1 00:00:00,240 --> 00:00:05,597 Huge congratulations to Pawan, Jacob, the teams at Microsoft and Nvidia. Um, I 2 00:00:05,680 --> 00:00:08,797 will say they were kind enough to send me a Surface Laptop Ultra for the last 3 00:00:08,880 --> 00:00:11,037 few days. It is amazing that I won the 4 00:00:11,120 --> 00:00:14,157 workstation for Christmas. Now, now 5 00:00:14,240 --> 00:00:15,997 there's a bunch of stuff about the 6 00:00:16,080 --> 00:00:18,957 exciting future to get into. But before 7 00:00:19,040 --> 00:00:21,517 we do that, I was looking both of you up 8 00:00:21,600 --> 00:00:23,757 last night on YouTube and it's quite 9 00:00:23,840 --> 00:00:26,157 rare to see both of these gentlemen on 10 00:00:26,240 --> 00:00:28,717 stage together. And I was thinking about 11 00:00:28,800 --> 00:00:31,357 how, you know, you both had a history. 12 00:00:31,440 --> 00:00:35,037 These two great companies had a history going back decades. You both 13 00:00:35,120 --> 00:00:38,797 individually had a history of working with each other going back decades. Ties 14 00:00:38,880 --> 00:00:41,117 to me. My first job was at Microsoft. My 15 00:00:41,200 --> 00:00:43,357 first paycheck a GeForce FX to play 16 00:00:43,440 --> 00:00:47,437 Halfife 2. So, so maybe to start us off 17 00:00:47,520 --> 00:00:50,877 in Jensen, what is your favorite memory 18 00:00:50,960 --> 00:00:52,797 uh or fun memory of working with Satia 19 00:00:52,880 --> 00:00:56,317 and Microsoft? Well, uh, Nvidia was 20 00:00:56,400 --> 00:00:59,517 founded because of Windows. It was 1992, 21 00:00:59,600 --> 00:01:02,957 late '92, uh, Windows 3.1. And we were 22 00:01:03,040 --> 00:01:06,877 just started to see the the vision of what a what a personal computer really 23 00:01:06,960 --> 00:01:09,437 can be. And at the time, frankly, we had 24 00:01:09,520 --> 00:01:14,157 never I'd never even seen a PC before. I largely worked on workstations. We had, 25 00:01:14,240 --> 00:01:17,837 you know, we had uh file servers and supercomputers that we used to design 26 00:01:17,920 --> 00:01:20,397 chips with. and and the founders and I, 27 00:01:20,480 --> 00:01:23,037 you know, talked about this future where 28 00:01:23,120 --> 00:01:25,677 maybe uh a personal computer with 3D 29 00:01:25,760 --> 00:01:27,997 graphics would enable it to do things 30 00:01:28,080 --> 00:01:31,277 like games and and be a workstation and 31 00:01:31,360 --> 00:01:35,837 design things. And so we we imagine about this world. If not for Windows, uh 32 00:01:35,920 --> 00:01:37,357 Nvidia wouldn't have been founded. And 33 00:01:37,440 --> 00:01:39,917 then and of course 34 00:01:40,000 --> 00:01:43,837 and of course we we started in 1993. Um 35 00:01:43,920 --> 00:01:48,477 uh Windows 95 really made reinvented 36 00:01:48,560 --> 00:01:53,037 Windows 95. >> Wow. 37 00:01:53,120 --> 00:01:55,037 >> Am I on 38 00:01:55,120 --> 00:01:58,477 >> Windows Windows 95 39 00:01:58,560 --> 00:02:01,117 uh made possible the GPU to be connected 40 00:02:01,200 --> 00:02:06,397 to uh uh a personal computer. And what 41 00:02:06,480 --> 00:02:09,597 revolutionized the GPU was DirectX. 42 00:02:09,680 --> 00:02:13,357 Direct 3D uh was the API that really uh 43 00:02:13,440 --> 00:02:16,877 enabled us to re uh reveal the 44 00:02:16,960 --> 00:02:19,997 capabilities of GPUs into applications. 45 00:02:20,080 --> 00:02:22,397 But then DX8 came along and we 46 00:02:22,480 --> 00:02:24,957 co-invented with Microsoft the world's 47 00:02:25,040 --> 00:02:27,437 first G programmable GPU which is 48 00:02:27,520 --> 00:02:29,997 programmable shaders which a few steps 49 00:02:30,080 --> 00:02:33,917 later evolved into uh CUDA as we know 50 00:02:34,000 --> 00:02:36,397 it. And so if you look at the entire 51 00:02:36,480 --> 00:02:41,197 journey of our company, Windows was at the core of it. And if not for Windows, 52 00:02:41,280 --> 00:02:44,237 there would be no GeForce. If no GeForce, CUDA wouldn't have come to the 53 00:02:44,320 --> 00:02:47,677 world. Uh without CUDA, Alex Kushki and 54 00:02:47,760 --> 00:02:51,917 Ilia Suskver, um Yan Lun, you know, uh 55 00:02:52,000 --> 00:02:54,237 Andrew Angone of them would have uh 56 00:02:54,320 --> 00:02:58,237 found a computer uh to do deep learning. 