get the idea.
You can’t ever remember all the names or all the code words.
Yeah.
China. It’s all over the place. But things have gone really well, and you’ve been a little inquisitive. Tell us about the branding that you’re announcing today, and then maybe some of the acquisitions and the evolution of the company. You’re not just renting kitchen space.
Those who know how I thought about things in the Uber days know that a lot of this stuff isn’t surprising. I would often talk about digitizing the physical world. I think I even did it at the Summit.
The quick version of this—I’ll try to do it quickly—is that we know the bits world, the computer world, the one that Michael Dell essentially invented for us: CPU, storage, and network. These are the 3 core computing resources. When you go to computer science class on your first day, you learn about the 3 core computer resources. The CPU manipulates the bits, storage stores the bits, and the network moves bits from point A to point B.
But if you’re digitizing the physical world, you’re treating atoms like bits. You’re building an atoms-based computer, and I’ll explain what I mean. I know this is a little out there. The CPU manipulates bits—what manipulates atoms? Manufacturing. Storage stores bits—what stores atoms? Real estate. The network moves bits from point A to point B—what moves atoms? Transportation or logistics. So you have these 3 core computing resources in an atoms-based computer.
The name of my company was very obtuse and purposely designed to be as boring as hell. It was called City Storage Systems. That’s digitized real estate in an atoms-based computer, our first computer being a food computer. What does that mean? Manufacturing, real estate, and logistics for food. The mission was infrastructure for better food.
The idea was: can you get a meal that’s prepared and delivered to you so efficiently that it starts to approach the cost of going to a grocery store? If you can do that, you do to the kitchen what Uber did to the car. But in the Uber days, the roads were there. The cars were unused. You just had to put an app in the App Store. It wasn’t that easy, but it was kind of that easy.
In this world, you can’t do this with a restaurant. A restaurant doesn’t have the capacity. When I left Uber, 13% of all San Francisco miles were Uber miles, and that was 9 years ago. You can’t get there with food and restaurants. They have about 20% capacity. Uber Eats and DoorDash fill it, but the infrastructure to do high-capacity, high-scale, industrial production just isn’t there. The logistics just aren’t there. It doesn’t work. That’s why, in e-commerce, you go through Amazon’s big-ass warehouses with awesome logistics. You’ve got to do the same thing when food goes to e-commerce.
That was a lot. Okay, so the bottom line is, it’s awesome.
We do this food computer stuff. We’re doing more computers now, and the name of the company is Atoms. Let’s say the mission is physical automation to transform industries and move the world.
We have the food computer we talked about. Then we’re doing mining—mining minerals.
We’re talking about Atoms, guys.
Yeah. Of course, you do some data mining, too, but the point is physical mining. We’re automating mines. The mission there is more productive mines to power Earth’s industries.
On the transport side, it’s Wheelbase for robots. If you’re doing specialized robots—not humanoids, specialized robots—you need to be able to move and act in the physical world. The minute you’re moving, you have to have a wheelbase. It’s just part of the equation.
A lot of people look at Tesla and say, “It’s great.” Look at Waymo—awesome. They’re cruising around Austin, of course. But there are so many things that move. It’s not just a ridesharing thing, including mining equipment that’s doing its thing.
That’s the general idea, and we acquired a company in the mining space called Pronto. Or we’re about to close it—we’re inches from closing, is the way to put it.
What were they doing? What was their business, Pronto?
Automating mining equipment.
Where are they based?
They’re based in San Francisco.
You and I were starting to talk about this backstage, but there are some folks I talked to in the mining industry who mentioned that the big issue with mining, number 1, is surveying—finding the locations, right? Is there an advantage to be created there? I know there are a couple of startups trying to be really smart about selecting locations to get the targets out of the ground.
Yeah.
The other one is, can you go deep? Pretty much anywhere on Earth, you can get whatever you want if you’re willing to go deep enough, but the cost is distance squared, right? So the energy cost is how deep you’re going, to the second power. It becomes geometrically more expensive to go deeper, but the deeper you go, the more you’re able to not worry about getting the right location. Does automation unlock that capacity?
Automation definitely does. Also, does The Boring Company have some good stuff going? I hope we’re doing the mining thing and Boring makes some good tunnels for cars to do their thing, but there’s some kind of boring mechanism—automated tunneling—to do some of this.
But, to be honest, they have this thing called rare earths. I don’t know why they put it in the plural. Rare earth—isn’t it? Rare Earth? I don’t know.
Yeah, but it’s not rare. It’s a very common mineral, guys. It’s not rare. It’s what you have to do to the land that’s aggressive. What’s rare is where they’ll let you do it and where you can also get people to go.
When you automate, you can go to a lot of places. First, all the existing mines become way more productive. Second, you can justify going to places you wouldn’t have been able to go before because you don’t have as much of a labor footprint, a safety issue, or a whole bunch of other things. So if it’s inhospitable, regulated, or somewhere people don’t want to live—if it’s the end of the Earth—
You can send robots and have people monitoring them remotely.
Yeah.
This is a future that feels a little bit like science fiction. Look, we’re here in Austin. You’ve got to do the shout-out to Tesla and all the things, because I like to break down the physical AI stack. It includes not just computation and physical AI models, but all the things you think of. What about land development? That should be in the stack. What about chemistry? That needs to be in the stack. Manufacturing needs to be in the stack. When you look at the stack, you’re like, “Damn, Tesla’s got this.”
They are the Google of this era. What I mean by that is, in the 2000s, if you were doing a startup, the first question you would get was, “Why isn’t Google going to kill you? Why isn’t Google just going to do it?” Before that, it was Microsoft—the late ’90s. Uber had that time in the 2010s.
Yeah, but if Uber puts that in the app—
Come on. It’s like, dude, this is Uber. But I think in the physical AI space, that’s sort of a Tesla thing. There are so many things to do. You’ve got to shoot your shot. You’ve got to do this stuff.
There are rumors that you might not be done with self-driving, something that you were very early on. How do you think about what you’re seeing in the playing field of self-driving? Waymo’s making great progress, and Tesla’s making great progress.
Pick a winner between Tesla, Waymo, and Uber. Uber seems to be building a network of stuff. I mean, pick a winner.
