>> from the heart of where innovation, money, and power collide in silicon valley and beyond, this is bloomberg technology with caroline hyde and ed ludlow. caroline: live from new york, i'm caroline hyde. >> and i am mike sheppard. caroline: we dive into the ripple effects in tech markets and the impact on u.s. chip curves.

we discussed deepseek's ai model . and president trump says microsoft is in talks to acquire the u.s. unit of tiktok. details later this hour but first we checked the markets and claw our way back, not nearly upsetting yesterday's selloff. we are up just a percentage point on the overall nasdaq. nvidia bounces back but 2% after a 17% selloff yesterday, $30 billion being added from the market cap wiped out by 600 billion dollars yesterday. this is surrounding china coming with a powerful, efficient, cheap open-source model. mike: nvidia called deepseek's new model and excellent ai advancement that complies with u.s. technology export controls and that it illustrates how new models can be created. let's bring in ian king to tell

us more about this. this was a rough day yesterday for nvidia. it bore the brunt of the deepseek shock across the markets, yet their statement was curious. tell us your interpretation. >> they could have said nothing. they could have ignored it. they could have said it is dodgy technology that we do not believe in, but they endorsed it and the underpinning message we saw was maybe they are telling the u.s. government export controls have not worked. we should be able to ship whatever kind of chips we want to china because they will do good stuff with whatever they can get hold of. caroline: let's go to the narrative on ship exports not working. we heard from president trump about what the impact could be. >> we are going to be placing tariffs on foreign production of computer chips and pharmaceuticals to return production of these essential

goods to the united states of america. they left us and went to taiwan, where -- which is about 90% of the chip business, and we want them to come back. caroline: a lot of nvidia trips are manufactured in taiwan. what you think the regulatory impact will be, let alone whether inferencing will be the future for gpu demand? >> the expectation before the inauguration was this administration would double down on this type of crackdown of chips and the flow to china and we are getting a different aspect of it, so i think we will have to see how much attention the new administration is placing on this issue, but the expectation is they are not going to let up. >> how much is this complicating their efforts to get their customers to keep buying those premium ships like blackwell and the next generations?

ian: it creates a massive question. if a chinese start up can find tune the models using three year old technology on the cheap, why can't you? if you are microsoft, if you are running these systems, why would you not do it cheaper if you can? that has implications for what nvidia is trying to sell. caroline: so far, the statement is all we get. let's talk about how the tech markets, roiled by china's sheep ai model. here is what some guests had to say. >> i do not think it is about china versus u.s.. it is about closed source versus open source and more and more american developers should leverage the open source and what deepseek had done >> pretty amazing, what they have done, a fraction of interference cost to produce a model like that, but it will unravel based on

spending. >> a healthy dose of skepticism is good. it is needed, but even if you multiply their spend and cost by 10 times, 20 times, this is still an order of magnitude. >> they are not a company trying to compete directly with big u.s. players, so in that sense it was an overreaction to the release of the model. caroline: let's get more reaction. the question for many is will capital expenditure remain for nvidia chips and ai hardware? is that what you were tackling yesterday? >> that is one of the critical questions because this was a surgical selloff in the sense that it is attacking not the case for using ai but the supply chain around ai and all roads on

supply chain back to nvidia. this question of whether companies need to spend this much is the question markets are grappling with, but it is hard to imagine that we are going to see companies massively change spending patterns because of this news. there is still this arms race going on, so that is something we have to watch closely but we may need more than just this news to unravel that. caroline: the unraveling could start when you have earnings expectations as they are and valuations at 41 times future earnings for a name like nvidia. how much are you worried this could have a similarity to what we saw in the.com era? >> that is something we are watching closely. when you're looking at earnings expectations overall, you are seeing celebration of earnings. certainly a continuation of the

powerful earnings we have seen and you couple that with valuations that are extreme. when we do our analysis of valuations, we are looking at the tech sector that is in the 90th percentile. associate and that -- associating that with returns got you can still see valuations growing higher, but that is where you also see some of the more meaningful selloffs. i think a lot of developments over the past few days are more to evaluation argument than anything that relates to technology in the sense that this is big news in technology but the bigger news is there is no room in the narrative to account for anything but world domination and this is allowing us to think about the range of outcomes. mike: to play the movie forward a few years, where do you see this taking us?

