China Open-Source, Compute Arms Race, Reordering Global Trade | BG2 w/ Bill Gurley and Brad Gerstner
00:00:0015% on all goods coming from Europe, 0%on US goods going to Europe. So anopening up of Europe markets, paying us15% and on top of that gettingcommitments like $750 billion, almost atrillion dollars of energy purchasesfrom the US. Or look at Japan, whichthey announced last week, another hugemarket. Again, similar. They're going topay tariffs to the United States. Notariffs imposed on the United States.and they're going to invest $550 billiondollars into the US in a way thepresident gets to direct. So I wouldlove I would I just think we need togive the president credit where creditis due. Everybody said this was going tolead to retaliation to trade wars wasgoing to be disastrous for the US andall we've seen so far is deals dealsdeals deals. And I have to say, if thiswas the CEO of one of our companies,right, let's say we had a board meetingat the start of the year and he outlinedthese plans and we said, "Hey, we'rereally nervous about this. You know,this is a high-risisk, highrewardstrategy. It's either going to backfireand we're going to fire you or it'sgoing to work really well and we giveyou a bonus." If we're measuring themhalfway through
00:01:16the year, I would saythat he's in line for a bonus.[Music][Applause][Music]All right, the summer pods back inaction. Uh, good to see you guys. Wehave our good friend Sunonny Madra, theCOO of Grock, joining from I don't know,Sunny, it looks like some fancy hotel inSaudi Arabia in the middle of the night.Good to see you.>> You picked it up. You picked it up. Youhave a good You have a good eye for theMiddle East. And Bill, Bill, you got offyour boat catching blue fin tuna. You'reuh you look like you're in some officesomewhere. So good to see you.>> I am uh borrowing Mitch Lasky'sincredible podcast setup in in ourWoodside office with the uh Oh, nicehigh highde SLR camera.>> Nice. You're looking you're lookinggood. Sunny, of course, our good buddySunny is the COO of Grock. Um I don'tknow, Sunny, probably a few hundredmillion of revenues, maybe doublingyear-over-year. I just saw something.You're rumored to be raising 600 millionat a $6 billion valuation. Maybe thathas something to do with you being overin Saudi Arabia. Um, and of course,you're hosting all the uh open-source
00:02:31models in in your inference cloudsaround the world. Is that about right.>> Yeah, you got you got it right. Youtouched on all the key points. You know,we don't comment on speculation, but youtouched on some good points.>> Well, it's great to see you in DC lastweek. Um, you know, our good friendDavid Saxs is really on a heater. First,it was the crypto summit a couple weeksago. Of course, the Genius Act gotpassed, which really teed up thesestable coins and and and now the ClarityAct around market structures making itsway through Congress. And then of courselast week was the AI summit where thepresident laid out um a multi-prongstrategic plan for American AI that bothextends the government's investment andand leadership um but also acceleratesthe distribution of the American AIstack around the world. Both of thosethings, both the crypto and the AIsummit that he put together, I thoughtare key contributions really to the nextgeneration of American technologyleadership around the world. It'samazing to see that much progress in sixmonths. Congrats to Sachs>> and the rest of the team. Catios, DeanBall, Sriram, they really all kind ofbrought it brought the heat.>> Yeah, for sure. Of course, the AI actionplan
00:03:46was focused on maintaining globalAI leadership, particularly over China.And I hate to say it, but we have reallybeen our own worst enemy. You know, itseems like excess regulation oneverything from energy production tomodel development to semiconductor chipdistribution. It's really been a bit ofan unforced air by the US over thecourse of the last 24 months and handeda lot of momentum to China. and I thinkreally threatened our leadership. So,this is kind of a 180 to get Americaback on track. You know, we'veunderestimated Huawei and the Chinese AIdevelopment, model development really atevery step of the way. So, I want tokick off today talking about recentdevelopments with the base models andreasoning models coming out of Chinabecause, you know, we had this freakoutmoment earlier in the year with Deepseekthat we all remember. Remember, Nvidiastock plummets. Everybody inWashington's talking about it,>> but since then, you know, people kind offorgot about Deep Seek. But the realityis China's been on a roll. I mean,they're dominating the global landscapefor open-source models. We've seen sixto seven highquality open- source modelproviders, many of the fastest growingin the world.
00:05:01And this at a time whenAmerican open- source, right, Llama 4,has been sputtering a bit, really losingits mojo around the world. Um, so Quen,the open- source model out of Alibabahas passed, I think, 400 milliondownloads. Of course, that's releasedlike these other models under the Apache2.0 open source license. So, very openas Bill's talked about. But, Sunny, youtweeted, and one of the reasons I wantedto get you on the pod this week is youtweeted, you know, that the all of theseopen- source models are really comingtogether in China. Um, they'releveraging one another. they can distilland generate synthetic data on eachother's work. And you went so far as tosuggest this might allow them to passthe best proprietary models coming outof the US um yet this year, maybe by Q4of this year. So why don't we dig inthere? What is your theory of the case?Why is China, you know, doing so well inopen source? And should US modelcompanies like OpenAI and Anthropic beconcerned?>> Yeah. So, uh, let let's kind of tie itinto, I think, three important thingsthat we see happening. The first onebeing, you know, the Chinese and and thepresident addressed this at the AIsummit,
00:06:17right? He addressed um the thepoint around using copyrighted work andhe says, you know, he used a greatexample. If you read a book and you useit, you're not violating the copyrightthere. And so, he addressed that concernand that was one of the major thingsthat, you know, a lot of people didn'ttalk about, but I think it's importantfor the model makers. And so the Chinesejust have been able to work around thatbecause of, you know, their position onIP. And what we're really seeing here,and I think, you know, Bill teed it upeven better off off of my tweet, whichis, um, you know, they're able tocompound. So what you're seeing veryquickly is both the open- source nature,the open weights nature, uh, allow themto basically compound on each other. Soinstead of working in silos and insteadof having to create giant trainingclusters um separately they canbasically take each other's work buildon top of it almost consider it like aremix of someone's model K2 sort of awell-known remix of what what Deepseekhad done and now we're starting to seethat happen really fast and we're seeingtwo dimensions of it going quickly. Onewe're seeing the leading edge modelsgetting quicker and we're then seeingthem distill down smaller you know turbomodels really really fast as well. Therewas a a release today of a quen you know30 billion parameter model which isperforming as good as GPT40.