The number of players in this space is crazy now, right?
Yeah, look, I think there’s more noise—there’s more bark than there is bite right now. I think Waymo obviously is ahead. The existence proof is there. Their issue is manufacturing and scale, urgency, and fierceness. Let’s win. Let’s go.
Uber had an autonomy project back in the day, and they have a different strategy these days. I haven’t been there for a while. But the point is, you’ve got Waymo, then you’ve got Tesla. Fundamentals. Science. Hard mode times 100. The question is, do they get there, and on what timescale?
Honestly, everybody’s like, “Could happen tomorrow. Could happen in 5 years.” I think the question is: When does the ChatGPT moment happen for vision? Let’s call it vision without other sensors. It’s super inspiring, but what’s the timeline on it?
Those are the basics. Then there are a lot of other little guys that don’t really have the stuff yet, I believe. There’s nobody standing out just yet among the others.
Do you think we’re at a point now, as you’re getting into more of these autonomous systems that move around, where we have these vision-language-action models tuned and ready for prime time?
There’s been a conversation about who’s going to have the Android—the operating system for vision-language-action—where I can use my voice, tell it to do something, and it knows what I’m saying. Then it identifies the objects and does the thing in the physical world. Do those models exist today, or are they still a work in progress? Is that like a Google OS, or where does that OS come from?
Look, I think this is an area of a lot of energy, a mix of research and implementation. I think there’s a lot of hope and interesting stuff. At a high level, we all remember what happened when you used ChatGPT 3.5 and you were like, “Holy—yeah, it’s legit. Whoa.” Then it went to 4, and you were like, “Okay, some stuff just changed. The world just changed.”
I can sort of connect some dots, and it’s getting real. Is it about to happen? Is it about to happen for physical AI? That’s what this is about.
The fun part about it is that machine learning, deep learning, this kind of thing, for many years—decades—was inscrutable. You didn’t know what the thing was thinking. It just spit out an answer, and you knew it was correct.
Now you can have a conversation with it. Imagine if it’s driving your car and there are different agents. One is just driving, and the other is like, “Yo, look out over there.” It’s like, “Oh.” It’s just like how we roll. Somebody does that, and you’re like, “Honey, that’s 200 meters away. We’re going to be okay.”
Yeah. Jason and I don’t call each other “honey,” but I got you. Like, “Sweetie.” You know.
Okay. So, anyways, that was odd, wasn’t it?
A little strange.
Okay, okay. I didn’t mean it that way.
It does that. I didn’t mean it.
Yeah, yeah. You know, I meant it.
Language is a beautiful compression mechanism. Humans use 100 watts of energy. When you put that in the scheme of things—AI training, AI energy, the power plants that are built to do the thing—it isn’t even at human strength yet, okay?
The Waymo machine takes 100 times more energy to drive a Waymo than a human does to drive a Waymo. There are still things that humans are great at, and at those things, we’re unbeatable. We’re the GOAT at certain things. Language is this epic compression.
We need to find ways to compress because, when you think about how we first started looking at the physical world, we saw everything. You know what, guys? This is sort of obvious: It doesn’t matter what the cloud is doing if I’m driving. But the car doesn’t know that. It’s pulling in every freaking data point and processing everything.
They’ve been carving out the things that don’t matter and things like this, but there are ultra-awesome versions of this. You can imagine how you can use language, or things that look like language, to communicate either among agents or safety systems with a driving system, to get very efficient answers and identify safety issues very efficiently.
People don’t know that you moved to Texas—or most people don’t know, but it’s out there. You moved here in December, so now you’re a resident of Austin.
Yeah, I was—thank you. It’s very exciting for me. We’ve been getting to play some backgammon.
Backgammon, cards—it’s cards. We’re having a good time. I’ve had a place on Lake Austin since 2021, and I go there. I’m an avid water skier. You’re impressive at water skiing, I have to say. I’ve had a place in Austin for 5 years. I freaking love it. It’s my weekend place; I would go 15 weekends a year.
What do you think’s going to happen in California? It’s pretty messed up.
Look, I grew up in Cali. I grew up in Los Angeles. My parents were born and bred in Los Angeles, which basically makes them the founders of LA, okay? So I have a lot of heart—my whole family, everything. It’s pretty heartbreaking.
I don’t want to get the violin out, but it’s heartbreaking. It’s just the place you grew up. It’s your home, you know, when you have to leave. But it’s getting weird out there, and it feels like it’s getting weirder. At some point, it’s just too weird. It’s too weird.
Do you think everyone’s going to leave? It started with Elon, and it was like—
Yeah, he was in the right.
“We don’t want Elon here,” and then he was like, “Message received.”
Like, cool. Right. Then it kind of worked its way down the tech industry, and then the world of people building businesses and whatnot. Now it’s gotten so broad in terms of the group of people—
Comedy, music, New Yorkers, restaurateurs.
I mean, this place is much— I’m not even talking about Austin. I’m just talking about everyone leaving LA—or, sorry, leaving California. It’s almost like working down the path of—look, the rest of my team is like, “When are we moving?” You know? They’re like—
And how are you dealing with that? That was the question.
You’ve got to buy a home on the lake. There are literally dozens of startup CEOs of successful or growing companies that I talk to who are like, “Dude, I want to leave, but I’ve got employees here. I’ve got an office here. I’ve got a facility here. I build stuff here. How am I going to leave?”
Yeah, I totally get it. It’s a real thing.
So, look, I think, like most things, when it’s time and it feels painful to do something, sometimes it’s actually not as bad as you think. You just have to make the move, lead, and do it. That’s kind of the process—a mourning process—I went through. That’s just what it is.
And you’re setting up a team here?
Yeah, of course. I got that office right on the lake.
Did you get that?
It’s—we are negotiating.
No, it’s all good. No, no, no, it’s all good. We’re negotiating right now.
Okay. But I’m going to jet ski to work.
No, literally, we were driving up there last year, and I was like, “Wow, I wonder who owns that?” He’s like, “I will.” And I was like, “Did you look at it?” He’s like, “I looked at that.” And I was like, “That would be a nice one.”