57 00:02:58,320 --> 00:03:02,077 >> And here we are. >> Is now going to do a little transfer of 58 00:03:02,160 --> 00:03:06,080 market cap back to me. >> [laughter] 59 00:03:06,560 --> 00:03:09,197 >> I do that every day. I do that every day. 60 00:03:09,280 --> 00:03:12,877 >> But it's extraordinary that journey uh 61 00:03:12,960 --> 00:03:14,637 that Jensen recounted. I mean it's kind 62 00:03:14,720 --> 00:03:17,597 of the the it's the history of computing 63 00:03:17,680 --> 00:03:22,157 in some sense. Um, and the thing, one of 64 00:03:22,240 --> 00:03:25,437 the things, uh, that Jensen sort of 65 00:03:25,520 --> 00:03:27,837 sparked for me was 66 00:03:27,920 --> 00:03:32,157 he has been perhaps the most consistent 67 00:03:32,240 --> 00:03:35,917 with the vision of what accelerated 68 00:03:36,000 --> 00:03:38,637 computing can be. And this going back to 69 00:03:38,720 --> 00:03:40,957 the trajectory, you've been the one 70 00:03:41,040 --> 00:03:43,197 person in our industry who has had this 71 00:03:43,280 --> 00:03:47,757 one idea whose sort of trajectory has 72 00:03:47,840 --> 00:03:50,477 been the most expansive quite frankly 73 00:03:50,560 --> 00:03:53,277 more so than even any single PC form 74 00:03:53,360 --> 00:03:56,077 factor or chip architecture. Because I 75 00:03:56,160 --> 00:03:59,677 look back, one of my first memories was 76 00:03:59,760 --> 00:04:02,477 we were talking about essentially saying 77 00:04:02,560 --> 00:04:05,837 in the cloud, how big can be accelerated 78 00:04:05,920 --> 00:04:07,357 computing or high performance computing, 79 00:04:07,440 --> 00:04:10,397 right? The first deal I remember was in 80 00:04:10,480 --> 00:04:14,237 Azure, how do we make HPC a bigger 81 00:04:14,320 --> 00:04:17,277 thing? Um and by the way if we 82 00:04:17,360 --> 00:04:20,477 >> Satio was the first user of Infiniban 83 00:04:20,560 --> 00:04:25,517 inside a cloud which brought right the first supercomputer to open AI which 84 00:04:25,600 --> 00:04:27,597 boom >> that's right if we had not done that 85 00:04:27,680 --> 00:04:31,917 like the fascinating you know like life is full of accidents but if we had not 86 00:04:32,000 --> 00:04:34,477 done that for HPC 87 00:04:34,560 --> 00:04:39,197 we wouldn't have had the supercomputer to give to open AI to go train a GPT 88 00:04:39,280 --> 00:04:41,997 model and rest would have been history I mean it was kind of one of those happen 89 00:04:42,080 --> 00:04:45,117 stance that we had done the HPC see 90 00:04:45,200 --> 00:04:50,157 stuff when we were building an Infinibanbased supercomputer and and 91 00:04:50,240 --> 00:04:52,077 comes Sam and says hey do you have 92 00:04:52,160 --> 00:04:55,117 computing and we said yeah maybe and and 93 00:04:55,200 --> 00:04:57,517 so it's sort of like but the thing that 94 00:04:57,600 --> 00:05:00,397 I really give Jensen all the credit 95 00:05:00,480 --> 00:05:04,157 quite frankly is that consistency of 96 00:05:04,240 --> 00:05:08,317 vision of what this can be not today not 97 00:05:08,400 --> 00:05:11,757 tomorrow but as a secular trend and 98 00:05:11,840 --> 00:05:15,357 that's kind of what you know brings us even to this moment 99 00:05:15,440 --> 00:05:19,597 >> and then it led to a conversation. It led to a conversation I remember quite 100 00:05:19,680 --> 00:05:21,917 well where Satia and I were talking 101 00:05:22,000 --> 00:05:23,837 about so so what does this mean to the 102 00:05:23,920 --> 00:05:28,957 PC? How do we make uh the the perfect PC 103 00:05:29,040 --> 00:05:32,557 in the post era of all of this stuff and 104 00:05:32,640 --> 00:05:36,077 you know of course the personal computer is the ultimate tool. It's my ultimate 105 00:05:36,160 --> 00:05:37,597 tool and for a whole generation of 106 00:05:37,680 --> 00:05:39,997 people it's our ultimate tool. Um, what 107 00:05:40,080 --> 00:05:43,357 happens in the era of agents uh when 108 00:05:43,440 --> 00:05:46,157 when when the agent is on your computer, 109 00:05:46,240 --> 00:05:50,797 it is now your personal assistant, but it wants to have access to all of the 110 00:05:50,880 --> 00:05:52,957 amazing tools that we have, but the 111 00:05:53,040 --> 00:05:55,437 amazing tools are spread out in the 112 00:05:55,520 --> 00:05:59,357 Nvidia ecosystem into uh gaming PCs. 113 00:05:59,440 --> 00:06:02,877 >> Yeah. >> Into workstations. And that's where most 114 00:06:02,960 --> 00:06:05,037 of the tools were, right? all the all 115 00:06:05,120 --> 00:06:07,037 the all the do tools and all the 116 00:06:07,120 --> 00:06:09,917 seammen's tools and uh autodesk and 117 00:06:10,000 --> 00:06:14,717 adobe and you saw many of them just now and you know we want the world's best 118 00:06:14,800 --> 00:06:18,157 blender the world's best unreal the world's best omniverse the world's best 119 00:06:18,240 --> 00:06:22,317 kia everything all sitting on the same PC 120 00:06:22,400 --> 00:06:27,677 >> um as the AI >> and so how do we go create that by the way took four years 121 00:06:27,760 --> 00:06:32,637 >> wow [laughter] >> that was a four year ago conversation and just incredible 