Bloomberg Audio Studios, podcasts, radio news. Bloomberg Tech is a live from coast to coast with Carolline Hide in New York and Eva Low in San Francisco.
This is Bloomberg Tech coming up. Meta agrees to buy AMD chips and data center gear and a deal worth tens of billions of dollars.
Plus AI whiplash in markets, It's erupts again as investors navigate the disruptive power of agentic tools and unveil by Anthropic and Warner Brothers.
Discovery is considering a new takeover bid from paramounts Guidearce, with its board set to review the proposal and respond.
First, we check in on these markets that bounce back after yesterday's sell off again, Anthropic Front and Center, the latest AI tools that have been battering software and now adding some fuel to them. We'll get into that story a little bit later, but at the moment, that's that one hundred up, my nine tensven percent reprieve. Remember also consumer confidence from a macro perspective coming in better than anticipated. I give a bitcoin, though does nothing better than anticipated with.
This particular asset.
We're currently at sixty three, nine hundred and ninety three. The pressure still there, D But what have you got?
Let's get to our top story. Shares of AMD pushing higher. Meta flat had been slightly softer. At one point in the session, AMD was up about nine percent and on track for its best day since November. This is a six gigawa deal. Metas agreed to buy AMD chips and data center supplies in a deal worth double digit billions per gigawat. This is the social media giant looks to
prioritize it's AI ambitions. Here to break it down, Bloomberg's Ian King on the chip side, Riley Griffin on the Meta side, around the table, Ian, I'm going to start with you the basics of the deal. Six gigawatts over time AMD accelerators, but also other gear. It's a big deal for them.
Yeah, it's a real endorsement from one of the biggest buyers of this kind of equipment. Obviously, they're playing catch up with the in Video who had their own deal with Meta last week. So this is a real strong affirmation and it also includes a share warrants as well that are going to Meta.
Riley, this is all about metacompute. Talk us through just how large the scale is for demand for this sort of infstructure right now.
It's insatiable demand from Meta. Mark Zuckerberg last month announced Meta Compute, with ambitions to get hundreds of gigawatts to fuel its data centers and ultimately reach that super intelligence score where AI can outpace human intelligence. And this is just the latest in a frenzy of deals. Again, we're anticipating one hundred and thirty five billion in capex this year, but now we know the spending won't stop.
The warrants part is very interesting. It's very similar to what AMD did with open Ai. There are operational and financial milestones in both directions in order for Meta to potentially become a very big shareholder AMD. Could you explain that in.
Yeah, the way to look at it, Lisas he said, Look, this basically ties as closer to Meta. They get these shares if we're in a very good position. I mean, the last chalm I believe is about six hundred dollars there. You know yesterday they closed at less than two hundred, so we've got a room to go over before that becomes a reality. But you know, her point was that this kind of locks us together and tightens our relationship.
It was really interesting to read through the materials in our reporting. Riley, like AMD has played this card before. The emphasis is on inference. So m I four fifty generation of accelerator. The engineering teams will work together, but the inference part. Is there anything unique about this different to what Meta is doing with nvidio because we had a very similar deal very recently.
It's a really good question. Not only is it similar to what Meta is doing with in a Video, but Meta also has its own internal pipeline of custom chips that it's building for AI purposes. What we heard yesterday is that they see different applications, different workloads here being supported by all three of those verticals, and so they're trying to diversify as they pursue this massive scale in terms of compute.
In take us back to in video because we've got their earnings coming up tomorrow. And how much does this show that there is anxiety that companies are diversifying looking at MD, looking their own in house chips, or actually does it still not a concerned in video is still dominant?
Yeah.
I mean at this point all we can do is say, look at the numbers, and the numbers assuming in video comes in and executes on what Wall Street thinks it's going to do. The numbers say there's no anxiety here, or boats are floating. Everybody is benefiting from the same massive spending in equal measure. I guess what would change that would be if in Vidia does not hit those targets, does not exceed those targets, that I think would change the conversation considerably.
And Riley, the conversation is nuanced when it comes to capital expenditure. Many would say, yes, a lot of it is about the chips, it is about the day centers, but it's also about the power. Can we understand really how much Meta is having to focus in on where they can spend on GPUs or their own homegrown chips, and where they do it globally as well.
So we asked yesterday where they thought they would deploy these chips, and they couldn't specify which data centers. We know that some of their biggest projects are targeting five gigawatts right, but they need to get certain regulatory approvals and they need energy companies to be able to deliver there on the ground. So as for the merging of the energy and the compute here time will tell as we home in on where those chips will actually go.