is this a parallel to the.com era which eventually after troubles around 2000 set us up for a long period of growth? what about ai? are we in that moment? >> that is the corollary i referenced. a lot of us are pointing back to the.com era. the important element is this is an innovation super cycle. what we saw in that period was massive expansion or growth for the broad market, not just the tech area. if we are looking at trend growth from the 1990's compared to subsequent years, we saw massive expansion, whether that is margins were enterprise value. the internet world had a huge impact on the broader economy and that is what we can expect with ai and this news is part of that in the sense that it says we can do this better and get

deployment more quickly than we had otherwise thought. mike: on the idea of getting cheaper ai out to companies and businesses across the economy faster, what sort of impact do you see and where do you see growth potential? >> i think there is this broad market where you have to understand you cannot perfectly anticipate. we are seeing more conversations around retailers and deployment of -- more quickly with fewer steps along the way. we are also seeing stocks hold up better. software can easily deploy or may more easily deploy ai. that is a possibility, so i think allocations are somewhat endless. the imagination has to be

brought here, so there is some argument for equal weight exposure to capture that range of experiences. i do not think this is something we are going to see in q2 and q3. caroline: i have this crucial point for the here and now with retail buying back into nvidia and institutional's not right now. what about a black swan event or ripple effect? have a listen to what he said. >> beginning of an adjustment of people to reality because now they realize you had a small chip on the glass. caroline: a little chip in the glass of the valuation question, as to whether this is akin to a dotcom crisis. we saw significant selloff.

>> when we are taking a look at nvidia or energy, i think there is more vulnerability because this is getting at the heart of the thesis, that they are the dominant provider of this capability, that all roads lead back to nvidia when it comes to supply and the story is tackling that idea that potentially there are more ways to skin a cat. you can focus on efficiency. the idea of a chip in the glass is important and getting at the heart of nvidia, who whether this is the beginning of the unraveling of ai, that is more extreme or a bridge too far because that unraveling would depend on the use case of ai being attacked and i do not think that is what we are seeing. it is more the supply chain question and at this point i think there is still more we would need to see to suggest companies are going to pull back on capital expenditures in any

meaningful way. mike: thank you. coming up, we will discuss deepseek's open-source approach and what it means for competitors. the chief science officer at hugging face will join us. this is bloomberg. ♪

i earned my degree online at southern new hampshire university. after i graduated, i started a new job. i was finally able to realize my purpose and passion in life. pursuing my degree gave me so many opportunities to grow

don't just think about yourself. think about the lives that you can really change. snhu laid the groundwork. i am doing what i've always wanted to do. if i was back at the beginning, i would choose snhu all over again. (♪♪) caroline: china's deepseek ai model up ended.

what about providers? clearly open-source is impacting ai development. he is the chief science officer at hugging face. is this the huge breakthrough the market thinks it is? >> it is a bit of an overreaction. we have seen a steady increase in open-source models but we also have to be honest. it is a first model. the gap is close now. caroline: let's talk about how derivatives have expanded because you are all about open sores. you are about community and innovation one model can provide to others. from what we understand from the ceo of hugging face, there have been 500 derivatives already created from deepseek. how are you seeing exponential growth of the use of this model? thomas: it is growing.