122 00:06:32,720 --> 00:06:37,117 >> now that's very good segue today and Satia so I think the last few months you 123 00:06:37,200 --> 00:06:41,357 know we've been sort of inundated with these stories of agents in security you 124 00:06:41,440 --> 00:06:42,797 know we all use agents I'm sure 125 00:06:42,880 --> 00:06:45,837 everybody here is using amuse or uh 126 00:06:45,920 --> 00:06:48,397 instinct and you are giving full disk 127 00:06:48,480 --> 00:06:50,157 access and you're having it act as you 128 00:06:50,240 --> 00:06:53,117 and today when you guys showed MXC it 129 00:06:53,200 --> 00:06:58,077 seemed to me like a fundamental new operating system primitive for agents 130 00:06:58,160 --> 00:06:59,997 which is a break from where we have been 131 00:07:00,080 --> 00:07:03,597 so maybe talk to us about why the OS 132 00:07:03,680 --> 00:07:05,597 is the right place for this and what 133 00:07:05,680 --> 00:07:07,917 this primitive if you will opens up. 134 00:07:08,000 --> 00:07:10,957 >> Yeah. No, first of all remember this all 135 00:07:11,040 --> 00:07:14,477 started with coding agents all running 136 00:07:14,560 --> 00:07:16,557 on the desktop. I mean in fact if you 137 00:07:16,640 --> 00:07:19,677 sort of look at uh where is the agent 138 00:07:19,760 --> 00:07:23,357 that needs this security primitive the 139 00:07:23,440 --> 00:07:25,997 most. It's the local harness and its 140 00:07:26,080 --> 00:07:30,557 access to the file system and to be able to take action. And so that's why in 141 00:07:30,640 --> 00:07:32,637 fact you know working with Jensen and 142 00:07:32,720 --> 00:07:35,757 team we said look how do we wire you 143 00:07:35,840 --> 00:07:37,997 know wire this in into the primitives 144 00:07:38,080 --> 00:07:41,197 whether it's in the cloud or in uh uh in 145 00:07:41,280 --> 00:07:43,357 the operating system locally and given 146 00:07:43,440 --> 00:07:45,837 the fact that essentially coding went 147 00:07:45,920 --> 00:07:48,237 there first where the harness is local 148 00:07:48,320 --> 00:07:50,477 even if you were using a cloud model we 149 00:07:50,560 --> 00:07:53,117 needed to make the desktop uh the most 150 00:07:53,200 --> 00:07:55,517 secure place for agents to execute and 151 00:07:55,600 --> 00:07:58,237 so that's where MXC comes from it's 152 00:07:58,320 --> 00:08:03,197 fantastic FIC to see open shell now fully integrated and native with MXC the 153 00:08:03,280 --> 00:08:07,437 controls that Pin talked about right which is containment itself has to be 154 00:08:07,520 --> 00:08:11,677 done at every level right it's not about sort of doing one thing and leaving the 155 00:08:11,760 --> 00:08:15,277 other right it has to be at the microVM it has to be at the WM it has to be at 156 00:08:15,360 --> 00:08:19,677 the session the piece that I'm most excited about is the Windows session 157 00:08:19,760 --> 00:08:23,917 having containment right it's so so critical then okay containment is 158 00:08:24,000 --> 00:08:27,117 wonderful but then I need full observability how do you get 159 00:08:27,200 --> 00:08:31,677 observability you get it If you have an identity for an agent that I can track I 160 00:08:31,760 --> 00:08:33,997 can then see the entire exhaust of the 161 00:08:34,080 --> 00:08:36,877 agent right all of it uh and then I can 162 00:08:36,960 --> 00:08:39,037 govern it that means I can like classic 163 00:08:39,120 --> 00:08:42,557 set policy and have it govern that is 164 00:08:42,640 --> 00:08:44,317 now what's going to give peace of mind 165 00:08:44,400 --> 00:08:47,837 whether it's to IT ops to sec ops uh and 166 00:08:47,920 --> 00:08:51,917 by the way and the other thing that we should also not forget is the ability to 167 00:08:52,000 --> 00:08:55,997 track the token spend right and where it's happening and so that's the 168 00:08:56,080 --> 00:09:00,157 entirety of what we have tried to bring together. It's not about any one thing. 169 00:09:00,240 --> 00:09:02,157 It's about stitching all these things. 