Future Outlook for the AI Ecosystem

Support for the show comes from AtTia, the CRM for teams who set the pace. You know that rep who closes faster than everyone else? They're not quite who they seem While they're walking into every call prepped, writing perfect handoff briefs and follow ups and catching every signal in time They've actually got ATiO working for them around the clock. AtTiO's revenue agents qualify and work every account so they can move at scale with ease It's how startups including Ganola, Mal and Whisper Flow win You can go to Aio. com slash proG and you'll get fifteen percent off your first year. That's ATt o d. com slash Rrov J. Support for the show comes from BCX, the public ticker for private tech. The US stock market started history's greatest wave of wealth creation, from factory workers in Detroit to farmers in Omaha. Anyone could own a piece of the great American companies. But today, our most innovative companies are staying private longer, which means everyday Americans are missing out until now. Introducing VCX, a public ticker for private tech, now available wherever you buy stocks Visit getvcx. com for more info. That's getvcx. com Carefully consider the investment material before investing, including objectives, risks, charges, and expenses, this and other information can be found in the fund's prospectus at getVCX. com. This is a paid sponsorship Goldman Sachs trading floor, in ten minutes or less. Investors and analysts shaare timely analysis on the week's market activity The Markets podcast from Goldman Sachs. Listen now If m is evil then that building is held Welcome to Profperty Markets. I'm Ed Elson. It is july twenty first. Let's check in on yesterday's market vitals. The S andB five hundred and the Tao declined as conflict in the Middle East escalated. The NSdAq was flat. Oil was volatile. moreore on that in a moment. The yield on tenenure treasuries rose BaseX stock hit a new low of one hundred and twenty dollars per share And finally, Warner Brrosothers shares fell nearly four percent A a judge temporarily halted its deal to get acquired by Paramount. The judge said the sale likely violates antitrust laws and scheduled a hearing for next month. What else is happening Conlict over the Strait of Hormuz keeps escalating, and now it is spreading to Saudi Arabia. As of Monday, the U.S. had bombed Iran for nine consecutive nights in response to Tehran's attacks on oil tankers. and Iran had retaliated with strikes across the region. But yesterday Iran's Houthi allies in Yemen declared a naval blockade against Saudi Arabia. This blockade stands to threaten primary way in which Oil has been able to get around the Strait of Hormuz through a Saudi pipeline to the Red Sea These developments immediately shot the price of oil back up. Crude is now about eighty nine dollars a barrel and the national average for a gallon of gasoline has yet again hit four dollars in America up fifteen percent in just the past week. So to discuss what is happening in the Middle East and also it's affecting the price of oil. We are speaking with Matt Smith, director of commodity Research at Kepler Uh Matt Great to have you on the show. a lot happening here If you could just give us your initial reactions and a quick rundown What has unfolded and how is it being reflected in oil prices right now? We're tracking those tankers that are passing through the Strait of Hormuz here. A jobers become increasingly more difficult as there's been different routes to try and traverse the strait. And so what you've essentially got is you've got the Iranian route, which is right at the top kind of the north. And then you have the pre conflict highway, which was straight through the middle and then at the bottom you've got the Omani route, which is the kind of the southern corridor As we've seen escalations increasing here and you've seen some of the tankers being hit, they were passing the Omani route All we're actually seeing now is essentially traffic grinding to a halt again, except for those Iranian tankers and friendlies that are passing the Iranian route. So it's been alulating right over the last few months. You March, April and even into May, the traffic was very, very slow. And then you know just over the last month or so, we've really seen it pick up because of the signing of the memorandum of understanding between the U. S. and Iran. Now that has basically been you know dissolved. and we're seeing an escalation here in attacks. as you mentioned, it's been nine consecutive nights. We'll probably have the tenth today. and so This is causing all prices to to start to kick back higher again just looking at what happened with Saudi Arabia and that blockade, it seems as though Oil ly was figuring out a way to kind of reroute itself away from the Strait of Hamz or around the Strait of Hamuz I guess my question is, to what extent was that successful And to what extent has that now been kind of ock now that we've got this new development It was working pretty successfully and so it was able to reroute about three and a half million barrels a day of Saudi crude across to the Red Sea. And so Saudi was exporting about seven million barrels a day out of the Midd East Gulf prior. So it was able to reroute half of that crude. so that put him in a better situation more than most. So that has definitely helped somewh push the supply shock because all of that crewe was then going across to the Red Sea and was heading into the likes of India, China, South