The truth is, I had a couple of people move here a couple of years ago, and they all had the same reaction: “Oh my God, I’m living in a place that’s twice as big for half as much. The people here are dope. The food is dope. Everybody here has this sense that we’re building the future, and it’s just fun and we’re all positive.”
For me, I got to live in New York, LA, and San Francisco. I did 3 of the great cities in this country. This one feels the most like home to me, which is a very strange feeling, but it feels like everybody here wants to build the future. It’s very diverse—all these different industries and people pursuing stuff. I think this is the future.
Yeah, here’s the thing. You go to San Francisco, and I still have a little nostalgia when I go to San Francisco, just having built Uber there and the whole thing.
I still get the butterflies, but it does have something magical. You just can't take it away. And then you look at all of these bike lanes and these bus lanes that never have a bus or a bike in them and cost $400 million to build 1 mile. It's literally sort of like this subconscious desire to choke the city off.
Now, remember, I look at things through roads. That's how I think. So I'm just like, obviously, the city is totally busted.
Yeah, no, but they literally took Market Street and they're like, “What would be the optimal way to fuck this up and virtue-signal at the same time?” And they're like, “Yeah, buses.” And it's like, “Nobody's on the bus. Nobody takes the bus. It's a beautiful small town. San Francisco's Austin.” That whole street is empty. It's painted red.
So we all complain nonstop in our group chat. “When are you leaving San Francisco? When are you leaving?” I'm number one, I get it. But Friedberg, when are you leaving?
Okay, well, so—
Sacks, you're the best by far. Okay, so let me just say: a couple of glasses of wine in, public-facing Friedberg is like, “You know, I think there's a better way to do this.” And then there's Darth Friedberg in the group chat. He's like, “These people, these morons are morons. They're destroying society.” He's like Darth Friedberg in the group chat. Am I lying? Am I lying? Is he the most—
Correct. Especially after a couple of glasses of wine.
When I start drinking, he takes a photo of a cliff. And he takes pictures. He's like, “Fourth beverage.” And we're like, “Oh, it's worth staying up on the group chat.”
And then I'm like, “I'll go and attack this congressman on Twitter,” which I realized—
Yes, and then you have to delete the tweet.
Yeah, I delete the tweets.
There's a group of people trying to raise $500 million to create a tech-business coalition to go to Sacramento, which arguably is something that everyone's left and avoided doing forever because no one wants to spend time in freaking Sacramento fighting politicians. But it's almost like we're all falling off a cliff; it's time to do something. Do you think there's a realistic path back? Do you think the people can actually get their act together? Even if $500 million came in, is there a way to kind of turn around the state and fix some of the policies? Do you think it's too late?
I think that—look, anybody who's doing anything to fix things, I'm like, “Hell yeah, let's do something.” The issue is we all grew up in the tech world, which was like a libertarian place where you stay out of politics. It was that kind of vibe. Everybody was like that: “Leave me alone. I want to make stuff.” I just—I don't do that. And that's obviously not a thing anymore.
In California, I think the ballot initiatives are very powerful, and there are very clean ways to get something on the ballot. I love that. I think that your DAs who have decided, “We do not enforce crime at all anymore,” that's a sweet spot.
I believe—I sort of have this aphorism—that truth and justice are the immune system for society. When the immune system is suppressed, all the social ills flare up. So look for the places where truth and justice are being deteriorated, where they're being degraded, and say, “How do we get at that?” Because if you get at that, everything else downstream will be better.
That's kind of how I look at things and how I also determine whether the world's getting better or worse. When I say “weird,” I'm talking about truth and justice. That's what I mean when I say, “Oh, man, it's getting weird. It's getting weirder.” It means it's weird. I'm just talking about truth and justice.
Well, I mean, you look at the homeless industrial complex, and you look at Chesa Boudin. The All-In pod—Sacks and myself—we literally led the recall of him. And then you have the same thing going on in L.A. with Gascón. Yeah, Gascón. I mean, we basically lost the script. You're running the city for the criminals. It literally is like a Batman movie. It's like Bane.
Nobody's going to arrest the criminals. Look, I was born in the darkness. I mean, these guys are lunatics.
I know police officers in Los Angeles who are no longer police officers. These are lifelong guys who protect and serve; that's in their blood or DNA. They want to protect people. They want the bad guys to be dealt with, and they almost have PTSD from what it's like to want to serve, see bad things happening, and not be allowed to stop it.
Nobody's got their back, and they're not allowed to do their job. It's crazy, and it's getting weird.
Okay, hey, I want to just go back to AI for a moment.
Sorry for the darkness.
No, I think it's good.
I was trying to induce—I'm trying to induce dark Friedberg.
You brought it up. I mean, someone bring me a tequila; I'll get going. Yeah, let's do it. Can we get a couple of tequilas?
It was funny. I went on this podcast yesterday, and the first hour was middle of the road. I was talking about tech and science, and then politics came up. He's like, “So, socialism?” And he said, “You lost it.” The energy went 10x. So that'll come out in a couple of weeks, but I was like, it got me going.
Okay, I want to talk about physical AI one more time. I saw a presentation the other day where someone showed a video of a squirrel jumping from one tree to another, and they're like, “A tenth of a watt” or something. The biology is tuned, and it's so perfect in terms of its efficiency of energy utilization to do physical things. We're taking these big things of metal and motors and actuators, and if you add up or compound all of the inefficiencies in the system, it's like 1,200 watts to get the robot to walk 4 feet.
Break apart not just the software, but the hardware layer. Where are we in evolving things like actuators, materials, and everything else that's going to make physical AI work and scale?
A lot of the questions you're asking are going down the humanoid lane, which is this thing. Everybody talks about, “How do you do the hand?” It's almost like a Terminator 2-type obsession with the hand, which is fair. It's a very critical part of it. I like to look at the Achilles—quote-unquote, the Achilles tendon—of any of these machines, and you're like, that's where the action is. There's a couple of other places.
Look, I'm in the non-humanoid space. Mechanical engineers have been dealing with actuators and all the electromechanical interactions that make machines do certain things. I'm in the food-machine space, so I can tell you how to open a paper bag and put a bowl in a paper bag without tearing the paper bag. But I'm less into proprioception—the senses that help you understand awareness and touch. I'm not in that game.