00:07:32So thinkabout that right and GPT40 was you knowworld class not that long ago. So thoseare the reasons that we're really seeingan acceleration right now.>> You know it I want to dig into thismodel development in particular. Youknow a year ago we were talking aboutthese models being these you knowstochastic parrots um you know and wereally had to compress the entireinternet. So you go back to GPT4, youknow, and and and you're compressing,you know, the entire internet. But nowwe really don't need to do it becausewe've trained them to use tools like theinternet, right? The they're truereasoning engines. When I ask a questiontoday, you know, it doesn't just spitout an answer immediately. It goes anduses the tool and it searches theinternet. Um, so if you don't have tocompress all this wikipedia information,I don't know, take take a subject likeWorld War II, you just need to know howto go out and use the internet to findthe information to summarize it in realtime. How has that changed the pace ofprogress and the balance between openand closed?>> Yeah. And so it's spot on, right, Brad?what we see now and you see it when youuse these reasoning models and what Isuggest everyone do is when you're usinga reasoning model you can usually expandout its thought process so when you aska question it'll say oh the person isasking a question about this
00:08:48what shouldI do let me go and maybe search theinternet let me go do a few differentthings it can have a lot of differenttools and so the push has been towardsyou know really really strong reasoningmodels and you know we have to givecredit there openai really started thatwith 01 that was really the firstreasoning model that was put out therebut I think on the back of the researchchurch in the back of what you knoweveryone talked about and you know thepod's good friend Nome Brown was theleader on that program everyone's beenable to look at that and say let'sreframe the problem and this allows usto build stronger reasoning models thatdon't have to compress like you said allthe internet's information and oncethey're coupled with strong tools youstart getting these really reallyincredible results that don't even justshow up in benchmarks because none ofthe benchmarks really allow you to use atool to answer the results and if theydid we're going to see a whole bunch ofnew set of results that happened there.>> Hey Bill, just in many ways I think thisvalidates what you were arguing over thelast 6 to 12 months. You said this waslikely to happen. Everybody knows you'reyou're one of the biggest proponents ofopen source in the world. Um and youwere telling people don't make you knowthe Chinese are going to use open sourceto their advantage and now we're seeingthat in a really profound way. Why isChina so successful here and what wewhat what what can
00:10:03we learn from them?Well, I we've talked about it in thepast. I won't dwell on it, but China gotexcited about open source about 20 yearsago. It's not a new thing that'shappened. And you can imagine when ummost of the world accuses you of IPtheft that that embracing something likeLinux and all the other software opensource products um seems very appealing,right? And so I think it became kind ofa common way of operating within China.Um and it, you know, it's a country thathasn't prioritized uh IP protection theway we have, you know, around patents.And I I could make an argument that, youknow, there's way more prosperity ifideas are shared instead of instead ofprotected. But um I don't know the onething I don't know is in the in thecurrent AI situation you know was thegovernment promoting open source andencouraging it or did it just developthrough competitive forces right>> but but now you have you know a scenariowhere these companies you know are youknow are first of all there's new onespopping up and so you know I almost feellike an idiot when when Kimmy comes outand moonshot and
00:11:18and and And this week,I guess I don't even know how topronounce it. Zippu, you know, releasesa model and I go on Pitchbook and I lookit up and they've already raised $1.4billion. So, shouldn't have been asecret, but I didn't know about it, youknow. Um, and all of a sudden they're inthe leader tables on Open Router andlike, oh my god, you know, they'recoming out of everywhere. Um, and whatit shows you is just that when you havea competitive dynamic where every singleplayer, and I think there may be sevenor eight deep pocketed players with openmodels in China, they they all learnfrom each other extremely fast. And andin this case, unlike software, you canuse one model to distill the other andmake it better. So, it's almost like aaccelerated, you know, form of that andyou just get massive quick co-evolution.And I I came up with a a uh a littleanalogy for people. I'll try and do itquickly, but imagine you had twocommunities. They're both farmingcommunities and and let's say there's 10to 20 farms in each. And in onecommunity, they come into the farmersmarket once a week and they just competeby selling their products, but then theygo back. in the other community whenthey come into farmers market they alsoin addition to competing and selling
00:12:33they they they're forced I don't knowwho would force them but they're forcedto share all their best practices fromthat week>> with everybody and everyone does it andeveryone shares their best practices andthen if you ran that exercise over twoyears or whatever obviously thecommunity where the best practices areshared across all farms is going to havea higher global output for the communitythan the one where you've just gotproprietary ideas uh driving you knowthe individ and and just competitionwithout the idea sharing and that maylike even me saying that may cause somepeople to scream that's socialism orlike you know they may not understandopen source or how it works or why umbut but I do think you end up with ahigher fitness level for a communitythat's behaving that way overall you mayend up with a lot less chance of abreakout monopolistlike we've had in many of the thesectors in in American technology.>> Well, I mean, let's let's just assumethe Chinese government is in factencouraging this in whatever ways,right? I mean, if you look at therelease of the AI action plan last week,the Trump administration, you know, hada section>> they did>> which which is about encouraging
00:13:48open-source and open weight models inthe US>> saying that these could become standardsin some businesses and academicworkloads. And it's important they'rebuilt on the American AI stack. So, youknow, as an aside, the Chinese quicklyfollowed uh, you know, the the AmericanAI plan. I think they released theirs acouple days later where they called forthe establishment of a global AIcooperation organization,>> which I thought again is is is isinteresting. So, you know, Bill, how doyou feel about, you know, like we we wehaven't seen that much traction in theUS labs uh on open source? Obviously,Llama has been probably the marketleader there, but this is for both ofyou. Uh, you know, handicap for me ifyou will, how you think this plays out.First, Sunny, maybe you start. What doyou see at Grock? Do you see a lot ofdemand for these Chinese open-sourcemodels? And if so, what would it takefor an American open- source model tocatch up?>> Yeah, one of the uh things that weshould pull up is the chart ofintelligence to price. And one of thethings that you see with the leadingopen source models now, which are theChinese,