Korea, these countries that were otherwise getting their crew from the Midle East Gulf and it had stopped. And so it was definitely providing some support there and helping in terms support in terms of supply and helping to keep prices in check somewhat Now The Houthis are threatening to do that blockade. They're not actually doing it yet. We're not seeing tankers or anything being hit. But this essentially is the ace that Iran has in its pocket because we've had this escalation that's been happening over the last few months here and some have said, oh know they could close Babam and debad but they've kind of held that until the point where the U. S would essentially start perhaps attacking infrastructure, energy infrastructure, bridges. and that's kind of the point that we've got to. So then it's for Iran to up the ante here and that's basically bringing Babel Mendeb into play. And so it's really just a sign that Essentially, Iran is getting to the point where they've really got nothing left to lose or you know they're just getting to the point where they're willing to do this kind of scorched tactic. And so we'll have to see how this plays out. But the threreat of stopping these flows will definitely have a bullish impact on prices. So the memorandum of understanding has been dissolved. we are now fully at war, striking Iran. on multiple consecutive nights Now they are, as you say playing their Ace card. they are trying to block any of the other ly routes that have been resorted to over the past several months It doesn't look good And we're at eighty nine dollars a barrel Gas in America has gone back up to four dollars Why should we believe that that number is going to come down within the next I don't know, several weeks. Yeah, no, we shouldn't. And actually, what has developed over the last few months here essentially since the beginning of March, when this has happened, is everyone's been watching that oil price and you haven't felt the biggest impact on that oil price. And the reason for that has been a number of different reasons. know China has really come out of the markets. China has just stopped buying oil.p they dial back on their imports by about five, five and a half million barrels per day So that has been hugely helpful. You've also had essentially a lot of these refineries dialing back on their activity. so they haven't taken that crude. And that has largely offset the production loss we've seen from the Middle East. But what that has meant is that the pain has essentially been transferred from the oil price across to the products. And so when you talk about gasine at four dollars a gallon on the national average We see diesel at five dollars breaking above five, and that's going to be really pushing higher because in barrel terms, it's about one hundred and seventy dollars a barrel for a barrel of diesel. And so that's where that pain is coming through is in the products because we're not seeing those produced, whereas the crude market is somewh remains somewhat in balance because of this rerouting and because of this less lack of refining When you look at that number, eighty nine dollars a barrel To you, does that say that investors are feeling optimistic about the current state of affairs or pessimistic I mean does that number Hold Any biases inside of it. One huge bias that it holds is that even if you are bullish on oil markets, you're not going to go and buy a paper barrel because you could have President Trump tweet something in five minutes time and oil prices could drop by ten percent, fifteen percent. So I'm not saying prices are manipulated per se, but they are definitely under the influence here of things other than fundamentals. And so because of that, you've got some that are simply not getting involved in the oil trade and that has been happening for a good number of months here. There's a lack of liquidity there. The flip side of that, that's why I point to the diesel market again is because the U.S. administration is fixated on the oil price, super fixated on the prices at the pump. It's not necessarily paying that much attention or putting that much emphasis at all on diesel prices. And so that's perhaps the least influenced market out of all of the petroleum complex. and that's the one that we're seeing absolutely ripping here. I mean, this is essentially the most important question for the US economy right now, which is what's going to happen to the price of oil? what's going to happen to the price of fuel As we saw in the previous inflation report, it was lower oil prices as a result of the memorandum of understanding that made the number go down more than the previous month. But now we know that whatever pricing was being priced into the market at the time was incorrect because the memorandum of understanding is over We're now back in war Uh Some would argue we continue to be at war the entire time. I won't get into it. It seems as though what we have seen over the past week is going to have material impact on U. S. consumers and the U.S economy. and perhaps that isn't being fully reflected or appreciated or priced in by investors and traders right now. How impactful and how bad do you think it will be going forward Well we could just continue in the status quo, right? in that there's this back and forth between the US and Iran in terms of the attacking of tankers by Iran, the attacking of infrastructure by the US. And then in the background there is talks and whispers of diplomacy, which helps keep