i was just checking stats. over 600 models already created. more than 3.2 million downloads of these models. that is really the power of open-source. we see more than 30% of downloads today, which china -- it is already starting to build around deepseek, which basically all these organizations are taking this open-source model and testing it on open use cases. we are seeing this ecosystem grow live day by day. mike: we cannot separate the story of deepseek from the geopolitics here. what concerns do you have that a

reaction from washington or other governments might be to restrict open-source use in some fashion as a way of keeping technology out of china's hands? thomas: i would say a way of reframing this story as u.s.-china, but the more general story is the open-source model. it could have come out of almost any country and out expect open-source models to keep coming out of china but we also have european companies. so hopefully we will see a move from a geopolitical interpretation of the story to interpretation of open-source versus closed source and how that is the way to folk -- to foster great technology and communities and more business use cases. mike: as a software creator, he

must have admiration for what deepseek has pulled off. apparently at a lower cost than rivals in the u.s., but i bet you have questions too. what would you like to learn more about what the company did and how? thomas: what is interesting is we already started to see if we can reproduce this model, so i was trying it yesterday and it does really offer what we have seen, which is it is a very powerful model, but we would like to know can you retrain it? can you apply the same to all? we have started a reproduction of the deepseek. thankfully, they shared a lot of details in how they trained the model so i am confidence in the

coming month we will be able to basically understand all the breakthrough that went into making this model. caroline: what does it mean for a closed source openai or anthropic in this moment? thomas: i think it is positive. you saw the reaction of sam altman. for competition in the field, it is something that is a net positive. a lot of -- are open. for openai, google. this model basically is a good thing. the good thing about open-source is you share a lot of information about your models. you lift the whole field up with explaining how you make your breakthrough. caroline: some people do not

want to see everyone rise up. they do not want to see china rise up. is that a false narrative can't even say china is behind the u.s. when there are open-source communities such as yours? >> the open-source community generally do not know any better. today it is a chinese team, but it is just a small team. there are teams like that in many countries. so i think i believe in fair markets. having competition is a good way to make progress together, so in the future we would like to see more actors active in ai. and that is maybe the first step in this direction. >> are we expecting similar innovations coming from europe?

there have been questions and concerns raised in the european union about the level of regulation and that it might be restricting development in ai. >> it is a big discussion. i was at davos last week and the discussion -- i think the teams are really good. the u.k., germany, france -- they basically help train the models. a lot of people are leaving to start their own start ups, someone take is basically the recipe to build a very good quality large language model now is something that is almost accessible to everyone. i think all these people will start their own teams. because you kind of just need a few minions i would be surprised that we do not see small teams coming out of europe.

mike: thomas wolf, chief science officer at hugging face, thanks for joining us. president trump says microsoft is in talks to acquire tiktok u.s. operations. details coming up next. this is bloomberg. ♪ at morgan stanley, old school hard work meets bold new thinking. to help you see untapped possibilities

and relentlessly work with you to make them real.

mike: microsoft is in talks to acquire the u.s. arm of tiktok according to president trump last night who did not elaborate further. microsoft declined to comment. let's bring in bloomberg's balance of power more. is this another case of trump trying to make a deal happen? he casts himself as the consummate dealmaker. we are short on details. walk us through what more we

know beyond last night. >> details are limited. he was asked by a reporter if microsoft was in talks to acquire tiktok and the president said i will say yes. microsoft is not commenting on this either. during the first trump administration in 2020, bloomberg did report microsoft was looking at acquiring tiktok operations back then, when the first pressure around divesting over security concerns was arriving. oracle was reportedly looking at it, too. the stargate project was being announced, donald trump said he would be open to elon musk buying tiktok. we know there are other players involved as well who have made bids and it seems that is what donald trump's preference is, a bidding war he said he thinks bidding wars result in the best deals. as for what that ultimately

looks like, that remains unclear. what he was clear on is that he does not want china involved in whatever happens here. china is going to have some say in that. caroline: will the u.s. government have involvement? >> he has talked about this 50% ownership structure sharing equity with united states, but it remains unclear whether that could be done by the april 4 deadline. there's also a question around whether antitrust concerns would be raised by individuals or companies like microsoft requiring operations of this size, which could be tens of billions of dollars. 170 million users in the u.s. while donald trump and people he has instilled may have some alignment on this, it still could face scrutiny from big tech skeptical lawmakers on capitol hill. caroline: thanks so much.