00:15:03is 90% of the quality in termsof intelligence, but at a 90% pricediscount. And I think whenever you offerthat to anybody, you're going to seepeople want to use that, whether it'sindividual developers or enterprises.And so we're seeing that>> let me interrupt there real quick. So ifyou're looking at this chart, right, inthe top right of that chart, uh you'llsee a cluster of these Chinese open-source companies, right? And thevertical axes here being reallyintelligence, the horizontal axes fromleft to right being the cost per milliontokens. And so you really want to be inthe top right of of of that model. Highintelligence, low cost. And what thismodel what what this uh uh chart showsis that you you know to Sunny's pointyou can get 90% of the intelligenceright for 10 or 20% of the cost. And theresult I assume Sunny is that you'reseeing huge demand at Grock and in SaudiArabia where you are right now for theseChinese open source models.>> We are. And then now just taking thatforward. What do people want? They wantsome accountability. And that's whatyou'll get. That's what you sort of gotout of Linux and say Red Hat, right? Asas great as Linux was
00:16:18and Bill wastouching on it, the majority of theenterprise was using, you know, adistribution which they could go and,you know, point to someone if theyneeded something. And so I think theworld wants models that they can getfrom companies that they can go to. Andso to answer your question, whathappens? I think if we look at Q4 thisyear or Q1 next year, I'd be willing tosay, you know, the a top three worldwidemodel will be a US-based open sourcemodel. And you know, we've got two bigefforts happening there. We know we havethe open AI open source which you know alot of people have been working on andopen AAI and even you know Sam hascommented on that and its release laterthis summer and then we have all theefforts by meta and if you take you youcombine both of those things together Idon't think you end up with somethingthat ends up further down the the listin terms of intelligence andor price.I um one thing I wanted to highlightabout the China situation that I thinkmight inform the US situation. I was umI was having a conversation with thisextremely young AI uh entrepreneur thatI know and he he he was pouring over thethe Zippu I hope I'm pronouncing thatright paper and um he asked me someinformation. I
00:17:33went on Pitchbook andsent it to him who had funded it. And hehe he he asked me, he says, "Why isAlibaba funding all these things whenthey've got their own model?" Andbecause they're in several of the othercompetitive plays and it reminded me of,you know, a lot of the the the pointsthat I've made about open source is likeif you're not if you if you're notconfident you're going to win onoffense, you want to play defense. Andso for any large tech company,commoditizing a a potential threat is isactually quite valuable. You know, youlook at what Facebook did with the opencompute initiative inside of their datacenters. And so, you know, it may justbe very well be that that Alibaba justwants to make sure there's no bitedance, you know, that equivalent in theeye space. And that would be prettyrational. And the reason I think that'san interesting data point when you thinkabout the US and you know there thereare several big tech companies that seemto be not on the bleeding edge of AI.You know you got Microsoft maybe I meanthey have access to open AI right nowbut they might lose that or whatever.>> You've got um Amazon, you got Apple.
00:18:49youknow, if I'm at any of those companies,I'd be funding a open-source competitor,you know, rather than funding, you know,like Amazon with Antropic. I thinkyou're in a much better position toencourage open source. And so, Iactually think we may>> But Bill, are there are there are therea bunch of open source startup models inthe US?>> That's where that's where I was goingnext. I I think you're going to see newentrance um pop up um that that that tryto co-evolve with the Chinese models.You know, is Linux American? Is LinuxChinese? Like no one thinks of it ashaving a a doicile, right? And so umthis is just me predicting. I don't youknow I I just think you're going to seejust like you saw um Kimmy and and Japupop up. I wouldn't be shocked if you seeother new entrance pop up that aretrying to be um like sanctioned or or oryou know cleaned you know use Redhead asan example Sunny version of these thingsbecause I think if you start with accessto those Chinese models it wouldn't takeyou long to move into a near place andyou wouldn't have to spend the kind ofmoney the foundational
00:20:04model companieshave. Um, and and then you may see a bigfight around regulatory capture wherethey someone tries to say that's notallowed or whatnot. But I I do expect tosee that. And if I'm the mistrol team,if you're not distilling on theseChinese models, I don't know what you'redoing right now,but I don't have any data on that front.>> OpenAI is rumored to be launching theiropen source model any day. Sunny, whatwould they have to do? So to your point,you predicted, go back to theintelligence and pricing chart, right?So if open AI was in the top right ofthat chart, i.e. if they're able todeliver something at the theintelligence of let's call it Quen, andthey're also able to deliver it tomarket at, you know, 20% of the cost. Itwould seem to me that actors around theworld certainly actor it would be uh youknow part of the American AI action planwe'd want everybody in the world to usethat model and do you believe they havea shot at out competing being the upperright you know by the end of the yearand and and if so do you think that willbe the outcome like these companies thatare using Quen on Grock do you thinkthey would prefer to use OpenAI so longas it was equally capable and equallyprice performant
00:21:20So two things that we see is brand andyou know the US doiciled or you knowsomeone that they can kind of point atthat wins and so if that shows up itwill win because if you're a company>> and you know at some point you have toyou know get your teams to sign off onwhat is it that you're using what arethe risks associated with it and likeyou know who is liable if something goeswrong and so I sort of feel like um withOpenAI's release and if you know Metacharges back or even if some of thesestartups emerge that you know we canpoint at I think we'll see a huge shiftback towards those models versus versusthe Chinese ones.>> Yep. And we really don't know at thispoint unless you guys have some insideknowledge like when when Meta makestheir second push with all these hiresthat they've made are they are theygoing to remain committed to open sourceor even be more open? We don't we reallydon't know yet at this point.>> Right. You certainly seen some of thoserumors out there. Um you know, I've seenrumors that on Twitter that they weredebating whether or not they should backaway from open source. Um my hunch isthat that's a misread. My hunch is thatthey're going to stay very