170 00:09:02,240 --> 00:09:06,077 Having seen all these adoption cycles, whether it's the consumer adoption 171 00:09:06,160 --> 00:09:08,877 cycle, the enterprise adoption cycle, I 172 00:09:08,960 --> 00:09:11,037 would say [clears throat] the thing that 173 00:09:11,120 --> 00:09:13,997 blocks broad diffusion is not having 174 00:09:14,080 --> 00:09:18,237 that complete thought on what is required. Right? So if you do one and 175 00:09:18,320 --> 00:09:22,477 leave the other, there's no trust and what have you. So I feel very very good 176 00:09:22,560 --> 00:09:24,717 that we have brought it together in the 177 00:09:24,800 --> 00:09:27,837 cloud in the in the client across it 178 00:09:27,920 --> 00:09:32,397 sack and fin and it's heterogeneous right it's not about any one piece of 179 00:09:32,480 --> 00:09:34,797 Microsoft technology it's about supporting it all 180 00:09:34,880 --> 00:09:37,037 >> what Satia just said what Satia just 181 00:09:37,120 --> 00:09:40,637 said uh is going to be the foundation of 182 00:09:40,720 --> 00:09:45,517 the next generation of it as we know it yeah if you go the words he used what he 183 00:09:45,600 --> 00:09:49,117 described the architecture deeply profound for the next generation of 184 00:09:49,200 --> 00:09:50,957 computers just as Windows Windows and 185 00:09:51,040 --> 00:09:53,597 and DirectX revolutionized how 186 00:09:53,680 --> 00:09:57,357 applications were built. MXC 187 00:09:57,440 --> 00:10:01,037 is going to revolutionize how agents are 188 00:10:01,120 --> 00:10:04,717 built and deployed >> without it 189 00:10:04,800 --> 00:10:09,837 started without it complete non-starter and so this big big deal now this is I 190 00:10:09,920 --> 00:10:13,117 think closely tied to the next piece um for both of you which is hybrid 191 00:10:13,200 --> 00:10:15,517 intelligence now this feels a little 192 00:10:15,600 --> 00:10:17,117 different I think everybody here has 193 00:10:17,200 --> 00:10:20,477 uses large model in the cloud you use it 194 00:10:20,560 --> 00:10:24,157 it runs a tool maybe using blender and you get back a response we're all used 195 00:10:24,240 --> 00:10:27,677 to that's the In a few years, this feels different 196 00:10:27,760 --> 00:10:32,557 where you have a choice of models, some running locally, some in the cloud, and 197 00:10:32,640 --> 00:10:36,317 you're going back and forth. And one sentence from the presenter who spoke at 198 00:10:36,400 --> 00:10:39,437 GitHub, which actually stood out to me was you're getting these tokens for 199 00:10:39,520 --> 00:10:42,157 free, right? And so maybe talk to us 200 00:10:42,240 --> 00:10:46,877 about what this opens up because this feels again something new where you are 201 00:10:46,960 --> 00:10:51,117 running part of the workload here in the locally on your on your laptop, part of 202 00:10:51,200 --> 00:10:54,157 it in the cloud and intelligently routing between it. What does this open up? 203 00:10:54,240 --> 00:10:57,597 >> Yeah. for first of all, you know, I 204 00:10:57,680 --> 00:11:01,597 think this is one of those moments where 205 00:11:01,680 --> 00:11:05,437 um some of the things that you've been 206 00:11:05,520 --> 00:11:08,157 using all click together, right? In 207 00:11:08,240 --> 00:11:09,517 fact, it's kind of I was thinking about 208 00:11:09,600 --> 00:11:14,157 this in the late 80s when I was in grad 209 00:11:14,240 --> 00:11:18,157 school, I kind of had a a a terminal, 210 00:11:18,240 --> 00:11:23,437 right? My VT0000 and a VAX. Um and um 211 00:11:23,520 --> 00:11:27,037 and then I had a PC at home. >> Mhm. 212 00:11:27,120 --> 00:11:29,517 >> Um and I kind of was using it all. In 213 00:11:29,600 --> 00:11:32,237 fact, it's interesting that on my PC, I 214 00:11:32,320 --> 00:11:34,797 had um essentially a modem with a 215 00:11:34,880 --> 00:11:37,117 terminal connecting to my wax and then I 216 00:11:37,200 --> 00:11:39,917 had all my local DOSs and uh you know 217 00:11:40,000 --> 00:11:43,597 early Windows and what have you and then 218 00:11:43,680 --> 00:11:45,277 suddenly it took all and by the way I 219 00:11:45,360 --> 00:11:47,917 had Usenet all of that stuff but except 220 00:11:48,000 --> 00:11:51,277 it took all the way maybe until 95 and 221 00:11:51,360 --> 00:11:53,677 even Mosaic right the first browser to 222 00:11:53,760 --> 00:11:56,957 bring it all together as one seamless 223 00:11:57,040 --> 00:12:00,397 experience right it's not about local 224 00:12:00,480 --> 00:12:03,517 cloud, what have you. I just want to use 225 00:12:03,600 --> 00:12:06,957 my computing power wherever it is in 226 00:12:07,040 --> 00:12:09,357 sort of a seamless flow of what I'm 227 00:12:09,440 --> 00:12:14,557 trying to get done. So when I was seeing even what Danielle was showing that demo 228 00:12:14,640 --> 00:12:18,157 is mind-blowing. I mean think about that moment. I think we're going to come back 229 00:12:18,240 --> 00:12:20,397 years from now and say, "Oh my god, that 230 00:12:20,480 --> 00:12:21,997 was when we saw 231 00:12:22,080 --> 00:12:25,357 >> what he did was he went to Comfy UI. He 232 00:12:25,440 --> 00:12:27,517 brought down a bunch of models, built a 233 00:12:27,600 --> 00:12:30,957 bunch of assets, took it to Blender, uh 234 00:12:31,040 --> 00:12:33,837 used that to even enhance those assets, 235 00:12:33,920 --> 00:12:38,477 then