oil prices in check here, which in turn helps keep the pump in check, But when we came into this thing, there was expectation you can't close the straight of Hormz for two, three weeks. It will cause like Armageddon. yet here we are four and a half months in. And so it's really realistic to try and consider the scenario. couldould this still be closed in November and December? Yes, there are workarounds. There's medium term plans here to reoute crude, but we really could be just continuing to scramble over the next four, five months here. and that's a reality If that happens, you're not going to be in an environment where prices at the pump and diesel prices are moving lower What kind of price do you think that that would result in if we find ourselves in the same situation that we're in today? And to be clear, I mean It seems like Thank saw making their way through the Strait of Hermuz Um I mean, is that right or is it just zero? There was like a week or so ago or even just before the weekends where you were seeing some getting through there, but the Iranians have turned their focus to targeting those because they were going through the Armani route, they were getting like in the US Naval escort. And so those they're trying to deter any kind of traffic. So the only traffic that we're seeing going through right now is related to Iranian or Iranian And it's just empty tankers. And so But to your point, we were seeing over the last month, this increase in traffic going through, increased confidence, increased risk taking. and that really helped the oil market, not necessarily on the oil price, but in the differentials, you really saw the air being let out of the tres there, the pressure taken out As we saw the stranded cargoos in the Mid East Gulf getting out of there. But then again we it's one step forward two steps back here where we're back to to essentially the doors being shut again. So I'll return to my question, which is if we find ourselves in the same position that we are in today four or five months from now What would you estimate the price of oil will look like. It's got to be higher. You know I willll hold up my hands here and you know, in March, if you'd asked me that question or you may have done, that' been like, you know, all prices should be one hundred twenty thousand hundred three thousand hundred forty dollars if you close the straight for months and months. And so perhaps I've been burned by saying that, right? But But maybe if we're going to be pushing above one hundred here, that doesn't seem unrealistic at all if we're in the same scenario that we're in now in four or five months time If we are in the same scenario that we're in now isn't above one hundred dollars a barrel. What would have had to have happened lower than that. What would what truths would need to exist in the world for oil to not be more than one hundred. Well, we'd be drawing down inventories absolutely everywhere that we could. You'd be having tankers sneaking out of the Strait of Himz. however that was possible, rerouting of crude as much as possible too. But it's just difficult to see a scenario where you're not pushing into triple digits if this is still the situation in four or five months time All right Matt Smith is director of commodity research at Kepler Matt appppreciate your time. Thank you. Thanks, Ed After the break why investors are so worried about the latest Chinese model. And for even more markets insights, you can subscribe to my weekly newsletter simimply put at simply put. profgmedia. com Monday AI agents took over my work. And I absolutely love it. Chasing deadlines, writing status reports, updating stakeholders. agents handle the daily grind now. They live inside Monday dot comot so they see the full picture, My work, my team, the whole company. And I don't have to worry about the data. It's safe, which means I'm free to focus on the big stuff knowing everything runs smoothly in the background. It's completely shifted the way we work. Create your own AI agent in minutes on Monday. com Support for the show comes from SoFI. One of the biggest advantages wealthy people have is the means to hire people who can help them make better financial decisions. But for most of us, there are six different financial apps that don't talk to each other. That's where SFI cooach can help Coach provides step by step chat based guidance for all your finances, helping you pay down debt, save more, spend smarter, and invest. It doesn't just give you advice, it can actually execute the moves for you with your approval. You can also ask important questions such as, can I afford to buy a home this year? How can I improve my credit score? And because it's connected to your linked accounts, the answers are tailored to your specific financial situation. If you ask Shi Coach, what should I do with ten thousand dollars If you're lucky you might need the answer sooner than you think. You'll be entered for a chance to win ten grand. Learn more and enter at sofi d. com slash propG cooach. That's sofi. com slash propG cooach This is a paid advertisement by SopFI teechnologies and is not intended to be financial advice. No purchase necessary. A purchase will not increase your chances of winning Op only to legal residents of the fifty U S slash DC eighteen plus void where prohibited by law. swweepstakes ends at eleven fifty nine PM Eastern standandard time onn september fifteenth, twenty twenty six Excuses are easy. An epic movie night, we don't have enough snacks. Dinner