00:22:35committed toopen source, but they may complement itwith a proprietary model. That would bemy guess. As opposed to scrappingopen-source alto together. And Brad, canI throw one thing out there? I thinkpart of the pitch to get everyone thereis that it's open>> because everyone that they're pullingover are coming from closed places. Andso if you're really passionate about thework you're doing and you're passionateabout where this is going to come andthere's only there's only one companythat can fund that to that scale and doit open is is those guys. So I thinkit's part of the pitch.>> Yeah. Some of some of the researcherslike have a religious belief in it whichwas evident in the in the interview withthe Deep Seek founder like it's it'soddly like higher on their Maslov youknow hierarchy than the money.>> Of course they're getting the money.>> Of course it's both. This is I I I thinka really important point I want to comeback to Sunny is that you're seeingmassive demand for these Chineseopen-source models today preciselybecauseenterprises around the world can utilizethem. They have 90% of the capabilitiesat 20% of the cost. 10 or
00:23:5020% of thecost, you know. So it turns out if youdeliver something really powerful andreally cheap that's more important tothese players than American valuesaligned. But if you gave them somethingthat was super powerful and super cheapand aligned with Western values, right,that you think that would be uh thewinning formula for an American OpenAImodel to top the uh uh the distributionleaderboards around the world. Is thatwhat I'm I think yeah on a place likeopen router you know where you can seewhere this is happening I see we wewould see it rise to the top in a fewdays>> u that's music to David Sax's earsbecause you know clearly in thestrategic plan they're worried aboutChinese open- source models dominatinguh globally and we're we're if you justwatch the pace of releases the qualityof the releases out of China the cycletime on the innovation in the opensource community out of China. It'sfaster and better at the moment. So, andthe only real big development, you know,we may see some of these new startups,Bill, Bill, we may see a reboot out ofmeta, but the the one that has, I think,everybody really
00:25:05holding their breathand hoping that we see something reallycapable and powerful uh is this opensource model that's been promised out ofopen AI. Now, of course, Elon says he'scommitted to open source as well. Gro 4is a great model is impressive what theyuh what they put out there. Any idea uhSunny about the open source plans out ofout of uh X?>> Yeah. So I think like Elon's been youknow pretty said it on Twitter thatthey'll always open source onegeneration back and so while they wereon three you know they should havegotten to two and now they're on four.So I think the the thinking is that theywill get there. You know, my my onlyguess would be is that um you know,right now if they were to open sourcetwo or even three, it's so far behindthat um you know, what's what's thepurpose in doing it? It may not even beutilized and you'll maybe end up havingto deal with just, you know, a bunch ofinternet or Twitter FUD. Just comingback to open AI, I I will tell you it'slike one of those things that really yourarely see in the enterprise. It'salmost like the demand for like a, youknow, like a Tesla Roadster or somethingor or Model Y before it came out.Everybody asked for it. It's like that'sthe model that everybody wants to useright now. And so, you know, we
00:26:20can'twait till it it comes live and and it itit you know, one it's on Grock, one it'sall over the world. I think it's goingto be a real big one for everyone.>> While while we have you, Sunonny, um youknow, there's really just been thisexplo explosion in the compute armsrace, right? Um there was a big debateyou were on the pod a year ago. We hadwith Bill, you know, had we topped outon compute demand, you know, were weentering an overbuild all Cisco 2000? Umand it's really been remarkable. I thinknow that's very clear. But just a coupletweets out of Elon and Sam Alman thelast couple weeks on this compute demandhas have really caught my attention. Ifyou look at this one out of Elon talkingabout the X.AI AI goal is 50 million inunits of H100 equivalent. And Clark Tangon my team tweeted something that brokethat down. Um, which showed that thatreflected something like, you know, 4million total GPUs and an energyfootprint of like 11 gawatt, right? Andthen Sam Hoffman comes out and and andtalks about their deal down in Abalenefor 4.5 gawatt and the fact that theywere going to come in well above the$500 billion estimate
00:27:36that they hadpromised to the government. So thesecompute clusters now that are beingtalked about being built over the next 5years, these are massively bigger thanwhat we were even talking about a yearago. And I think they reflect this movetoward inference time re reasoningagentto agent interaction reasoningengines you know Jensen's comment on thepod last year that inference was goingto 1 billionx and the consequence youknow what we were going to need in termsof compute power to power all that. Ummaybe just reflect you're in the middleof all this you're building out your owninference clouds around the world. Um,is this a lot of hyperbole and chestpounding or do you actually see uh thedollars going into the ground in placeslike Saudi Arabia and around the US?>> Yeah, I'm going to I'm going to justquantify it with uh with Google for asecond.>> So, you know, in the in in in the um inthe Mary Maker bond deck, you know, theyhave a slide there that shows Googlewent from 5 trillion tokens a month to480 trillion tokens a month. And theyhad just put some press out that theycrossed like, you know, 800 trillion.And I saw something today. They crosseda quadrillion. And I had to look thatup.
00:28:51That's a,000 trillion. And so in ain a course of a year and a few months,they've gone from 5 trillion to a,000trillion. So that's 200x right there. Imean, that just shows it to you. Um,without having to look at anything else.>> Uh, the amount>> every single search query on the planettoday is now an inference transaction.Correct. And so um you know and you seeit uh you know in anthropic uh with youknow their continued fundraisers goingyou know through the roof uh that'shappening because they're they're seeingthe amount of token consumption. And soanywhere we lay infrastructure um wefire up a rack it becomes fully consumedwithin a few hours.>> You're talking Grock.>> So you have demand far outstrippingsupply even at Grock. Yep, we do.>> So, Bill, you see these fundraisingannouncements that are being discussed.Just CNBC's reporting tonight. I thinkthat Iconic is going to lead a $5billion round into Anthropic at 170billion. In the case of Anthropic, it's,you know, uh 170 billion on rumored $5billion in revenue. X has been rumoredto be raising at 150 to$200 billion.