used again brought it back to Comfy. Oh, went to Photoshop, used a 236 00:12:38,560 --> 00:12:40,557 local model perhaps to sort of enhance 237 00:12:40,640 --> 00:12:44,157 it, took it back to Comfy UI, then took 238 00:12:44,240 --> 00:12:46,397 it all to the cloud with Astra, with 239 00:12:46,480 --> 00:12:50,557 Unreal, but he just used apps on his 240 00:12:50,640 --> 00:12:53,117 desktop, models, local, models in the 241 00:12:53,200 --> 00:12:57,437 cloud. It all just work. That is what 242 00:12:57,520 --> 00:13:01,517 this is about. Like that's kind of where today was about multiple fantastic 243 00:13:01,600 --> 00:13:04,237 devices, all these things. But the 244 00:13:04,320 --> 00:13:09,597 trajectory for me is so clear. There will never be a moment where we will go 245 00:13:09,680 --> 00:13:11,197 and look and say, "Oh, this runs 246 00:13:11,280 --> 00:13:13,117 locally, this runs in the cloud." You 247 00:13:13,200 --> 00:13:16,877 will expect this hybrid intelligence to 248 00:13:16,960 --> 00:13:21,197 be everywhere and pervasive. And I think that that's what to me is pretty 249 00:13:21,280 --> 00:13:25,277 exciting. Yeah, that one application what you just said Blender was 250 00:13:25,360 --> 00:13:27,997 accelerated, Comfy UI was accelerated, 251 00:13:28,080 --> 00:13:32,317 Astro was in the cloud, the agents were locally. I mean, it's just the the 252 00:13:32,400 --> 00:13:34,797 surface area surface area. That's cool. 253 00:13:34,880 --> 00:13:39,037 The surface area of computing was magnificent. 254 00:13:39,120 --> 00:13:41,437 >> Yeah, >> it is comp. It it brought together 255 00:13:41,520 --> 00:13:46,237 literally the entire might of lit every single technology company in human 256 00:13:46,320 --> 00:13:48,317 history came together for that one application. 257 00:13:48,400 --> 00:13:50,877 >> That's right. And and by the way going back to sort of the way you talk about 258 00:13:50,960 --> 00:13:52,877 accelerated computer that was I mean his 259 00:13:52,960 --> 00:13:55,357 point about why he was excited what took 260 00:13:55,440 --> 00:13:58,077 him multiple days takes him hours like 261 00:13:58,160 --> 00:14:03,037 that the fact that he was accelerated in his ambition and that to me is pretty 262 00:14:03,120 --> 00:14:05,757 cool >> which I think you know this ties closely 263 00:14:05,840 --> 00:14:09,997 to something else because these models have gotten very good at tool use but 264 00:14:10,080 --> 00:14:14,877 the tools often live locally. you know the files live locally by the way you 265 00:14:14,960 --> 00:14:16,637 know Jacob and I are still both of us 266 00:14:16,720 --> 00:14:18,477 are filing taxes at the last minute 267 00:14:18,560 --> 00:14:22,237 apparently um and and I think there is a 268 00:14:22,320 --> 00:14:23,757 aspect here about 269 00:14:23,840 --> 00:14:26,717 >> I want my data to be secure and safe and 270 00:14:26,800 --> 00:14:29,597 controlled I want the tool use to happen 271 00:14:29,680 --> 00:14:33,917 near where the data is where if you look at what Jacob showed with co-pilot like 272 00:14:34,000 --> 00:14:37,357 that it really opens up something new >> yeah I mean to me like one of the things 273 00:14:37,440 --> 00:14:41,837 right if you look at it in fact one of the when the coding agents became the 274 00:14:41,920 --> 00:14:46,317 thing. I was excited to see everybody who's in love with, you know, one the 275 00:14:46,400 --> 00:14:48,477 console uh and everybody was love with 276 00:14:48,560 --> 00:14:53,597 file systems. Again, it just shows how powerful of an abstraction of file 277 00:14:53,680 --> 00:14:57,357 system is. >> I think agents are going to be the 278 00:14:57,440 --> 00:15:00,477 largest users of file system. They need to be secure. 279 00:15:00,560 --> 00:15:02,157 >> They need to be because they're a 280 00:15:02,240 --> 00:15:05,197 heterogeneous collection space. Uh and 281 00:15:05,280 --> 00:15:07,677 what his tax example, Jacob SP, was a 282 00:15:07,760 --> 00:15:09,597 great one, right? Basically what it did 283 00:15:09,680 --> 00:15:12,877 was it organized from as he said from 284 00:15:12,960 --> 00:15:16,317 downloads from even cloud files, email 285 00:15:16,400 --> 00:15:19,037 everything into a file system container 286 00:15:19,120 --> 00:15:21,517 and then used models locally and in the 287 00:15:21,600 --> 00:15:24,477 cloud to go do stuff with it and that is 288 00:15:24,560 --> 00:15:26,637 the new computer right so it's a I use 289 00:15:26,720 --> 00:15:29,357 the computer my agents use the computer 290 00:15:29,440 --> 00:15:33,837 the agents need these primitives as well and so I think like one drive when was 291 