It's probably time for a bit of a reality check in, and I'm hoping that you'll provide it for us. Like in any given quarter, right now, AMD is going to do ten billion dollars of revenue all told, every single segment in Vidia's data center business? Is it more than fifty billion dollars a quarter? Yes, that includes networking. But they're now one and the same, right, you can't have chips without the other. How do we know if this is evidence that MD's catching up.
I mean again, you have to go back to the percentages, the growth percentages. At the moment, both of them would say, you know, this isn't about competition. This is about there's so much demand a company like Meta, which makes his own chips, is buying from both of us. That would be the answer. But ultimately, when the market slows down, and you can be the judge of when that happens, then we'll see the market share shift, because then it'll really matter.
Meta is not a hyperscaler, but it operates its own data centers at hyper scale, if you know what I mean. I'm thinking back to when Mark Zackobog was sat next to the president six hundred billion dollars over the next few years, and then he's also talked about this idea that they might misspend two hundred billion dollars here or there, but that's no bad thing. He's front loading purchases. Just explain his thesis a bit.
Yeah, His strategy is just that he's used the language of front loading capacity. They're going to try to get as much as they can gobble it up while there's still availability. And this, he says, could be applied not just for AI purposes, but for that core social media business, which still drives more than ninety eight percent of their revenue. Right so they see applications across the board and they're not worried about getting too much in the interim.
In why do is AM do you have to keep giving it shares away?
Again, we asked Lisa about this and she said that this is Lisa to the CEO, and she said, this is different than the open air idea. The open air ideal is a slightly different arrangement. That is a company obviously that is seeking liquidity. Meta clearly does not need liquidity. This is just a way of showing Meta's commitment to what they're doing. You've got to remember that, you know, AMD is only a couple of generations into being any
type of presence at all in this market. So if a big buyer of this stuff comes along and says, oh, we like your gear, oh guess what, we also like you as a company, like your prospects, and that is somehow a bigger validation than just a straight purchase agreement. And that really was what they were pushing as an idea. Look, this is how closely we are tied.
Together, beneffecting from the upside.
Bloomberg's in King and Riley Griffin, thank you so much for joining.
Us on that roundtable.
Meanwhile, coming up software earnings, they take center stage two.
What to expect as AI fears enter the picture. This is Bloomberg Tech.
Shares of IBM had the worst day in more than twenty five years Monday, after Mthropic announced that its clawed code could help to modernize the dated coding language Cobol, largely run on IBM computers. This is in Fropit continues to unveil new AI tools for its cowork agent software across human resources, investment, banking, and design. Bloomberg's Brady Foyd joins us with the latest Cobol was invented in nineteen
fifty nine. There are two hundred billion lines of OBO called the underpin financial systems, banking payments around the world. And basically anthropics said it would be really useful to use Claude to help maybe investigate looking to change that, and then one of the most story technology companies in history had its worst day in twenty five years. Take it from that.
You've got to feel powerful if you're anthropic right now, right if you just say the name of a product the company associated with that project tanks. I mean, IBM has been talking about using AI for cobal modernization and it seems successfully offering that tool for a couple of years now. And yeah, markets are really jumpy, right, I mean, the potential that Claude or another AI tool can disrupt the leaders in a given software category is really frightening
investors right now. And it's just it's an incredibly jumpy market.
Ibmci A VP Rob Thomas was pushing back against this writing in a blog, and you quote him in your story that the value of IBM mainframe delivers has nothing to do with cobol is trying to talk about the platform more broadly brody, but what you say is so brilliant. How powerful anthropic must feel. Also to the upside, because look, they do a partnership with Intuit today and that fuels into it shares a little bit on the upside, so it can be make or break in either direction.
It fuels into it a bit on the upside. But the company's still down what forty percent this year. I don't know if it's quite that dramatic, but it's a scary time to be an application software vendor, right. I mean the most extreme idea that companies are just going to vibe code their own solutions. I don't think anybody really believes that, But the idea that software vendors lose the kind of pricing leverage they've enjoyed for so long
because of that potential disruption. It's really something that markets are struggling to grapple with right now, and it's going to be a really interesting earning cycle.
It's basically pitching itself now as a platform. Right. So historically Claude is focused on code and making engineering more efficient, and like if you're an into it customer, you're aready used of into its platforms and data. You just have an agent inside of it. If you're a small business or or a consumer. It's the same with DocuSign. We're seeing all these names move. How is it making these companies better, Brodie, you cover this beat inside and out.
Well, it's forcing them to really try to drive adoption and use among customers. Like we all know that a lot of the AI height was a bit premature in terms of how mature these tools were actually and how ready they were to be used in the enterprise. And it's really holding toes to the fire that hey, it's time to make sure your customers are actually using your AI tools or else the market's going to sell your stock and act like you are going to go poof in a couple of years.
Plombo's Brodie Ford always a perfect way with words.
We thank you.