00:30:07So,you know, you've never in the history ofventure, you've never seen fundraiserslike this. One of the topics that isbeing hotly debated on on the Twitter,uh, is that there's massive interveningdelilution in these rounds, Bill,because of the employee option grants orthe employee RSUs that are needing to begranted to keep the employees uh, youknow, uh, in these businesses. So justobserving this a little bit from afar. Idon't think you guys you're a directinvestor in in any of these uh majorlabs.What what do you see? What do youobserve? What are your warning signs uhyou know about the size and and thedemand that you see in these rounds?>> Well, I yeah, I've never seen anythinglike it. I mean, I saw you know, throughthe Uber andyft situation, I saw aprecursor to this, but this these thesedollars are even bigger. and the amountof money that these companies arewilling to lose in a year. You know,it's I still think it's particularlyinteresting that Google has to competewith Open AI because OpenAI is going tolose 7 billion this year and Googlewon't like like they would never allowthemselves to do that. And and thisstarted back in that previous
00:31:22era wherewhere for the first time ever you sawprivate companies have a competitiveadvantage and that they can be morerisk-seeking with capital than thepublic companies are allowed to be.>> Um>> but I think in the past 12 months we Iwe've seen um OpenAI Meta and certainlyX you know move into this place. I callthem I call them the cost is no objectthe CNO group um where they're justthey're just putting out press releaseafter press release and opening datacenter after data center and you know wethere there are other people I think youknow you look at Microsoft you knowchoosing not to extend their capexbudget you look at that Amazon examplewhen we spoke to the the Levant brothersat codeu where they're they're notkeeping up with the the Nvidia purchasesrelative to their AWS share. And sothere are a few companies that are backon their feet and there's a fewcompanies that are really pushing thegas pedal. And I can't there's a few inthe middle. I can't tell um whetherAnthropic has the uh audacity and themeans to raise enough money to to startbuilding data centers themselves. Theyhaven't so
00:32:37far. Um but it's uh you knowit's a sport of kings, you know. It's itis a sport of kings like there is umnever been this amount of money spentright now. Nvidia and others building inthe stack like like uh Dell that we wespoke to a few weeks ago like they'rethe winners um in a pick and shovel gamethat's got this amount ofaggressiveness. It's uh and maybe maybeSK Heinik too. I don't know.Sunny, Sunny, do you see um you knowgoing back to the well-worn cliche thatevery glut or that every shortageultimately ends in a glut? Um do youhave any evidence on the horizon whereyou see uh uh supply outstrippingdemand?>> No. I mean I I was going to ask this toBill as he was just saying it. Bill,like on a daily basis, are you consumingmore tokens or are you consuming more,you know, just traditional um you know,web lookups, right? And I'd be willingto guess you're consuming more tokens.>> Oh, it's insane.>> And and and tokens are are increasinglike when you're using those reasoningmodels, you don't see all the tokens,right? They don't publish it, butthere's Right. This is where
00:33:52totallyYeah. 10 to 100x more than the your veryfirst AI search. Yeah, for sure.Absolutely. The one thing the one thingI will So you're right. I I'm doing>> Go ahead. You finish.>> No, no, no, no, no. And I just wanted tosay even yesterday, Anthropic had to putthis uh you know, press release plus,you know, product change out saying,"Hey, we've got to throttle everybody,right?" And this is like cuz we have allthese people using way too much of ourour services. So I think you know if youhave intelligent models right and youhave the capacity for it it's one ofthose things people are consuming Jven'sparadox or whatever. Yeah. Sorry. Goahead, Bill. You're>> No, I was just going to offer one caveatto to this uh super uh exciting line ofof of thought, which is because of theamount of venture capital out there. Umcompanies are not pricing the cost likeno one no one none of the modelcompanies I don't think anyone even inthe verticals like no one's pricing thecost because they're pricing to takemarket share. And you know, you and I,Sunny, we had a previous discussionabout the unlimited pricing andinference has variable costs. So, isthat
00:35:07even sustainable, right? And andeven when people say to me, "Oh, well,we're going to run out of power." Youwouldn't run out of power if you justtook the price up. Like, like the thingthat throttles demand is price, but noone here is raising price. LikeAnthropic doesn't need to throttle. Theyjust raise price, but they're notwilling to do that because they'reafraid to lose share. And so everyone'spricing to share. That means they'repricing under cost. There's rumors ofeven some of the best known brands in AIhaving negative gross margin. And so Idon't know when that settles out, butthat there that will create a bump inthe supply demand curve if that ever hasto be, you know, fixed. Um but for nowit doesn't>> because because>> Can I ask a question there?>> Yeah.>> Yeah. No, no. which is going back to thepoint that we made on open source, butif you know in the back of your mindthat there's something that's 90% asgood, but it's 90% cheaper. How doesthat factor in because we've also neverhad that that factor as we're goingthrough this growth curve. I mean sincealmost since we started the pod you knowI've I've routinely highlighted that thesteepness of that price curve on youknow
00:36:23as you as it kind of becomes youknow less less cutting edge is somethingI've never seen before. I've never everseen it. And I'm sure that a lot of umpeople sit around and say, well, it'sokay if I'm losing money here because,you know, six months from now, I'll justuse the older model. And we also talkedin the past about how in the internetage all the startups began with Oracleand Sun and eventually they all moved toLinux and MySQL. And so there was athere was a we got to win at all costphase and then there was a phase whereyou started worrying about cost andoptimization. And so one day, one daywe'll likely, you know, make that makethat move. And and and I a few of thecompanies I've talked to that arerunning inference at scale, they arealready starting to think that way. Likethey're looking right they're looking atit, you know, from that from that lens.But I but>> well and and I think that's why Grockand Cerebras are doing so well. ButSunny, give us give give us an, youknow, an an example. I would imaginethat the wind surfs of the world and thecursors of the world and all these folkswho are building these you know thesecoding agents they've got massive
00:37:38youknow demand for their applications rightbut they're paying through the nose toanthropic or to these underlyingproprietary model providers to be ableto do that so what's the dynamic thatyou see there do you see them running toimplement Quinn or some of these cheapermodels>> yeah without kind of getting intospecific any one of them, but like youknow multiple folks are building theirown models uh based off open source and>> right so they could just go distill anyone of these models.>> Correct.>> Right. And you know given that they'rethese very uh lenient Apache licensesand eliminate the entire cost of sonnetthat sits under it for 70% of use cases.>> Yeah. And and like Bill said, turn thatinto a premium offering, right? and saythat's the you know the gold and youknow the silver and the bronze are builtoff uh you know something that's like Isaid oneten the price>> that seems to me Bill you know if if Ihad to forecastyou know if I'm open AI I'm running aconsumer business with really high grossmargins right because consumers are uhuh less sensitive