00:15:33,920 --> 00:15:36,077 the last time I thought of one drive as 292 00:15:36,160 --> 00:15:38,237 a cloud thing or a client thing It's 293 00:15:38,320 --> 00:15:41,037 there everywhere. And so to me, that is 294 00:15:41,120 --> 00:15:43,117 what I think we're beginning to see is, 295 00:15:43,200 --> 00:15:46,077 hey, I want my primitives for my agents 296 00:15:46,160 --> 00:15:49,917 to use 24 by7. The form factor when I'm 297 00:15:50,000 --> 00:15:54,957 using it on my PC should feel like I'm using it on my PC. When I'm roaming, it 298 00:15:55,040 --> 00:15:58,477 should roam with me. >> You know what's amazing, as you know, is 299 00:15:58,560 --> 00:16:02,237 is that these agents are going to be better users of tools than us. Yes. And 300 00:16:02,320 --> 00:16:07,677 the reason for that is just we know 10 15% of the features of any application, 301 00:16:07,760 --> 00:16:09,437 but the agent knows every single feature. 302 00:16:09,520 --> 00:16:11,997 >> That's right. >> And so so the the the fact that the 303 00:16:12,080 --> 00:16:15,117 tools, these incredible tools are sitting on your PC, they're they're 304 00:16:15,200 --> 00:16:17,197 accelerated in in this incredible 305 00:16:17,280 --> 00:16:22,317 computer Surface Ultra. Uh now the agent has access to all of that capability. 306 00:16:22,400 --> 00:16:25,037 Who knows what we're going to be able to do? I mean, just look at what it did. It 307 00:16:25,120 --> 00:16:29,997 was the Blender, the Unreal, the Comfy AI was incredible. 308 00:16:30,080 --> 00:16:33,837 >> Could show of hands. How many people here have tried using one of these apps 309 00:16:33,920 --> 00:16:38,237 locally with a model in the last couple of months or few months? Great. Right. 310 00:16:38,320 --> 00:16:41,517 And so I think this just kind of show goes to show where you can actually 311 00:16:41,600 --> 00:16:45,437 bring the tool use locally. Now Jensen, I want to ask you about the hardware 312 00:16:45,520 --> 00:16:47,677 piece of this. Right. We very often see 313 00:16:47,760 --> 00:16:50,957 you talking about these big massive you 314 00:16:51,040 --> 00:16:54,957 know Verra Rubins Blackwells NVL72s 315 00:16:55,040 --> 00:16:56,957 you know these you know intelligence 316 00:16:57,040 --> 00:17:02,077 token factories um you know that's you know that power data centers here today 317 00:17:02,160 --> 00:17:05,597 we're talking about something different right like we're talking about the spark 318 00:17:05,680 --> 00:17:09,997 the workstation which I really want for Christmas you know which runs in your 319 00:17:10,080 --> 00:17:14,477 office talk to me a little bit about how you see this evolution about tokens 320 00:17:14,560 --> 00:17:18,957 being produced like right next to you as opposed to you know Ver Rub in data in 321 00:17:19,040 --> 00:17:22,957 the data center. >> Well uh the number of engineers that 322 00:17:23,040 --> 00:17:25,757 worked on this project is incredible 323 00:17:25,840 --> 00:17:28,957 over 4,000 engineering years. This is a 324 00:17:29,040 --> 00:17:33,277 massive project. I mean look what we've done. We've reinvented the personal 325 00:17:33,360 --> 00:17:37,117 computer completely. For the first time we took all the capabilities what used 326 00:17:37,200 --> 00:17:42,317 to be in a gaming PC, in a laptop uh in a workstation in a supercomput in the 327 00:17:42,400 --> 00:17:44,077 cloud. We unified the architecture and 328 00:17:44,160 --> 00:17:47,197 we made it run on one super chip. This 329 00:17:47,280 --> 00:17:49,437 super chip, you know, it's rectangular 330 00:17:49,520 --> 00:17:51,517 because it's two giant chips fused 331 00:17:51,600 --> 00:17:55,357 together. And so this it's a beast. It 332 00:17:55,440 --> 00:17:57,037 has one pedlops. Just to put in 333 00:17:57,120 --> 00:18:00,397 perspective, the computer that started 334 00:18:00,480 --> 00:18:03,757 the AI revolution called DGX1 was one 335 00:18:03,840 --> 00:18:07,597 pedal >> and it was a quarter of a it was a 336 00:18:07,680 --> 00:18:10,237 quarter of a billion dollars and um it 337 00:18:10,320 --> 00:18:13,517 weighed 500 lb and now it's in this in 338 00:18:13,600 --> 00:18:17,357 your laptop. Um, but what's what's amazing is this is also the only 339 00:18:17,440 --> 00:18:19,997 computer, the only computer in the world 340 00:18:20,080 --> 00:18:22,797 that runs DirectX and all of its 341 00:18:22,880 --> 00:18:27,197 applications, OpenGL and every single version of it because we need all the 342 00:18:27,280 --> 00:18:30,477 tools and design tools, right? Creative tools. We need everything to run. We 343 00:18:30,560 --> 00:18:34,877 need Unreal to run incredibly