coming up, we are going to dive back into how the deepseek model innovation could affect private sector's generative ai adoption. that is next. this is bloomberg technology. ♪

the way i approach work post fatherhood, has really trying to understand the generation that we're building devices for. here in the comcast family, we're building an integrated in-home wifi solution for millions of families like my own. in the average household, there are dozens of connected devices. connectivity is a big part of my boys' lives. it brings people together in meaningful ways.

caroline: welcome back to bloomberg technology. caroline:coming up on mike: let's discuss the impact deepseek is having on u.s. tech companies. coming up on thanks for all your reporting today. there is a lot of soul-searching going on at silicon valley over the past 48 hours. tell us what is happening. did they set up more rooms? what is conversation like? >> the top aides are try to figure out how the chinese start up deepseek was able to catch up

so quickly. you have their latest model really excluding on the scene in the past week and everyone is astounded by how competitive it is, actually leading by some metrics some ai models that have taken western companies years and money to build. >> we can all debate how much was building off of western technology, but will cap extra mean the same? and alternately what it means in terms of reducing the cost of their models. what do you think the ripple effects are? >> people are starting to question whether we need these astronomical budgets for building the most advanced ai models and whether because deepseek was able to make gains and how efficiently they used the computing power if those gains can be applied now to u.s. companies and questioning why u.s. companies did not come up

with that first. caroline: it is a fascinating take. we appreciate it. now we go to assess what president trump had to say about the competition from deepseek. >> if it is fact and true, and nobody knows if it is, i view that as positive because you will not be spending as much and you will get the same result. the release of deepseek ai from a chinese company should be a wake-up call for are industries that we need to be laser focused on competing to win. caroline: let's talk about that competition. i have a feeling a lot are rushing to understand how they can use this cheap and innovative ai model. >> i think 2025 is the year for enterprises to get roi from ai efforts and the roi calculation

just changed so that is the headline for businesses, which is use cases that were not possible a few weeks ago or even last week are now possible and i think we are about to see an explosion of ai experimentation and value delivery to organizations and enterprises. caroline: have this organizations been reticent because of where it was born in china? >> i have the same question and dove into this myself. i think it comes down to how you use these models. there are lots of ways to use open source models and ways to set up to be more secure and run locally on your own environment. in many ways is almost easier with open than closed source models, so everything comes down to the mentation of these models. that is in many ways why they are coming to help navigate the

landscape across these different models, what our best to use and when and for what types of use cases and how to set it up in optimal ways. >> one of the questions that has come up is the one on spending. how is that factoring into the way you are guiding people through the ai landscape now? is it a different message you are having to deliver about efficiency and bringing a product to market for less? >> the short answer is yes. i compare it to a highway. we have just added multiple lanes to the highway. when you do that, traffic does not go down. it only increases because driving and going on road trips gets more attractive. the same is true now for ai, so i think we are about to see an explosion of activity and that is across companies and enterprises and to appoint on

efficiency and product innovation we were going -- i think we are going to see proliferation across these use cases. that is not even include these large hyperscalers. nothing changes there. if anything, competition has increased dramatically. mike: the open-source model, is this something feasible to be scaled at large enterprises and big corporations that may have security considerations and other factors to bring in that might make them question whether open-source is the right choice for them? >> there are a lot of considerations that go into right -- what is the right model selection. it is clear we have entered an era -- i always thought we were in this era, which is a multi model world. businesses need to be able to

build to swap between models seamlessly and optimize for performance, accuracy, and cost. that means they also need the drails and infrastructure to set things up in proper ways. they also need evaluation frameworks to compare and contrast across models so they know what models to use, so i think while there might be some hesitance we have done lots of development on open source models and the considerations still come down to what is that company trying to achieve and what are the things we are optimizing for. that is what determines the model. caroline: i want to go back to the paradox piece, that the more ai innovation there is, the more innovation. but also the same amount of compute. that seems to be the silver