Let's get more on the wider AI impact on tech and software with Clearbridge Investments senior research analysts for Software for IT services. Anyway, Fresh Henry the perfect person. I go back to in many ways what Brady was just saying. We're questioning just how good the tools are. How have we seen a sudden shift in how brilliant these AI agents are.
In the last few months we have.
We're seeing this sellof and software in particularly SaaS because so much is changing so rapidly, and these tools are evolving and tremendously. There's probably a pretty big disparity between some of the capabilities of the tools and what's happening and likely to happen, probably even intermediate term in the marketplace, particularly the enterprise marketplace. But investors shoot first and ask
questions at some very distant dates. So we're seeing a pretty dramatic compression in terminal multiples of many of the names.
At what point are they going to become discerning? Because at the moment, it's very hard for a CEO, for Aarvin krishnover IBM, for whether it's service now, Bill McDermott, to disprove that they're not going to be impacted, that's right. So how do they prove a negative? How do you start to see people willing to catch falling knife and be like, oh, these are platform companies are not going to be disrupted.
Or maybe they will be sure sure, It's so hard to tell the near term reaction to any kind of event it's been asymmetrically skewed to the downside. I think what the vendors really have to do is show that AI is additive to the business, to beat and raise and show that AI isn't just off setting into clients on the core business, but it's actually adding to the business, and that could really change the narrative. We actually saw that with the Snowflakes of the World two years ago,
with MAMO d B last year into this year. But that's typically what's required for investors to realize it's not just a zero sum game.
On the Bloomberg terminal, there are let's say a dozen reports of single name stocks that are moving because of the association with Anthropic, who held an event this morning. You studied the software sector right very closely. Any evidence those names are moving because people genuinely believe there's a fundamental change to how they do business, or is it just simply name association At this point.
I think it has to do with the idea that if Anthropic sees the need to partner with these companies, and Thropics sees that they provide something that Anthropic doesn't, and they understand these are incumbent vendors with installed customer bases and loyalty and processes and things like that.
It doesn't mean.
There isn't future disintermediation disintermediation risk, we all know that, but I think it's a signal to investors at these companies are viable and important to the ecosystem, and investors have not been assuming that.
Over the prior few weeks.
Hiary, there was a research report, a bearish research report from a little known outfit called Catrini that basically outlined various risks to different sectors from AI. One of the co authors joined Bloomberg Television. I just want to play you some of that conversation.
I thought there was going to be a small reaction. It was definitely larger than we expected. But I think it's not that surprising when you take a step back and consider kind of where the markets are in the US. You know, the AI trade has been going on for
three and a half years. It's been more or less a straight line up, and essentially like everyone is max long today and so there really aren't many incremental buyers left, and so it on the one hand, you know, spookspeople when you do consider what is negative about this, But I think specifically the market right now is trying to digest this idea that AI has gotten a lot more powerful in the last six months.
You know, in that he's kind of saying, well, this is kind of the function of the market of the last three years, but still poses the central question, which is is AI rendering existing software obsolete or is it making it better? Just your reaction to that research report and the co author's thoughts there.
Sure, I think it's both.
I think AI represents represents a risk and it also represents an opportunity for many of the incombent vendors. We could take sas separately, but we've been talking internally about what I call the great catchdown to sas. Every other sector, every other part of software selling down to SAS on disintermediation risks. But there are reasons why a lot of subsectors should be relatively more defensible, and we could talk about those a.
Few like, Okay, let's talk about what's just spent defensible Because there's this drip feed of AI dooomerism, shall we say, in these various reports and blog posts coming from the CEO of Anthropic himself to over at Szuctrini. We've also had Harvard coming out with their piece that we're seeing it being incredibly effective when it comes to financial analysis and portfolio decision making. The Harvard led study is talking
about how trading mutual fund training decisions. Seventy one percent it predicted right the model was trained over a five year window. But they're also talking about maybe the outperformance, but is what it misses? So from your perspective, where's the outperformance going to come from? Some of the companies that can withstand this, that can show that they're really effective while using the AI tools.
Sure Sure, I think of security as an example of that. Anthropic introduced a vulnerability management tool, not even management, just a scanner for vulnerabilities and code that's produced on the platform, and that's a build time phenomenon. Most security vendors are run time, but they bring a lot to bear that we could talk about from a technological perspective that it's going to.
Be hard for an l ELAM or an agent to have.