well. So, DirectX has to be amazing and it runs 344 00:18:34,960 --> 00:18:39,197 every single CUDA application incredibly. what what uh Satia and I 345 00:18:39,280 --> 00:18:42,397 were talking about four years ago is, you know, we're going to have a whole 346 00:18:42,480 --> 00:18:45,917 new generation of software developers and these software developers are going 347 00:18:46,000 --> 00:18:49,997 to be AI software developers. They're either going to be creating AI 348 00:18:50,080 --> 00:18:54,557 applications and AI models or they're going to be using AI to to develop their 349 00:18:54,640 --> 00:18:58,557 software. And so the computer has to be revolutionized and has to be really good 350 00:18:58,640 --> 00:19:01,197 at CUDA from the ground up. And so this 351 00:19:01,280 --> 00:19:03,917 is the only computer on the planet that 352 00:19:04,000 --> 00:19:07,197 runs everything natively on the device. 353 00:19:07,280 --> 00:19:11,757 the first time ever. And so it took thousands of us to do this, but it also 354 00:19:11,840 --> 00:19:16,557 took the entire industry. Look behind us. Every single laptop maker, every 355 00:19:16,640 --> 00:19:20,237 single do desktop maker, every single workstation maker joined us. 356 00:19:20,320 --> 00:19:24,637 >> Yeah. Wow. >> Yeah. >> This is the first time I think first 357 00:19:24,720 --> 00:19:28,237 time I think a new piece of hardware, new computer architecture have come 358 00:19:28,320 --> 00:19:30,877 together and the entire industry top to 359 00:19:30,960 --> 00:19:35,517 bottom, left to right, everybody's involved, every single geography. And so 360 00:19:35,600 --> 00:19:38,797 this is a complete reinvention. We understand the seriousness of the work 361 00:19:38,880 --> 00:19:40,717 that we're doing doing here. We've now 362 00:19:40,800 --> 00:19:43,837 test tested some over 363 00:19:43,920 --> 00:19:46,317 1,200 applications. 364 00:19:46,400 --> 00:19:48,317 >> The most intense applications in the 365 00:19:48,400 --> 00:19:50,877 world. All fully tested, performance 366 00:19:50,960 --> 00:19:53,597 tested, compatibility tested, everything 367 00:19:53,680 --> 00:19:58,397 has been optimized. We're going to keep on going. The support for this system, 368 00:19:58,480 --> 00:20:01,997 the the amount of software we're going to bring to it, middleware, algorithms, 369 00:20:02,080 --> 00:20:06,477 I mean, it will never end. This is just a magnificent thing that when you when 370 00:20:06,560 --> 00:20:10,877 you finally look at it, you look at this beautiful laptop. Pav stands up here 371 00:20:10,960 --> 00:20:14,397 with this beautiful laptop, you know, but the number of engineers between our 372 00:20:14,480 --> 00:20:17,597 two companies and all of the companies around the world that did this 373 00:20:17,680 --> 00:20:20,557 incredible all because we believe that 374 00:20:20,640 --> 00:20:24,637 the personal computer was our tool tool for people. Now it's going to be tool 375 00:20:24,720 --> 00:20:28,797 for people, but it's also going to be our personal assistant that uses the tool. 376 00:20:28,880 --> 00:20:32,317 >> Now, one last thing for both of you. I think if you've been on X for the last 377 00:20:32,400 --> 00:20:36,077 few days, there has been this intense conversation about I would say personal 378 00:20:36,160 --> 00:20:39,277 agents. We've been talking about productivity and maybe enterprise agents 379 00:20:39,360 --> 00:20:44,317 here whether it is muse instinct there's this question about hey when this agent 380 00:20:44,400 --> 00:20:46,157 goes out and makes a purchase is that me 381 00:20:46,240 --> 00:20:49,037 or the agent who is liable for it and if 382 00:20:49,120 --> 00:20:52,237 I'm a service provider how do I react to 383 00:20:52,320 --> 00:20:56,557 this human being no longer showing up on my website and people are trying to 384 00:20:56,640 --> 00:20:59,597 build standards and as an industry I think we're trying to figure this out. 385 00:20:59,680 --> 00:21:02,877 So I'm sort of curious if you kind of had to peer into the crystal ball a bit 386 00:21:02,960 --> 00:21:07,437 where do you see this headed right is it standards is the industry coallesing 387 00:21:07,520 --> 00:21:10,157 around some new behaviors where do you see this headed 388 00:21:10,240 --> 00:21:12,557 >> yeah I mean first of all I think it starts with a little bit of the 389 00:21:12,640 --> 00:21:14,717 conversation we had uh in the early part 390 00:21:14,800 --> 00:21:18,797 which is trust um and trust starts with 391 00:21:18,880 --> 00:21:20,637 things like hey are these agents 392 00:21:20,720 --> 00:21:23,357 contained do they have identity is the 393 