When you're mining, you're not threading a needle. You're not playing tennis. Certain things may be equivalent to tennis. The bottom line is we're seeing—obviously, all you have to do is go online and look at where the humanoids are going over time and how much better they're getting. It's wild, and it's happening so freaking fast.
But any humanoid demo starts with dancing and martial arts. We're in the specialized-robot lane, which is gainfully employed robots.
I know I didn't totally answer the question on the technology piece, but do you agree that there's probably a big opportunity for venture money and research to go into materials science?
For sure. Because the physical AI stack—manipulation and all of the related things around it—is massive. If you get the software working, it's almost like the hardware has to catch up.
I do. Well, actually, it's good that you bring this up. One of the things you pioneered at Uber was capital as a weapon, and you were very thoughtful about, “Hey, if we can take this capital off the table, then that's going to—let's call it what it is—it's going to be an advantage versus the competitors,” and these other competitors couldn't get that capital.
Now, I think people have seen that playbook and they're like, “Someone was like, ‘That was smart. Let me try.’” And it's at a different scale. Now that you've come out of stealth, now that you've got people starting to understand how big your vision is—capital as a weapon—this is, I guess, in your plan, yeah?
Well, I mean, here's the thing, right? Capital as a strategic weapon for its own sake is not a thing. When it is actually a strategic weapon, then it is a thing. What I mean by that is, in the Uber world, in the early days, if you didn't have capital, it didn't matter how good your app was because Masa was going to put $1 billion in your competitor and you were going to lose 20% market share tomorrow.
So a critical competency—in fact, one of your world-class competencies has to be raising capital. You need to do it better than everybody else, and if you don't, you are going to lose.
Let me ask one follow-up to that. Sorry, you go ahead.
I've heard theories the last couple of days that big capital seekers are kind of skittish right now because of what's going on in the Middle East with the Iran war. Dubai, Qatar, and the Saudis are kind of going to close up capital flowing to the U.S. right now. Is that real? Do you think that's a real threat?
So look, our Middle East business was supposed to go public in January, and the Saudi market went down 20% over like a 2-month period. That was a massive damper on the situation.
Now, part of that was because oil prices had gone down so dramatically. And if you went into KSA, you went to the kingdom, everybody was like, “We need oil prices to go up.” That’s the other side of the equation.
So, I don't know what happened. I'm not in the market raising money right at this moment, and this is a 2-week-old thing. I see the news just like everybody else, and I'm not out there calling while the war is going on and saying, “Hey guys, you got some money?” So, I don't know exactly what's going to happen.
Yeah, I don't know. It was shock and awe, and then, hey, now we've got a steady state. Let's hope that's what happens in Iran, that we can depose these evil dictators and replace them with something a little more stable.
And related to this, before we wrap, they're going to China, and there's a big trade deal being negotiated. What do you hope comes out of this Chinese thing? What did you learn in China, and what do you think would be great for America? What would you like to see and be like, “Man, that's going to set us all up”?
Look, here's the thing: If you go to China right now and just take a tour of the manufacturing that's going on there—the manufacturing base and the cities, especially if you've gone to China for a couple of decades in a row—you’re like, “Damn.”
So, let's just do 2 things. You go to Shenzhen, which 50 years ago felt like Kansas City, but really humid, which I guess Kansas City sometimes is. But you go there now, and it's like one-upping Singapore. So that's the city view. You're just experiencing something very awesome, and you're like, “This is advanced.” You just get the vibe, and it's everywhere.
Then you go and start seeing the manufacturing base, and you see what Xiaomi is doing—or any of the others. There are so many scrappy guys, badass guys everywhere, and you're like, “Fuck, they're hungry.” Does anybody remember the 2008 Olympics in Beijing? Does anybody remember the opening ceremony? You're like, “These motherfuckers are taking over.” At least that's what they want to do. That shit's happening.
So, I don't have any issue—or this is not negativity for me. I'm like, “These guys are killing it. The best idea is winning.” They're fiercely going after truth and progress, and they're making it happen. Let's step up our game, okay?
But we can also have a friendly game. We don't have to be like the Detroit Pistons in the ’90s, you know?
Yes, we can. There’s a way.
Yeah, yeah, you know. There's a way to do this right, and there's a way to do it like adults. I hope that's where we would end up.
I have an employee who—we were, for a long time, the largest cloud-kitchen builders in China. I have an employee in China who has an American wife. They both live in China. They're both from China originally. But one of them—it would be great for him to work here on some things I'm doing. It's very hard to make that happen right now.
Now, that's selfish. Maybe selfish. I'm like, there's a person I've been working with for over a decade. I'd love to continue here. Maybe there are other bigger-picture items that I'm not dealing with. I'm not the geopolitical guy, but I'd love for there to be good relations. If you have a significant other who is an American citizen, do we have to make that hard? That's an example.
Some normalcy would be something, you know. I'm just saying.
Now, I agree. There are ways to do immigration properly. We effed it up super bad. Don't even get me started. But there's also good migration, too. A lot of great innovators all over the place came from other places for their own version of the American dream. God bless. Free bird.
That doesn't have to be a negative thing. So I'd like to see more of that. And, yeah, China's wild. Let's keep our eye on the ball, and let's give them a run for their money, too.
All right. Well done, brother. Thanks, brother. That's good. Good to see you, brother. Wow—Michael Dell.
Born in Houston.
Well, I missed the opening. We jumped to the music, but you started Dell Computer here in Austin with $1,000 42 years ago.
It was in my dorm room at Dobie at UT, about 10 days before I finished my freshman semester. And it's been working out pretty good. Yeah, there have been some bumps in the road, but it's generally worked out okay. We'll have about $140 billion in revenue this year, so, yeah, it's okay.
Hey, it compounds over time, doesn't it?
Yeah, yeah. You start small and just keep adding, and there you go. That's how it goes. It's just that easy.
But why Texas? I think this is an important thing. We're in Austin. Jason lives here. David Sacks lives here now. More people are moving from California to Austin. Why Austin? Why Texas? Why does it work here? And is it getting better, or has it just always worked?
I think Texas has had a low-tax, pro-growth, and pro-business environment for a long time. If you look at the growth of the Texas economy relative to the rest of the United States without Texas, Texas just kind of looks like a better version of the U.S. economy.