00:38:54to what they're payingtheir their their their intensity of useis lower. Whereas when if if somebody'swriting code, you know, there's thevariable intensity is high. And so, youknow, for them, it would make sense tolaunch an open source model back towhere we started the pod and price itreally low to, you know, drive shareknowing, you know, it reminds me alittle bit of Amazon back in the day.Amazon had this monopoly retail businessthey could use to subsidize AWS gainshare for a decade and then begin totake price. That would be a rationalstrategy for open AI to follow. So youtake the profitable consumer businessyou use it to subsidize you know the thethe market share in other applicationsthat you hope to build.>> Certainly a reasonable strategy. I I I II'm not inside that company. you haveway more knowledge than I do, but um youknow, I pay the $200 a month and I doevery one of my searches on four or fiveand I'm probably I'm probably negative,I would think. Yeah.>> And so I do think there'll be somerationalization where these models umkind of self-pick which one they'rerunning based on what you need. I
00:40:09probably don't need to be using>> and move to more of a consumption logic.Yeah,>> move more to more of a consumptionlogic. they're already uh they'realready doing that in in parts of theirenterprise business. Um and and youknow, as they've transitioned to more ofthis consumption logic, I think it's ledto some real unlocks. Uh you you know,for the business, I think it that makesit harder if you're in the the lab gameand you don't have a consumer productand you don't own an application thatyou can drive high gross margins. Ithink then gets back to this question,how long can you run the your businessfor share, right? Hoping that somedaybecause you know they all exist at thebeneficence of the capital markets andthe capital markets are willing toprovide an incredible amount of capitalto these businesses today. They sureare.>> Um but but but you and I have we'velived through these periods where thatdisappears quickly. And can I can I putsomething there just to kind of hearyour feedback on it guys which is lookat Google and the TPU and Google isclearly you know by these numbers thatwe're seeing right putting out moretokens than anyone else right do they doyou guys believe they have a strategicadvantage because they have their ownhardware they're not having to pay
00:41:24youknow 80% margin on something that theycan generate tokens with like how do youguys look at that business and sayclearly they look to be sort of thelargest at least openly saying um thelargest processor of tokens. I thinklook the key data point in that casewhich I don't have the data but I'd beglad to repeat it if someone shared itwith us is how many non-G Google uhapplications are running on the TPUslike how many third party customers areusing because I what I've heard or whatthe you know the general perception isis that that most of their proprietaryTPU transactions are their ownapplications.>> Yep. But that's probably where most oftheir tokens are being processed anywaysat this point, Bill, right? Liketranscribing YouTube videos and you knowall in Google Meet you can turn it onand all the searches.>> I I was just inferring in your question,maybe I shouldn't have been that thatthey'll have an advantage for GoogleCloud. And in in order for that to betrue, they need to to have thiscrossover moment. One quick thing, Brad,on on Open AI, you know, I'm I've I'vebeen writing a book, which I've talkedabout frequently, and I've been quite uh
00:42:39although I guess there's some privacythings now you need to be worried about,but I've been quite open with Open AIabout the book and doing research, youknow, along the way. it knows atremendous amount about my book rightnow and I can ask follow-up questionswithout having to put the whole bookback in the prompt again um because ofthat. And so I can I would I continue tobelieve that OpenAI'sum most likely chance to long-term umsuccess comes from switching cost andlock in more than it will come fromstaying on the edge of the of the modelrace because I think>> and the pricing and the pricing powerthat comes with that brand dominance,right? No doubt.>> Because the the fact of the matter isyou said you're paying 200 bucks, you'regetting more than $200 in value. I don'tknow what the price is, but I know thatif it was variable, you would pay a hellof a lot more money to use that service.>> I would I would but but but the the thelock in once of these systems starts totruly understand you and have all yourhistoric knowledge um I think theswitching cost will be very high at thatmoment in time.>> Sunonny, back to your question about theTPU and the advantage of that
00:43:54verticalintegration um you know to to Google. Ithink it's it's it's too early to know.What I would tell you is that uh theythey've absolutely made some changes Ithink over the course of the last youknow 3 to four months to accelerate thebusiness. You you've seen the news aboutOpenAI uh leveraging TPUs for some ofthe inference uh you know uh demand thatthat they have. Um ultimately what I'vesaid all along about Google is theremany ways the best position company inthe world but a lot of their advantageright is no longer much of an advantageright and namely that you know they werethe dominant place where the consumerstarted every single query and we knowtoday that's just not true in you knowwhen people are looking for answersBill's book is not in Google bill's bookis in you know chat GBT and that's the>> actually it's in Google doc but butyou're right you're Right. The knowledgeof it, the knowledge of it,>> the knowledge of it and the interactionand the the token generation. So my onlypoint is this,>> the battle for the consumer isultimately where the value occurs,Sunny, not who runs, you know, whathardware. And so Google's dominance, theit's been the greatest b business in
00:45:09thehistory of capitalism for 20 yearsbecause they owned the consumer. Theyowned the verb in something that wasextraordinarily high margin. And all Iwould say is the first real threat in 20years came about because in the chat GBTmoment it's continued to accelerate. Ithink chat GBT will cross a billionweeklyslike maybe this year. Um you knowprobably this year I would guess. Um,and so that to me has always been thecase, right, for open AI. And when youlook at the rest of these frontier labs,the case you've made throughout thispod, about seven of these models in inin in China, being able to open source,distill off one another, drive upintelligence, drive down cost. What thattells me is the model layer is beingincreasingly commoditizedand that there's not going to be a lotof intrinsic value in in in thatintelligence layer, that operatinglayer. You're going to have to buildapplications that guys like Bill Gurleyand and you and I are using every day.And that's where the battle is on theconsumer side. You're going to have theexact same battle when it comes tocoding agents and general enterpriseapplications. And I've said there, Ithink it's going to be more of a uh uhyou know, heterogeneous world. I thinkthey're going to be lots of players
00:46:24thatcompete. I think the margins will belower. Yeah.>> Uh in that world, but it may very wellbe that the tide is going up so muchhere, right? The whole world istransforming so much around this thatyou're still going to have lots ofplayers who do incredibly well. I haveto say I'm surprised if you would havetold any of us a year ago or 18 monthsago, right, that the combined enterprisevalue of Open AI and Anthropic togetherwould be over half a trillion dollars.And you throw X.AI in there, it' be atrillion dollars, you know, across thethe three of them roughly. um it'sbigger and faster and the compute demandis higher than any of us I thinkanticipated. Sunny, hopefully we cankeep you on here for a bit. We're justgoing to wrap up with a um you know atopic that I think has dominated reallythe conversation in the markets over thecourse of last year and that's beenabout tariffs and kind of the reorderingof global trade. Bill, you I know youhad some some questions, some thoughtsabout it and I'm I'm happy to dig in andtalk about it as well. Well, I mean, Iwould I would really just love to hearfrom you, Brad. the the the markets gotvery nervous when the um what was itliberation day when when the