00:21:23,440 --> 00:21:26,157 exhaust uh of the agent with the 394 00:21:26,240 --> 00:21:28,557 identity separate from the user right 395 00:21:28,640 --> 00:21:31,357 which is one Pawan's thing was how do we 396 00:21:31,440 --> 00:21:33,837 make sure the agent ID that's even 397 00:21:33,920 --> 00:21:36,797 working on behalf is separate from the 398 00:21:36,880 --> 00:21:39,437 user and you need it even for everyday 399 00:21:39,520 --> 00:21:42,077 uh usage but you also need it for what I 400 00:21:42,160 --> 00:21:43,997 think you're referencing which is how do 401 00:21:44,080 --> 00:21:46,877 I then trust what do I delegate in fact 402 00:21:46,960 --> 00:21:49,437 if you look at the first decision that 403 00:21:49,520 --> 00:21:53,437 we are making is what is my delegation 404 00:21:53,520 --> 00:21:55,757 um you know authority that I am giving 405 00:21:55,840 --> 00:21:59,277 to the agent uh on my behalf right and 406 00:21:59,360 --> 00:22:02,877 that is a decision that's going to be based on trust and it'll keep increasing 407 00:22:02,960 --> 00:22:07,997 the more autonomous it is uh you trust it and go so then I think on the 408 00:22:08,080 --> 00:22:11,997 consumer space this will come down to even you know I I think of it as 409 00:22:12,080 --> 00:22:16,637 insurance will price this >> insurance >> because at some level we can all talk 410 00:22:16,720 --> 00:22:20,477 about it's not like oath is hard to do right I think the protocols are good we 411 00:22:20,560 --> 00:22:23,277 should do them and t but ultimately it's 412 00:22:23,360 --> 00:22:28,237 going to be pricing the the essentially 413 00:22:28,320 --> 00:22:32,157 the agent working on my behalf. Uh, and 414 00:22:32,240 --> 00:22:35,197 how does the marketplace price it? 415 00:22:35,280 --> 00:22:39,357 >> Your question at the at the foundation of it, MXC is so foundational. Open 416 00:22:39,440 --> 00:22:42,637 Shell, the work that we're doing together, it's about it's about trust. 417 00:22:42,720 --> 00:22:46,077 It's about containment. It's about observability, monitoring. If we can't 418 00:22:46,160 --> 00:22:47,437 get that right, that's right. Oh yeah. 419 00:22:47,520 --> 00:22:51,197 >> If we just can't get right. So that's the first part. We have got to secure 420 00:22:51,280 --> 00:22:54,717 that part of it so that the agents that are operating on our behalf, doing 421 00:22:54,800 --> 00:22:57,117 things on our behalf, we give it minimum 422 00:22:57,200 --> 00:22:59,517 rights. It has credentials. It is 423 00:22:59,600 --> 00:23:02,157 monitored and trusted. Only then can we 424 00:23:02,240 --> 00:23:07,677 allow these agents to really go off and do do uh do tasks on our behalf. 425 00:23:07,760 --> 00:23:11,037 >> You know, today's an awful moment. I remember when open clock came out from 426 00:23:11,120 --> 00:23:15,597 there with the terminal hacking to get their own um gateway to here where it is 427 00:23:15,680 --> 00:23:19,517 an embedded piece of Windows, you know, with tokens running accelerated. It's 428 00:23:19,600 --> 00:23:21,997 quite something. >> So much engineering went engineering 429 00:23:22,080 --> 00:23:23,517 went into that that arrived today. 430 00:23:23,600 --> 00:23:26,957 >> Amazing. And it is a great device. Let me just tell you that honestly. Well, 431 00:23:27,040 --> 00:23:30,477 Jensen, Satia, thank you so much. It's a great day. Thank you for having me. And 432 00:23:30,560 --> 00:23:32,317 I'm going to throw it back to Pav 433 00:23:32,400 --> 00:23:35,517 >> and Satia and Microsoft team. 434 00:23:35,600 --> 00:23:39,437 >> We have reinvented the computer as we know it. 435 00:23:39,520 --> 00:23:43,037 >> Thank you so much, Jensen. Thanks for the partnership. Really excited today. 436 00:23:43,120 --> 00:23:45,117 And I think as you said like one thing I 437 00:23:45,200 --> 00:23:47,917 if I can end with Jensen just in passing 438 00:23:48,000 --> 00:23:52,317 talked about the vision for this was how do we I mean Windows has always been 439 00:23:52,400 --> 00:23:54,717 about how do we not leave any piece of 440 00:23:54,800 --> 00:23:58,077 code any piece of framework behind and 441 00:23:58,160 --> 00:24:01,837 bring it all. The fact that he talked about hey we're going to bring all of 442 00:24:01,920 --> 00:24:04,477 Windows apps all of the the graphics 443 00:24:04,560 --> 00:24:08,397 apps all of CUDA all of OpenGL and now 444 00:24:08,480 --> 00:24:12,877 with the token factory that is what is mind-blowing. Windows has been with me 445 00:24:12,960 --> 00:24:15,037 for 34 years. Looks like it's going to 446 00:24:15,120 --> 00:24:20,920 be with me for 34 more. >> There you go. Thank you so much.