Now you've got Austin just about in the top 10 cities in the United States. When that happens, you'll have 4 of the 10 largest cities in America in Texas. 1 out of 10 children born in the United States are born in Texas. There are more New York Stock Exchange companies in Texas than in New York or anywhere else.
And you've got the University of Texas here in Austin, which I would always think of as kind of the wellspring for a lot of the companies that are here, certainly ours. There's a long history of innovative, pioneering spirit and entrepreneurship, and it's been a fantastic place for us.
Part of this, I think, Friedberg and Michael, is what's happened in the other great cities—or what were once great cities. My hometown, New York, got to spend 10 years in L.A., and in the last 12 in the Bay Area. What's happening there is incredibly un-American, and they're decelerating when compared to here. I think maybe the gap, the disparity between these 2 locations, has gotten greater, yeah?
You're seeing a lot more people say, “Life here in Austin seems a lot better than the life I'm living in New York, L.A., or the Bay Area.”
Yeah, well, I've got a lot of new friends and neighbors that have come. Certainly, if you look at the migration statistics, Texas has attracted an enormous number of people.
And look, when you compare the environment here to the other situations that are going on, it's very attractive. But it's kind of been great for a long time, so it's not really new news to those of us who have been here a while.
Elon had a great experience when he was building the Gigafactory over here. They let you do stuff here.
Yeah, basically. They let him build it, which he said was an incredible experience for him, because in California they didn't let him build these factories. In fact, the Tesla factory in Fremont was just an old, ancient factory that he was able to retrofit.
There's something going on here as well with the data centers, and that's actually very close to what you're working on at Dell. Maybe you could talk a little bit about the data-center boom that's going on in Texas that people maybe aren't paying attention to.
Sure. Well, there's obviously been an enormous build-out of AI infrastructure, and that requires lots of new data centers and lots of power. Texas has an enormous advantage there relative to other states: a lot of power, a lot of land, and you can build stuff, right?
There's been a massive build-out, particularly in some of the cities and towns in West Texas where there's not a lot of population. They're not too opposed to having data centers out in the middle of nowhere, where there's land and power.
The demand for tokens is enormous. We've been building these AI data centers not just here in Texas, but around the world, and the growth in that has been tremendous.
We introduced the first H100 server literally a couple weeks before ChatGPT was announced. The progression of our business in that area has gone from $2 billion to $10 billion to $25 billion. This year it'll be like $50 billion, so tremendous growth.
When you think about what these models are creating, there's this phase change that's happened in computing, right? We had 60 years of calculating and computing.
Now we have machines that are thinking and helping us think. The demand for that kind of intelligence is growing, and the models are amazing, but they're also the worst they'll ever be, and they're continuing to improve. We see a lot more demand than supply, and it's happening not just in the hyperscalers and cloud service providers; it's happening in 4,000 enterprises where we're building these Dell AI factories. It's happening in sovereign AI, like Palantir, where people want to protect their data but also use AI on it. They want to bring the AI to where their data is.
When this kind of thing started a few years ago, we had some really sophisticated large companies—think of Fortune 100 companies—and they started buying these AI servers from us. They kind of knew what they were doing, right? They were building their own models, taking open source models and running them. Some of them were algorithmic traders or using derivatives of machine learning. Of course, they needed a lot of help in doing that because it was sort of a complicated thing.
About 2 years ago, we put together this product that we called the Dell AI Factory. Now we've got 4,000-plus of these, and it's kind of running rampant across enterprises.
How do you think about the payback time on the investment that's being made? The administration put in place this accelerated depreciation rule for the company.
Very helpful, actually. Yeah.
Just for folks to understand that a little bit, if you spend $100 billion this year building out data centers and buying infrastructure for those data centers, you get to write off 100% of that this year, correct?
Correct. So you don't pay taxes. You pay way fewer taxes. It's in place for 10 years, I think.
That's a 10-year deal, so it's accelerating the investment. How much is that helping versus how are you seeing folks rationalize the investment relative to the return they're going to make, and over what time scale? This is still the big question: Is the money really there? The hyperscalers, maybe they're starting to come up, but end usage, end states—are we kind of, "Hey, wait and see, we don't know yet," or are folks getting 20% ROIC starting in year 1 after they've made the investment?
I can tell you in our business, in our company, we definitely see plenty of use cases where the ROI or the improvement in productivity and efficiency is 20% or greater.
The ROI right away gets there?
Yeah. I mean, it's not like you just hit a button and you get 20%, right? There's work required in thinking through the processes, and it's worth describing that a little bit. When you have any company, its processes, tools, and technology are a function of what was available at the time it created those things. So what you have to do is step back and say, "All right, what's the trajectory of the improvement of the tools? What outcome are we trying to create?" Now let's simplify and standardize the processes, get all the tools together, get all the data together, and then apply the technology.
This really has to be done in kind of a top-down way. You can't sort of do it spontaneously in silos; they're not going to spontaneously improve themselves. Often, that means you're completely changing the way the organization works. It's like a wholesale re-architecture. It's a reimagining of the way a company works.
The way I described this to our team about 3 years ago was, "We're going to have a new competitor 5 years from now that would be 2 years from now, that was in every business that we're in, except they were going to be faster and more innovative and more successful and lower cost, and they were going to put us out of business. The only way we're going to prevent that is we're going to become that company, and here's how we're going to do it." It excited some people, it scared some people, but I actually believe that's what's going to happen. We've been dramatically changing our business.
I would say the biggest benefit by far is speed. We're much faster at being able to apply innovations. You look at our infrastructure business: Last quarter, it grew 73%, which is kind of unusual for a business of this size. This quarter, we guided that it would grow even faster, like 100%.
You've lived through a couple of paradigm shifts here. The PC revolution, obviously—you led that—and then, of course, client-server, the network revolution, online, the internet, mobile.
Mobile.
Yeah. So each one of those, we saw massive disruption. We were talking in the green room about how we used to have a typing pool, and there was a mail room. All these things got abstracted away by the PC and networked PC revolution, but it took a decade or 2. This one's happening a lot faster, yeah?