00:47:40kind ofunpredictabililityof how big some of the numbers were andwhat that might mean and whether we werewalking away from the notion ofcomparative advantage and and I thinkthe markets got spooked and you know Ithink you turn around and look at wherethe markets are today and and we'vereally gone through an evolution of howthe Wall Street is interpreting the boththe initial um launch of the tariffs andthe reality of where they're landing.So, how would you how would you describethat? And and why do you think themarkets are getting um very comfortablewith the where where they're landing? Imean, not only comfortable, we're atall-time highs,>> you know, and and April 2nd, I was uhgoing, you know, on CNBC saying thenuclear Navaro was going to be adisaster and I'm out, right? And that'sthis the the the amazing thing aboutthis administration is there's really uhyou know it's a team of rivals withinthe White House and you had Besset andLutnik who were basically outlining thisplan for you know let's call it 10 to 15to 20% tariffs across the board thatwould amount to about 300 billion intotal tariffs up from 75
00:48:56billion uh in2024. But you had Navaro who wasbasically saying we're going to replacethe Internal Revenue Service. We'regoing to get rid of the income tax andwe're going to have two trillion oftariffs. Okay? And I was very clear andI think the market was very clear. Weall voted with our wallets and we saiduntil the president tells us whetherit's door one or door two, we're out.>> The market shot first and askedquestions later. And that's where yousaw that huge draw down in the market.The NASDAQ was down 21%right at its trough this year. Now theNASDAQ's up over 10. It's a 30% move inabout 60 or 70 days, which isextraordinary even by the historicalpatterns that we've seen over the courseof the last 5 years. But let me back uphere for a second.>> I think the consensus view of alleconomists, right, 90% of economists isthat tariffs are going to be bad.They're going to be a tax that gets paidby the US consumer. There was a smallgroup led by you know Scott Bessant andKevin Hassid at the National EconomicCouncil that said no it's actually goingto be different this time and the reasonit's going to be different this time istheir theory uh
00:50:11argued was that theworld had become dependent uponexporting to the United States so thatthe total trade deficit to the UnitedStates of goods and services was about915 billion last last year a $1.2 2trillion goods deficitand that basically meant that China wasselling a lot more to the United Statesthan than they were buying of US goodsand and so what Bessant and Hassetpostulated was that these countries haveno choice if we impose a tariff on themso long as it's not draconian 70% 80%what Navaro was talking about if weimpose a 15% tariff on them they have toeat it the producers have to eat itbecause otherwise they're going end uplaying off millions of people inVietnam, in China, in these countries.And politically, they can't afford tolay these folks off. So that was theirtheory of the case. The consensuseconomist said, "No way is that true.You're going to see massive inflation."But what have you seen? You have notseen the in the the inflation percolatethrough. I will caveat this by sayingyet.>> Okay. So here we are in July. The e theconsensus economist said it
00:51:26would havealready happened and the nationaleconomic council was out with a paperlast week that deconstructed core PCE.So that's the uh uh uh what the the bestproxy the Fed watches for inflationsince the start of the year. And itshowed this was really interesting.Import prices have been going up at aslower rate than domestically producedgoods. Okay, so this is the exactopposite of what you would have expectedfrom tariffs. Of course, you would haveexpected the imported prices would havebeen going up more than domestic uhgoods. And so we'll show these charts umand we'll we'll put the link to thispaper. People ought to take a look atthat. But to me, um, when I look at thepresident's, uh, the deals he's landing,the deal he just got announced yesterdaywith the European Union, 15% on allgoods coming from Europe, 0% on US goodsgoing to Europe. So, an opening up ofEurope markets, paying us 15% and on topof that getting commitments like $750billion, almost a trillion dollars ofenergy purchases from the US. Or look atJapan, which they announced last week.Another huge market. Again, similar.
00:52:42They're going to pay tariffs to theUnited States. No tariffs imposed on theUnited States. And they're going toinvest $550 billion into the US in a waythe president gets to direct. So, Iwould I would I just think we need togive the president credit where creditis due. Everybody said this was going tolead to retaliation, to trade wars, wasgoing to be disastrous for the US. Andall we've seen so far is deals, deals,deals, deals. And I have to say, if thiswas the CEO of one of our companies,right? Let's say we had a board meetingat the start of the year and he outlinedthese plans and we said, "Hey, we'rereally nervous about this. You know,this is a high-risk, highrewardstrategy. it's either going to backfireand we're going to fire you or it'sgoing to work really well and we giveyou a bonus. If we're measuring themhalfway through the year, I would saythat he's in line for a bonus based uponthe trillions of dollars that are goingto be coming into the United States andthe fact that we now have a 300 to a$350 billion recurring,you know, stream of revenues intoTreasury in the form of these tariffswhich are being paid. and and and Ithink Bessant said last week in
00:53:58themonth of June we had our first surplusin the United States monthly surplussince 2015>> right because of the tariff revenuesthat are coming in. So I will say this Iwas on the fence I knew the nucleartariffs the trillion or two trillion Iknew that was a disaster. I said if welanded the plane where Bessant wanted tocome in at 300 billion, I thought therewas a decent chance that those pricescould be passed on in the home countriesand so far it looks like that's thecase.>> Do you have any concerns? What's the uhanything you're watching out for?Well, I think the the number one thingis core PCE h did bottom last year andit started to tick up and so we have tokeep our eye on inflation. And ofcourse, I think it's almost impossibleto conceive that we would have totallyreordered the entire global tradingsystem on the first pass with zeromistakes. So, we're going to have somesome goods and some products that, youknow, consumers are US consumers aregoing to end up paying the taxes andwe're going to have to go back and fixsome of these things. But, I will saythat it's turning out massively betterthan consensus criticisms. I mean,remember Larry Summers at the CO2 event.>> I mean, this was just a few months agoand he was
00:55:13saying this is the biggesteconomic disaster of of his career. AndI I don't think you can describe it thatway. The markets are the voting machineis telling you and these are a lot ofsophisticated investors. The votingmachine is telling you that noretaliations, no trade wars, all thesedeals getting done um uh works for theUS economy. And I will tell you that theAtlanta now Fed tracker which tracksreal-time GDP has now ticked up back to3%. So after going down a lot in April,it's bounced way back up. So economicactivity uh appears to be going up. Andthen one final thing here,remember one of the key reasons fordoing this, right? Wasn't just becausewe were, you know, there was I mean theEU conceded in the trade negotiationsthat they that our relationship wasunfairly balanced in the direction ofthe EU. I mean they said that's thestarting point. So we have to rebalanceit. But on top of that, a key reason fordoing this was to support the domesticproduction of critical nationalindustries and to make our supply chainsmore resilient chips, data centers,energy production. At Altimter, we justled the series A in a company, I don'teven know if we've announced it, butI'll announce it here, which is anall-American