party with the girls 'd have to decorate. Surprise date night? thing to wear But Amazon's prime same day delivery lets you say yes before the moment slips away Try that new popcorn maker, order those cheeky drink glasses, get that new perfume, and turn that I wish we could into an I'm so glad we did Visit ammazon. com slash prime to find millions of items delivered fast. Same day delivery. It's on prime. Available in select areas. Terms apply proroperty markets China just gave Wall Street its second deep seek moment. Chinese startup Moonshot AI released KimiK three, the world's largest open source model on Thursday. On some benchmarks, including front end coding, K three beats the best models from open eye and anthropic. The biggest story may be the price tag. Running K three costs roughly a third of what Anthropic charges for its flagship model and businesses are starting to notice On open Router, a marketplace for AI models, Chinese open weight models now occupy the top five spots by weekly global token usage. The NASDAQ fell about one point a half percent on Friday as U.S tech stocks sold off following the release of Kimy K three. So We wanted to speak with an expert who works hands on with both open and closed models. So joining us is Charlie O'Neill co head of model training at Base ten Charlie, thank you for joining us. so This Kimmy K three model that was just released has Everyone kind of with their hair on fire. We obviously saw the NasDAq rerased one and a half percent, Chip stocks sold off. A lot of people saying that it was a problem. David Sachs, the former AI Zar called the release quote concerning. What do you make of Kimy K three Yeah, I think the big story here is not necessarily Chinese models versus American models. I think the big story here is open source versus closose source So obviously the story we've been sold for the last, you know, several years is that close source is going to continue to dominate The American frontier closose source labs are going to continue to pull ahead and open source will never catch up to that And I think what we're seeing with Kimi with other Chinese models like GLM, GLM cores a very, very big wave. It may not have done the rounds in the way that Kimi did, but it was certainly a great model and even like releases like inkling from thinking machines, which is an American company What we're seeing is that basically the recipe to build these things, there's no secret source. The big labs, they don't have anything that the open source labs don't have And open source is going to continue to improve the capabilities and intelligence of the models they release as we scale up the size of these models and the amount of data and compute that goes into them. And so yes, from one kind of aspect it's concerning that this is like a Chinese model that is leading the charge with this sort of like open source versus closed source debate I think there's really promising signs for the open source ecosystem in general, and I think a lot of people are starting to realize that potentially a better world to end up in where compared to where you have maybe a gopoly with O openpen air anthropic having these models that pull away from everyone else and they dictate all the terms of access and control that intelligence. Just for the uninitiated, what is the difference between an open source model and a closed source model Anthropic, openping E, Google, their flagship models are what we refer to as close source in the sense that I can ask it a question, that question gets sent off over the internet goes to their GPUs which run the model They do the number crunching, and then they send the answer back to me. I never get to touch the moder weights, which you can think of as this big collection of numbers that do a bunch of multipliers to give me my answer Whereas with open source, I can actually download those numbers. Not only can I host out of my own GPU's, I can also do things like continue to train it myself for specific tasks. So it's really about being able to download the actual weight of the model rather than just being able to send a question to it. So you can think of this as like, you know, owning the disc for an Xbox game versus like haaving that Xbox game installed through the cloud on your particular Xbox, I can actually like see the physical disk What would be the pros for developing a closed source model instead of an open source model? Why would open AI and anthropic pursue those methods instead I guess there's two answers here. The first answer is the one that open air and anthropic will tell you, which is that know these things, as they become increasingly intelligent, we have to think very carefully about how they're applied in society. There's obviously real safety concerns, there'sybersecurity concerns, there's biological weapons development concerns. and so We should really think about who we trust to build and control this intelligence. And anthropic and open air's argument is you should trust us. We are the best at developing this intelligence and hence, we should be the ones to dictate how it's used and how it's applied basedically in perpetuity. There should be a very small number of actors who can choose what we do with LMs and intelligence And I think the real argument is that obviously this stuff is so lucrative that if you do manage to prevent anyone else from developing it, you can capture insanely high margins on the tokens that you're producing. Anthropics rumored to have, you, margins