Yeah, this one, I think, is like a quarter is like a year. Maybe it's 5 times faster or something like that. But back to your question, I would say maybe 10% or 15% of large companies have really figured this out, and the rest of them are kind of fumbling around. There's a tendency when you hear about a new technology to say, "Oh, let's just go do it and show the boss, 'Hey, we did AI.'"
The board said, "We've got to do AI. We've got to do AI, guys. We need AI. Are you proud of me, boss?" And, "Look at what I made." Exactly. I also think an important point about this is that the barrier to technology adoption is not technology. It's culture, leadership, and courage, right? It's willingness to change and to adapt yourself.
If you're in a business that you don't think is changing very much, hard change is really hard, right? You have to—it can be very uncomfortable. You're like, "Well, we're going to stop doing that. Well, maybe we don't need this anymore." Particularly if your bonus is dependent on not messing things up.
But let's use the internet as an analogy, which you saw up close. There were businesses that were successful in the internet transition. They made the transition—maybe Macy's.com versus Sears Roebuck, right? Maybe Macy's did a better job of taking advantage of the internet than Sears. But then there were internet-native businesses that seemed to blow them all out. Amazon's a good example, or CSN Stores—whatever they'd be, Wayfair, et cetera.
What's the right way to think about this evolution in industries generally? Are we going to have businesses that are going to transition successfully and those that aren't, and they're going to die? Are we really going to see AI-native businesses in every industry come in and just disrupt everything?
I believe we will. Certainly, when you talk to the Collison brothers at Stripe, they'll tell you that the rate of growth of the 2025 cohort companies is about 4 times faster than the 2018 companies. Every year, the new batch of companies is growing faster and faster because they're starting with all these new tools.
Well, because they see all the new companies on their platform.
Exactly. When you think about an incumbent company that already exists, it has brands, balance sheets, customer relationships, and whatever other assets, right? Those are expiring-value assets. If it doesn't change quickly and get on the other side of this, I think it will go out of business—which is exactly the speech I gave to our team 3 years ago.
You have to be bold, and you've got to go make those changes not only to survive this but to thrive. I think about it as: How do we prepare our company to be ready for the 2030s?
Right. Isn't it much more—it's kind of the storyline—that there's more to do than there ever was? When the internet arrived, Sears didn't just need to sell locally; they could sell to the world.
Well, sure. I mean, this is the point.
AI lets everyone do everything.
When we have better tools, we can do way more things, right? When I hear people say, "Maybe we're just going to have all these great tools and we won't do more things. We'll just do the same things with fewer people," it doesn't sound right to me. There'll be some of that, but I think most of it will be that we're just going to do a whole lot more things.
We're going to solve a lot more problems. We're going to accelerate scientific discovery. We're going to invent all sorts of new things. We're going to solve all sorts of problems that haven't been solved. That's super exciting.
What do we have wrong on infrastructure? The original build cycle looked a lot like everything's in a data center. Everything's got to sit there. That's where all the intelligence—it'll all be in these hosted, proprietary cloud models. Do you think it's open source? Is it distributed on the edge? Where does the intelligence—where does the inference sit? And how does that really change or re-architect the industry, do you think?
It's really all of the above. It's not like there's one answer. Certainly, if you go to any industrial company or natural resources company, advanced manufacturing, retail, or logistics, there's tons of inference at the edge.
And that's growing very, very fast, and we make a lot of that embedded equipment. Certainly, telcos are doing that too. I mean, it's pretty much every industry. Think about wherever data is being created: you want the AI infrastructure and the inference close to the data.
There has been this sort of rebalancing as companies have figured out that everybody loves the public cloud, right? Until they get the bill. When they get the bill, they're like, “Wait, this is supposed to save us money, but it costs quite a bit more.” So the lowest-cost token is going to be the one that's generated right where the data is, on the device. You're going to have tokens being generated on your phone, on your PC, in every embedded piece of equipment.
And look, we have an interesting perspective on this business because we have 10,000 customers where they embed our product in their product. Think medical devices, security, all sorts of things in hospitals and industrial plants. Any kind of data-driven activity requires some kind of computing, network, and storage infrastructure.
So, when you look at the desktop where you started, it's coming full circle, and this must be at least very interesting or intriguing to you: You see this OpenClaw movement, everybody trying to buy the most powerful desktop they can, and all these hobbyists who were your customers, calling you up and ordering their bespoke PC from Dell. Now they're on Dell.com.
Dell.com. What did I say? Dell.com. Yeah, you said ordering from Dell, calling us up. They order online usually. They order online now, yes. They have this thing called the internet, Jason.
Yes, it works out pretty well. But this is incredible, that they're stacking computers and running local models. I was just thinking back to how much the first couple of computers I owned cost: $4,000 in 1980. Then the prices came down. You could buy a Dell for $500 or $800, really nice laptops for that price.
Do you think there's a world where we're going to start to see the desktop come back because people want to protect that data? They want to protect the skills they're building. They don't want to give it to Sam Altman in a cloud somewhere. They don't want to give it to Google, whoever it happens to be. The desktop revolution comes back, and everybody's got a $10,000 desktop. Is that coming?
I don't know if everyone will have a $10,000 desktop, but that would be great. We have this Dell portal on Hugging Face, and we have all these open models. We qualified them on every kind of machine we have. There's been enormous progress in the open-source models.
Google has these Gemma models, G-E-M-M-A, and they work really, really well on small machines. OpenAI has its open-source models. You've got NVIDIA Nemotron models. There's an enormous ecosystem of open source that's thriving, and certainly OpenClaw will be part of some good discussion about that.
How many people have set up OpenClaw? Raise your hand. Oh, my lord, that's about, what, 20% of the audience?
Yeah, autonomous agents—big deal. Certainly inside companies, there are going to be a lot more autonomous agents. There are significant security requirements that need to go with that. We need to be able to authenticate and validate who these agents are and what they're doing, and have the right controls and that sort of thing.
And your take on AGI, and when we're going to hit it. Do you actually think about superintelligence and AGI, and the 2 sets of problems they could solve there? Do you have a personal definition that you like to use for those when you're talking internally with your team about how things are moving?