lining that nvidia is trying to see. you are still going to need gp use for training. >> we have not cracked the code on the hardware needed. compute is still needed at high levels. there is no way even with these new models -- we have not found the answers yet. we are not stopping and that is how and why you are going to see continued investment here, that we are in a continuous innovation cycle and that drives utilization and the question is for watch? the other piece is ultimately consumers are the biggest winners here and that is the most exciting story. there's a lot of fear, a lot of china versus u.s. dynamics. the reality is that competition is good and what i'm hearing from u.s. companies is nothing

has blown their socks off. this is innovation they feel they already have and the difference is that there has been optimization for open source and cost in ways they have not done, so i think we are about to see a lot of fast follow and sam altman signaled as much last night. caroline: is he going to drop the cost? >> definitely. i think we are likely to see a cost curve decrease that is dramatic from them but also a latency decrease and that is how they will play the game, to try to one up completely on both dimensions. mike: thank you. caroline: let's check on these markets because these are the questions investors are on at the moment. what does it mean for competitors in the generative ai

model space and what does it mean for software that we are clawing back our losses from yesterday? we will talk more about llama there. bitcoin up .9%. the individual movers they have to keep trained on have been some of the contributors. we have seen apple doing well. nvidia bounces back hardly at all off a 17% sell off. it lost $600 billion yesterday. earnings are almost upon us. we will have the likes of microsoft as well. coming up, questions on his approach to efficiency because apparently he is good at it. this is bloomberg technology. ♪

caroline: check out the bloomberg technology podcast. this is bloomberg. elon musk's x has announced a new partnership with visa, enabling digital wallets on its app and website in what marks the first step for the company to create the so-called everything app as bankers are looking to offload $3 billion of debt. apollo is said to be among asset managers looking to purchase a portion of the debt. interesting sweeteners in that offering. staying with elon musk, many are scrutinizing his approach to efficiency with his own companies as he gears up to cut

spending and waste in the u.s. government with the doge initiative. how good is elon musk at efficiency in the private sector? >> it is worth taking a look at his track record and maybe be open to the idea that there is more similarities in how he has run his companies and his criticism of the government is characterized that he is possibly let on. it has been the case that tesla for example took a good decade before it was making its investors any money and there was plenty of waste even according to him or inefficiency according to him within the company that was sort of allowed to fester for some time before he turned things around, so no one is disputing that tesla and

spacex are not quite efficient and now quite lucrative companies for their shareholders, but you can find your fair share of inefficiency or waste in his own companies and even on his own earnings calls him sort of talking about ways in which is companies have not necessarily been as efficient as you would expect from somebody who is about to run d.o.g.e.. mike: your report during -- your report pointed to some visible examples of maybe a profligate approach that had to have been expensive. how does that approach translate to government, where the consequences and stakeholders are different? >> i think what was pointed out in this story is there is a

willingness to sort of allow for some inefficiency if it means other deliverables. within the government, i think the answer we are going to get is something different in terms of not so much forgiveness of we were inefficient but we delivered x or y. what we are seeing is this relatively new x account highlighting ways the government is sort of blowing taxpayer dollars and all the ways in which his team are coming in and sort of writing those wrongs -- righting those wrongs in his view appeared he is not shown a lot of willingness to cut slack in terms of trying different things or trying things that are not going well as he has been in his sort of time as ceo of companies.

mike: thank you. turning to crypto, the deepseek sell off posed a threat across markets including crypto, with bitcoin seeing its biggest drop in more than a month as president trump signed an executive order last week calling for creation of a white house advisory group on digital currencies. joining us now to discuss all of this is the crucible capital group general partner and founder. we have to get right to it. we were talking about elon musk's department of government efficiency. one of the biggest questions facing our industry is regulation. what are you looking for the trump team to clear away? >> i am going to say something controversial. the lack of clarity on regulatory policy in many ways i think was actually constructive for the crypto industry in the sense that there was not a lot

of external pressure. clarity i think can be challenging for markets. there are bright spots. this will now allow banks to hold crypto on their balance sheet and make it easier for institutions to hold crypto. that is great, but clarity also creates more perspective on what will work and what will not. so far, the view has been number go up. we sometimes joke it is number go up technology and i think what is happening in week one of the trump administration is this clarity is helping people and markets understand we are not all going to make it. there are going to be winners and losers and this administration has made it clear they are content picking winners and losers. caroline: are ai coins going to be winners of the future?