north of eighty percent I think like when there is a case for a world where there's only two major players and you end up in a Guopoi is a very real possibility to continue. And I think That's obviously very lucrative to openir anthropics. And so open source is a threat to them in the sense that those margins are going to remain at eighty percent for long. Of course, there are like security concerns. We have to really think carefully about how these things are used. But at the moment, it doesn't seem like open source versus closed source, the intelligence ceiling that we've gotten to hasn't led to any increase concerns around, you know Can I use this model through open source or close source? Like the risk of developing a bioeapon, for instance, is about the same in either case. It seems that there has been kind of a shift towards both Chinese models, but also open source models. Most of these Chinese models are open source or open weight U why is that happen do you think? What is the value proposition that developers are deciding is greater when they use these types of models as opposed to one offered by open eye or anthropic. I think there's developers who have a very You know, in elastic demand for the frontier intelligence that we'll always want to use the most intelligent models. and then there's the ecosystem and the economy in general The way I like to think of it is that for all the economically valuable tasks that we could plausibly use an LM for, there is some intelligence threshold. which below that, it's very difficult to do the task. and above that, you're getting very diminishing returns to having more and more intelligent models. And usually intelligence is correlated with cost So the obvious argument here is that there is margin pressure on all these startups, all these companies, even enterprise now who are doing these particular tasks with LLMs. They've hit the threshold of intelligence probably even a while ago with open source Open source has been accelerating rapidly, and you just don't need a fable or mythos level model in order to do some of these things and you get exactly the same performance if you use a model that's a tenth of the size or even a fiftieth of the size. Post traraining is also really important here because it means you can teach a much smaller model to do one thing really, really well as opposed to taking an off the shelf open source or close source model and trying to prompt engineer your way to doing that task The post training really changes the economics here and of course, you can only post train on open source models because you can actually touch the weights as opposed to closed source. And so I think margin pressure is a big one. Another one is like, Anthropic and open AI, I think are realizing that the recipe is the same amongst all these companies. Like there is no secret source. Yes, there's probably a long tail of optimizations, small optimizations that Anthropic and openp A have that the rest of the ecosystem doesn't have. But their moe is no longer in there them being the only ones who can post train or sorry, train these very, very large multi trillion parameter models Their me now is starting to shift towards okay If we have a little bit of a head startart, what if we try and like hit particular verticals? And so Anthropy is very clearly doing this. They're going after the verticals of, you know, finance and legal open eye as well. And so I think Companies are really feeling this pressure. If you're a startup or a company in legal or finance and you're using LN to do these particular things and you have previously just been an anthropic rapper You've just got some logic calling anthropic models You don't have a distinguishing me between you and Anthropic. and so you're starting to think about, oK, what's the one thing I have that Anthropic doesn't have? And that's a really nice feedback cycle. I have users who love and hate my product for various reasons and they will tell me what they love and hate. and I can use that to improve the intelligence of a model. And again, you do that through training. And the only real way to do that is with open source models I think it's this combination of margin pressure and companies wanting to develop their own mode to protect themselves against their vertical being eaten by these like close source frontier labs. It seems like a big piece of the story for an enterprise for a company that's trying to leverage AI as much as they can. And Alex Cobb talked about this in his video interview with CNBC that has since gone viral is basically just the price Anthropic tokens are expensive, open ee tokens are expensive tokens from Chinese model providers are less expensive So my question is To what extent is there a relationship between Pice and being open source. Why is it that these Chinese models and these other models that aren't, you know, frontier lab models How is it that they can offer a product that does the job pretty well, but at literally a fraction of the cost? The answer to this used to be simply that the Chinese and open source models were much smaller So the big labs are the only ones that had the compute to be able to train the really large models. and of course The scaling laws that we have predict that intelligence increases, but with diminishing returns in model size. And so yes, of course, the big labs had better and bigger models, but you often could use a much smaller model to do the task I think now it's more of a case of