I don't really know, Jason. With the latest releases—we were talking about this backstage, the Gemini 3.1, the Opus 4.6, and the OpenAI 5.4—it feels like we've sort of hit some kind of threshold where the quality of the models is just tremendous.
When I listen to what our teams are able to accomplish in a day or 2 weeks, which would have taken them a few months or 9 months' time, it's just amazing, the speed of innovation.
And so it seems to be continuing. We get all the reinforcement learning, and there's also tons of private, dark data that these models haven't been applied to. That's sort of what's happening with these—I think autoresearch is the key. With autoresearch, you have the capacity to take a standard model, retrain it on your private data, keep it private, and build an advantage for your organization based on the history of your data that no one else has. That seems to be what a lot of folks who have the capacity are thinking about.
But if you were to start a company today that was not in computing and build a business from the ground up, how would you architect your people and your organizational principles as an AI-first company, knowing what you know about computing and where things are headed? Are you hiring people? Are you hiring a bunch of people to run a bunch of agents? How do you think about architecting a new business today?
It's a great question. I don't really spend a lot of time thinking about that. I'm thinking about how to run our company. That's hard enough.
Yeah. Everyone I talk to, that's the question. Everyone goes to these off-sites and says, “I'm actually doing this with my management team on Monday. We're doing a teardown: Hey, how would we build the business differently today?”
Yeah, I mean, what we've been thinking a lot about is this reimagining question.
Yeah.
You know, sort of, all right: We know the trajectory of the tools. What are the tools going to be in 2027, 2028, and 2029? And how do we accelerate our path to that?
How worried are you about social issues? AI recently ranked as the most unfavorable term on a list of terms including “president—”
Saw that.
—somewhere between ISIS and the Democrats.
ICE, ISIS, and the Democrats.
ICE was better than AI. People liked ICE, the masked agents, more than they liked AI.
Yeah. Well, I think part of the problem is that it's been sold as something that presents itself like a human would.
Yeah, right? Maybe if we called it “linear algebra matrix calculations”—
Instead.
Right. “Matrix multiplier.”
That would be more friendly. I don't know. But do you think we're going to have—
No, I think you're right. The positioning is wrong, and we're not communicating to people: “Hey, this could help healthcare. This could make you live longer. This could help your kids get educated more. This could help with housing costs. This could help with food costs.”
Messaging aside, how much do you actually worry about disruption or dislocation in employment, about acceleration of earnings for some people and deceleration for other people in society who feel left behind? That starts to fuel more of the social concerns and politicians saying, “Hey, we have to stop building all the data centers.” How much do you really worry about that?
I tend to be more optimistic. In all technology cycles, you get these network effects, and that's kind of inevitable. But I also think we're going to do more with the tools. You do have this acceleration of all sorts of great things. Education can dramatically improve—scientific discovery, healthcare, energy. All sorts of unsolved problems can be accelerated. Ultimately, I think it's an amplification of human potential and capability.
And extending the frontier, too. By the way, we should also remember that, beyond some of the advanced semiconductors in the big data centers, we're talking about software. Right? It's like software that runs on your computer.
[Snorts] So, somebody says, “Well, we don't want that.” How do you stop total software? I mean—
Yeah, I know. How are you going to—are you going to stop someone from putting an open-source model on their computer at home and asking it for medical advice? In New York, a law was just passed saying AI models can no longer give medical advice.
Yeah, right? It's being proposed.
Yes, it's being proposed. You can't give legal and health advice. We're anti-software. It's like, yeah, well, we're also anti-my books and advice. If you were going to look it up in a book—were you dropping into your Bernie Sanders right there?
That was my Bernie Sanders.
Michael Dell, do you have a problem with the 1% of the 1% that you're enabling with your data centers? Why are you doing this to the people of our great nation while you give your money to children in the Invest America accounts? This is a good one you're doing. Can we talk about Invest America?
I think Bernie might have even criticized that, but, you know—
Well, that's the problem. The billionaires are giving our children money, and they're not asking us permission. Then those kids are going to buy things that their parents never asked for.
Well, they don’t actually get the money until they’re 18 years old. So, that’s—
But what gives you the right to give our children an education? What is this philanthropy? It makes no sense.
No, I mean, honestly, I see my great friend Brad Gerstner here. Brad’s here? There he is. Brad, come up for this little segment. Sit for a second. Let’s talk about Invest America. We have 5 minutes left here.
Everybody in the fifth row, give him a round. I didn’t know he was going to be here. Here you go. All right. How did this go down, Michael?
I heard about this idea in 2021 from Brad, and I thought, you know, that’s a great idea. That’s an awesome idea. I think there were some discussions with the prior administration, but they didn’t do anything about that, unfortunately. And here we are—a miracle. The Invest America Act was passed, and now we have thousands of companies joining in and matching the government’s contribution. Susan and I made a big announcement giving $250 to 25 million children.
I mean, Michael, let’s just talk for a second here. What do you think, Bernie? Do you approve? I’m going to get out of this place. I just want to pause on this because it is one of the greatest philanthropic gifts in the history of humanity. People have just kind of glossed over it because there are a lot of big numbers in the world, but we’re talking about you personally—you and Susan sat down and said, “We’re going to give a number.” Was that number $5 billion, $6 billion, $7 billion?
Well, it’s $250 to 25 million children ages 2 to 10 in ZIP codes where the median income is $150,000 or less. It’s $6.25 billion.
I just want to say something.
But they have to sign up to claim the accounts.
Yes.
They have accounts, but they have to sign up to claim the accounts. I think we’re getting 100,000-plus kids signing up every day.
Yeah. First, thanks for having me up. What a national hero and national asset my friend Michael Dell is, but he understates this because I’ve been working on this for 4 years. We’ve been talking about it on the All-In pod, and we had a lot of momentum.
Behind the scenes, Trump gets elected. It’s April, and we’re in the middle of the tariff strife—April 25. We realize there’s only going to be one piece of legislation that gets passed during Trump’s first 2 years. It’ll be this big, beautiful bill, the reconciliation bill. I call up Michael and say, “Michael, we’ve got to go. We’ve got 5 days. It’s drafted in the Senate. We have bipartisan support, but we have a window. I have to get in the Oval Office. We have to get in the Oval Office.”