they have taken a brutal hit in the current narrative. >> the ai crypto narrative is confusing. how does crypto make ai better or safer? the jury is out there. and the reverse is how does ai make crypto better or safer? there we have used ai to create new casinos and crypto in the form of online accounts tied to llm's creating their own clients and launching their own coins. to me, that is not interesting. what i'm looking at is how can crypto and opportunities around aggregation and financial is asian -- financialization make crypto and ai better than i'm looking at what apple is doing. obviously the big story around deepseek is what is going to happen to all of this energy and compute capex and some of the

early efforts we are seeing and crypto to aggregate the long tail of compute into these marketplaces where you can pay in stablecoins is interesting. caroline: i am interested in what the regulatory spirits have meant underlying what has felt like number go up on bitcoin. what has the trading told you? >> if we look at markets, what is interesting is you see a lot of sentiment online and it is easy to say something but if you want to know what people are truly thinking you have to look at markets and flows. the biggest week in flows we had into crypto etf's, $4 billion the week that trump won the election. last week, with all the regulatory clarity or setting up directions with the trump administration and executive orders being signed, $2 billion

in inflows, so the expectation is always better than reality. markets are not reacting positively. the bitcoin futures contract had the biggest drop in open interest in is trading history yesterday, so it is clear on crypto traders and markets are feeling overextended like they did on the big ai names and there has been pullback. the question is how much of that is going to come back with clarity and how much is going to come back when we see reality come -- catch up with hype. caroline: so good to have you on. coming up, we will take a look at what to expect from meta and all the narrative around those open-source llm's. ♪his is bloomberg technology.

mike: meta is set to report fourth-quarter earnings after the bell tomorrow. kurt wagner joins us now. meta somehow escaped the big sell off from deepseek yesterday. why is that and what does that tell us going into the results tomorrow? kurt: it might be two things. last week, they announced their

big plan for the year, all the spending they were going to do on ai. they announced threads was going to start running ads. with some of the biggest news they have planned for 2025, so maybe some people saw less uncertainty with meta here in the second is this deepseek stuff in a way validates the ai strategy that meta has been pushing toward. mark zuckerberg has an arguing for more than a year that where this was going was open-source and that is where meta was going to go as well, to create an open source model other people could build on. that is what deepseek is building, so there is more competition for meta here but the strategy they were employing might be seen as the best way forward, so perhaps some investors were less spooked because they thought the road ahead made more sense than others. caroline: they are thinking

maybe the ai models becoming cheaper will push on the overall profitability. is that we have to here? the return on ai investment as well as his investment going forward? >> this has been one of the biggest questions for these tech companies. when is this investment going to pay off? in 2024, meta did a better job at conveying where ai was impacting the product not only in making ads more efficient but in dispersing their ai assistant , having the smart glasses. they had tangible products they could show people. the massive investment in 2025, it will be even more important to show that, but given what they did in 2024 it seems like the street knows what to expect from them so we will see if they are able to continue that

tomorrow. caroline: shares at a record high. that does it for this edition of bloomberg technology. you do not want to forget our podcast. you can find it on the terminal as well as online. this is bloomberg technology. ♪

the way i approach work post fatherhood, has really trying to understand the generation that we're building devices for. here in the comcast family, we're building an integrated in-home wifi solution

for millions of families like my own. in the average household, there are dozens of connected devices. connectivity is a big part of my boys' lives. it brings people together in meaningful ways.

>> welcome to bloomberg's crypto. we will look at the people, transactions, and technology shaping the world of -- global markets get rocked by ai jitters. assets are now recovering, raising fresh questions about the outlook for crypto in 2025. the sec is set to launch a