like, okay, some of these open source models are actually very large. And I think K three was a massive shifting point because you know previously we'd gone into the just four it into the one trillion parameter model range with the previous Kimi models and deeep Sk very, very recently. But this is almost three trillion parameters. Like this is a big boy. And so now it's much more about, okay, we're really seeing under the hood that The reason that anthropic and open air models are so expensive is because they have great margins, because they were sitting at the frontier and there was no real competitor at the very frontier And again, a lot of this stuff like it is in elastic. you do demand frontier intelligence. But now we're really saying, okay, if we do have know multirillion parameter open source models that any company can host on their own GPUs and can post train and then host on their own GPUs then what that's telling us and a lot of analysis is telling us is that The frontier labs margins are just massive. And so I think the shift that's going to happen now is if there is an alternative that is essentially the same And to ninety nine point nine nine percent of people doing ninety nine point nine nine percent of things is indistinguishable Kimy is ind distinguishable from a fable or a GP five point six Sl, we're just going to see those margins shift. So instead of being eighty percent to the person who train the model, they might up being end up being forty percent and the rest of that margin is going to be distributed one to the consumer and then two to the rest of the ecosystem. So the compute providers and the infence providers are going to be big wins all this competition amongst you know, model trainers It's no longer the case where there's only one or two players who can do this and capture those massive margins. There's going to be much lower margins for model trainers and the rest is going to kind of be spread out amongst the ecosystem It seems to me that these models, Kimy K three and plenty of others that seem to be released practically every month and then we see all these benchmarks where they are performing either inline, with open ay models or outperforming them It seems like that combined with the pricing pressure could literally bring the frontier labs to their knees. If we know that they're already struggling to generate more revenue than they spend. If we know that they're also stacking up billions of dollars in losses and they essentially need to develop pricing power, if they want to get profitable and get cash cash flow positive over the next few years. And that's been open eyes objective It seems like this is exactly the kind of thing that will get in the way of that. Is this Dia to the AI ecosystem, how does this actually play out for the largest names in AI I've obviously been a big advocate and proponent of open source for a long time and want open source to win in some reasonably significant capacity. I think my honest take here is that This isn't the death knell for anthropic opening eye I think ideally and probably most likely now we're going to live in a world where there are a few key core frontier players and then a large diverse ecosystem of open source model providers. The reason I think that is because of kind of the distribution of tasks in the economy that we're currently trying to tackle with LMs and the distribution of tasks in the economy that we should be tackling with LLMs in the next ten years I think what we're going to see is a little bit of a bifurcation. I think tasks that we can currently conceive of as being economically useful and all the jobs that we currently do, we are going to rely more and more on open source to be able to do those things. I think very, very frontier things For instance, science and maths discovery, which you know have a longer they have a lagging period There's a lot there's a bunch of labs like Kiotic Labs who are really looking forward to tackling science over multi decade horizons with LLMs and this new intelligence. I think the frontier labs are going to gain a lot of like, you know economic benefit from tackling those tasks. I just don't think we're going to live in a world where the labs subsume everything. I think we're going to see this like rising tide of intelligence. Open source is probably going to continue to lag behind a little bit to some extent I think those are going to be fairly parallel lines that that go up together. But you know, if you're doing frontier sides and you are planning these very, very long you know, scientific endeavors in order to extract economic value from whatever it is you're doing, you are going to be wanting to using the best intelligence. And I think Anthropic and OAI and other players like them will make significant profits and contribute significant value on those fronts. It's just that it's not the world we thought it was going to be two years ago where they would also get all the value underneath that of like you current GDP and the things that we currently conceive of as economically valuable. And I think that's a good outcome for everybody. No one player wins. I think we still have significant capitalistic pressure to advance the intelligence of these models, and the frontier labs will feel that at the very frontier And then that's going to diffuse throughout the rest of the ecosystem as well. All right, Charlie O'Neill, co head of model training at base ten. Charlie, we appreciate your time. Thank you. Thanks having you.