Michael said, “What should the text say?” You and I had a conversation, and I sent the text to DJT. Yes, I’m not talking out of school. Listen, Biden—wherever you sit on the political divide, I will say this: Trump seeks out ideas from business leaders. He has deep respect for business leaders like Michael Dell. Wherever your politics are, that’s just the truth. The last administration didn’t. If it was Hillary Clinton, she may have done the same thing.
And I just have to say this: Invest America is not a red idea or a blue idea. It’s a red, white, and blue idea.
So, to the prior conversation, when Michael and I first talked about it, it was the right thing to do. We have to reconnect the 70% of people who feel left out and left behind to the American dream. But this is in our self-interest. This is about defending the ownership society and capitalism that, for 250 years, created the greatest experiment in the history of the world, but that’s at risk. Less than half of people under the age of 40 have a favorable view of capitalism.
Michael understood both sides of it. It’s the right thing to do, and it’s the right thing for the country. At any rate, Michael Dell, tremendous American. I have one punch-up: the name. Invest America, Trump accounts. What do you think? Were you considering this in the context of other philanthropy? How do you put this together in the spectrum of how you think about giving back?
Yeah, great question. We have a foundation that’s very focused on children in urban poverty. That’s basically the central focus of the foundation, although folks in Central Texas would know that we do a few other things here in our local community.
When I heard about this idea, one of my thoughts was, “Wow, this is like a platform for directly giving to the people that we’re targeting.” We actually thought about doing it just in Texas first. Things have gone pretty well with the company and all that, so we thought we’d just go bigger.
And what happens, Brad, if 10 more Michael Dells show up?
There are a lot of Michael Dells, let’s be honest, but there are a number of folks who could make an equal-size or even greater gift. There are 1,000 people who could make a gift of significance. What if this actually becomes a movement and we change the dial?
I think it actually is becoming a movement instead of a moment, and we’ve got a lot of that queued up, Brad.
Wait, have you called anybody, Michael? Have you called people?
Michael and I chair the Invest America Giving Committee, and we’re ambitious guys, so we’re knocking on doors. We’ve had a few conversations. Yeah. And so, there was a question earlier. First, it’s really important for you guys to spread the word: Every child under the age of 18 is eligible to claim their account. Number one.
Number two, you’ve heard this, like, “Oh, kids born between 2025 and 2028.” No. This is forevermore. The legislation creates this account forevermore. Every child born in America starting January 1, 2027, will automatically get a Trump account at birth, stapled to their Social Security card.
The $1,000 has to be reauthorized every 4 years, but the accounts don’t. Every kid—this is Social Security 2.0. This is the biggest change to the social contract in America in 50 years. 3.7 million kids a year will get an account that can compound as a 401(k) from birth. We’re going to have a lot of announcements, but it’s not just billionaires. It’s going to be companies donating stock on their IPOs into these accounts, wealthy people, states, moms and dads, corporations, and normal people.
The estimate is that, over 15 years, we can move $5 trillion into the pockets of families that would otherwise have had zero. $5 trillion.
Right. And to me, the leadership that Michael showed—not only in helping me get the meeting that ultimately got this passed into law—really matters. It takes people. Those moments either happen or they don’t happen. If they don’t happen, there’s no law, and this doesn’t change kids’ lives.
By the way, 2 things on this. If this $5 trillion moved through government programs, it would get incinerated.
Exactly. That’s what we see happen. There are just a million crony structures that take it away and destroy it. So, giving it directly into the accounts is the answer.
The second thing is, it makes a lot of sense. You guys can—I’ll be the lead—but can we replace Social Security in this country with a defined benefit or defined contribution like this? Eventually, everyone could have a Trump Account, or whatever you call it, and we wouldn’t have to have this fake Ponzi scheme that we call Social Security.
We can do it.
Well, they have a defined benefit program, but I’m saying everyone has an account and they all own a piece of their future. Every time you get a payroll tax deduction, instead of it getting eviscerated, destroyed, and vaporized, that money actually goes into an account. You buy a piece of a company, and maybe you can direct it.
Friedberg’s getting angry.
On July 4th of this year, you’re getting me wound up. So, for all the—there are 4.5 million kids who’ve claimed their account. Almost 150,000 a day will have, on the trajectory we’re on, 10 million by July 4th, the 250th anniversary of the country.
Every one of those kids’ accounts—the parents and the kids—will have an app on their phone on July 4th that looks a lot like a Robinhood app. They’ll see that they own something. It’ll say, “You’ve received your $1,000 or your $250,” and it will show a little bit of NVIDIA, a little bit of Walmart, and a little bit of Dell.
We decompose the S&P 500, which they own, into its constituent parts so they can get excited about being an owner in the upside of America. When moms and dads double-click Apple Pay and send $5 or $10 into the account, when they send their QR codes to their friends on their birthday and their friends add to the account, or on Christmas or at bar mitzvahs, and companies add to the accounts, they see all of it growing. It unlocks human potential.
It's not just the money. It's that I'm in the game. I have a shot. To David's point, I think the biggest crime of Social Security is—and we made this very clear—Social Security is a sacred promise.
We refused, and many people tried to get us to take on the broader struggle, but we didn't do it because we knew it would kill this program. But let's be clear about this: Our government requires all of us to give 10% of what we earn into Social Security, right? It was the social contract's evolution in the Industrial Revolution that kept the country together. The only problem is that it goes into a black hole. Nobody sees it. Nobody knows what's there. But it is your savings.
Now, imagine if that same money was required to—you know, the government took it away—but it was in an account with your name on it. You could see it grow. You know exactly what was there. You could get excited and say, "Hey, I'm going to add a little bit more to that." Right? And you had a little bit of choice.
That, to me, is the possibility, and I think we will end up there. And, Brad, thank you for—
Yeah, let's give it up for Brad Gerstner.
Finally! Brad also adopted his home state of Indiana. We have Ray and Barbara Dalio, who adopted their home state of Connecticut, and many more to come. Look, it's going to be super easy for anybody to add 100 kids in your neighborhood, adopt a ZIP code, adopt a school district, or adopt a town.
It's going to be amazing. Give it up for one of the great entrepreneurs of our time and an incredible philanthropist.