00:56:28producer of rare earthmagnets. Right? So these are now viableinvestments because of the tariffs andthe resolve of the government toreonshore these critical supply chains.So I mean that's a huge benefit that youwould be willing to pay something forbut we're getting the benefit and on topof that we're getting paid.>> Yeah. Can I add one thing guys like noton the economic side but like you knowrunning the supply chain you know Grockand look we're we're fortunate that likethe majority of our supply chain is US-ccentric including our chips but>> um you know we still have small discreetcomponents and Brad exactly what youwere saying is happening is that the theproducers the manufacturers of thosethings are coming and you're negotiatingand we've had pretty you knowsignificant negotiations with thosefolks and that that was I think like notlike you said wasn't anticipated. Theother thing is it hasn't been um static.The one thing again I'll go back to thisadministration they moved like we'vewe've seen you know our tariff schedulemove around probably six to eight timessince this all has started because theykeep evaluating they understand theylisten and I think that's also afunction of how this administrationoperates and we want to give them creditfor that like people can go and sharewith them hey like these things are notavailable we can't replace them rightaway and I think that that's also
00:57:43helping with that last thing you saidwith the investment you're making iscompanies get establishedon shore to take advantage of what thesetariffs are causing.>> Sunny, has your supply chain planningand that that dynamic nature, is thatstarted to settle down? Do you see this,you know, are we reaching the end of thetariff negotiation such that everythingcan kind of settle down and operate?>> It's gone from week to week or even dayto day when it first started and tryingto figure out what happened to like nowwe're looking at it monthly. So, it'sdefinitely settling. And Brad, we stillhave a big thing looming with a Chinadiscussion, correct?>> Yeah. You know, listen, so we landedEurope, we've landed Japan, we're goingto have, you know, the long uh the longlist is is is going to be coming out,but when you look at our big tradingpart partners, the EU, we do, I think,$900 billion of trade with them a year.With China, it's about 600 billion, butwe have about a $300 billion goods tradedeficit with both Europe and China. Sothey were the the the two big ones. UmChina is the big enchilada because it'snot just trade with China, right? It'sstrategic. It's national security andits trade. Um and it's the AI race. We
00:58:58know that, you know, the rare earth banon Chinese magnets was devastating to USindustry. We know the retaliation that,you know, where where H20s were cut offin terms of Nvidia's chips back toChina. reading the tea leaves. I'm goingto I'll go out on a limb and I'll say umthe consensus still believes like thatthe China thing is going to be a problemor that it will be small. I think thispresident wants to do the biggest dealever done with China. I don't think he'sdogmatic at all. I don't think he's somebig China hawk. He said at the AI summitlast week, I'm a deal junkie. I like todo deals. You can't be the biggest dealmaker in the world without doing a bigdeal with China.>> All right.>> Right. And I and I, you know, if youjust look at today, the Chinese uhreciprocated in a way I think the the USgovernment was looking for. They said,"Hey, we'll postpone all of ourretaliatory tariffs." They've invitedthe president to China. The presidenthas suggested he's going to go to Chinathe first week of September or sometimebetween September and November. I thinkthere is a very very big deal that'sgoing to get done with China that'sgoing to reorient the relationship in abig way. And let me just like teasethis. The president said earlier thisyear something that
01:00:13caught all of ourattention. He said, "You know, if Icould wave a magic wand, I would cut thedefense budget in half for the UnitedStates, for China, and for Russia."We've never heard a US president inhistory utter anything like that.>> Be amazing. That is a what I would callan extraordinarily flexible mindset. Andif you go into this deal negotiationwith China with that sort of flexiblemindset, I think it could include all ofthe above. Rare earth chips, maybe evenmilitary cooperation, certainly arebalancing of trade. I think China's,Listen, we entered the year with Chinapaying 15% in tariffs. That waspre-Trump. They were paying 15% to theUnited States in tariffs. So I don'tthink we're going below 15%. I thinkthey're going to pay>> that in in Trump for>> Trump won and Trump won, right? So Ithink that they will continue to pay atleast 15%, but I think it's going to bemuch more structured, much more nuanced.You know, Besson's been very clear Chinahas to rebalance to domestic consumptionand away from an export economy that'sreally uh sticking it to the US in termsof the trades def trade deficit. I thinkChina gets that. I think they want thatfor their own
01:01:28country. I think they'rewilling to do that. I think the UnitedStates understands that can't happenovernight. It has to happen over aperiod of years. I think that there'sgoing to be a big Chinese deal donebefore the year's out.>> So So let let's close with this, Brad.You've you've often on the pod um beenwilling to speak about your your owntemperature for the US markets and andand where you whether you're net long ornet short. and you>> you've on this podcast been moreenthusiastic than I've ever seen youboth about uh about AI, but but thisthis China theory that you have, Ithink, would would cause the markets torip, if you're correct. Um, but you alsosaid we're at all-time highs, you know,and so you want to buy low and sellhigh. So, where where's your headhanging?>> Uh, we did a pod, I think, around May2nd. Well, we did the pod in March and II I said we're out of the market. Iremember.>> Right. And you and you said you're earlyand liberation day came and we werehappy to be out of the market. On May2nd, I said we're all we're all back inbecause the Bessant consensus has won.It's going to be 300 billion. We'regoing to land the plane. And I outlineda flight path. I said you can land theplane, no inflation, you get rate cutsand
01:02:44you know, and it's kind of off tothe races. And we're up 30%, you know,off of that bottom in the NASDAQ sincethen. So 30%'s a huge move, but whenwhen you when you telescope out, we'reup 10% for the year.>> Okay? And if I had told you guys on dayone of this year, here's what's going tohappen.>> We're going to rebalance global tradeand we're going to land the plane around300 billion. We're going to have theeconomy grow at 3% accelerating. We'regoing to um uh you know have noinflation heading toward rate cuts bythe end of the year.um you know that is the backdrop youknow that I and we're going to have allof this AI demand and acceleratingdemand for AI compute I I would havesaid the market can be up at least 15%for the year I think we've captured alot of the return for the for the yearBill I will tell you this though we seetons of opportunities and so I would saythat we're also bullish on what we seehappening in AI Sunny's going to raise ahuge new round here we're happy to beinvestors with Sunny as well and andit's extraordinary to Watch watch whatSunny's uh No, seriously. He>> wasn't supposed to disclose that. He canedit it out.>> He has He has it, but his face isturning all red.>> Have a good week. And>> thank
01:03:59you, Sunny, for coming on with ustoday. Thank you so much. I know it's>> Thanks for having me. Thank>> Yeah. Thanks, guys.>> Byebye.>> Take care, guys.[Applause][Music]As a reminder to everybody, just ouropinions, not investment advice.