They have the physical infrastructure, they cover all the enforcement points. It's partly a physical phenomenon, not just a digital phenomenon per se. But on top of that, when you think about it. I don't know how entities, and especially regulated entities can have the fox guarding the henhouse when it comes to security, and kind of the same existed in
cloud as well. Because AI creates so many vulnerabilities, so many new vulnerabilities that we've never had, that it's a major issue when you're looking to what's becoming the greatest attack surface to be the solution to the problem. Similarly, the data platform vendors are really the platforms for the
next generation of AI applications. We're seeing them build on top of the snowflakes and Mango dbs, even potentially the Oracles to some degree data dog that's slightly different, but what customers are looking for is higher levels of automation to make their lives easier when there's a tsunami.
Of data coming at them.
These infrastructures and oftentimes are quite different. The data models are different from llms, the infrastructure on which they run are far more efficient than what lmms do and seem to be able to do. And so I believe the greatest source of automation is going to come from these
incumbent vendors. I should have mentioned data bricks of course, in which Clearbridge has a private investment, and so solving and headaches for customers in the face of a ton of complexity is going to be a Paramount Imports.
Hillary Fresh clear Bridge Investments, thank you very much. Now coming up on the program, Paramount is said to have raised its offer for Warner Brothers More on the Drama to buy one of Hollywood's most story brands, the snackt This is Bloomberg Tech.
Time now for Talking Tech and First Up.
Athropic has accused Deep Seat, Minimax and Moonshot of distilling.
Its clawed models to bolster their.
Own AI capabilities, adding to concerns in the US about Chinese firms improperly gaining an edge.
Representatives for the three firms didn't respond to requests for comment.
Sticking with Anthropic look, sources say the company will let some current and former employees sell their shares at evaluation for about three hundred and fifty billion.
Dollars, the level reached during a recent fund raise.
Anthropic has lined up five to six billion dollars from outside investors, though the amount and details have not yet been finalized, and Canada Well It's summoned open AI executives after it was revealed the company debate but ultimately didn't refer a chat GBT user to the police. That person became the sole suspect in one of Canada's worst ever mass shootings. Opening I bound the use of months before the attack, but said it found no credible or imminent threat ed.
Okay, let's take a look at shares Warner Brothers, Discoveries, Paramount, and Netflix and get the latest.
Paramount has raised.
Its offer for Warner Brothers from its thirty dollars a share all cash proposal. That's according to Bloomberg Reporting. Now Warner Brothers Discovery and its board is reviewing that revised proposal. Bloomberg's managing editor for Media and Entertainment leading the screen Time team because sure, we actually don't have the specifics. I don't think yet of how improved this offer is. But it's on the desk of Warner Brothers Discovery, It's leaders and its board. Take it from that.
Yeah, I mean, you see both sides taking this more seriously and being quiet. I mean there's been a kind of a progression with these talks, where back when the first bidding war was happening, you had Warner Brothers reviewing offers from Comcast Netflix in Paramount, it was very quiet, right, people didn't say a lot because everyone thought they were in it. Then Warner Brothers picked Netflix, and I got really loud, Paramount accusing Warner brother of doing things that
weren't right. Netflix and Warner Brothers often firing back. And now it's gone quiet again, which to me is clearly a sign that the board is seriously reviewing the Paramount offer. If Paramount has increased it five a dollar to a share, which we assume they have, it seems hard to believe that they would outright reject it as they have the recent ones, which means that they would have to go to Netflix and sort of say, what do you got for me?
Well, exactly, we'll get a drip feed of what it is exactly that they've got. But if it is markedly better, do we expect Netflix to respond in kind?
You know, it's a tough question to answered. Netflix has communicated to some of its shareholders. I've heard from people at Netflix that they feel they have the balance sheet to go up.
Right.
We're talking about a growing business with a really significant market cap. Even if this stalk's taken a little bit of a beating during these negotiations. So relative to paramount and even relative to the Ellison family, Netflix should have the money to go there. It's a question of whether they want to. You know, their stock is down. I've lost track at this point. It's more than thirty percent since these negotiations started, and they may decide that there's
a number that is too high for them. We just don't know what that threshold is.
It continues to evolve and we know that you always have the story, but mostly Beka, sure, thank you so much. I'm coming up more on Meta and AMD's multi billion dollar deal as the race to power AI intensifies and AMD up coming off of those previous highs.
Yep, and then Meta flat as a pancake. Much more to come. It is halftime. This is Bloomberg Tech. Welcome back to Bloomberg Tech. We're continuing to track the impact of AI on legacy software. Essentially lots of names actually moving to the upside.
The story is.
An anthropic and basically taking its clawed AI agent and linking it in a platform for everything from HR investment banking documents, and that is pushing a lot of names higher. IBM is up four percent, but remember yesterday Monday it fell by the most since two thousand when it said that Claude could be used basically to improve archive change update cobol coding bases, a program that was invented in
nineteen fifty nine. Otherwise, our top story is a deal between AMD and Meta six gigawatts of capacity over the balance of the decade, largely focused in the first instance on AMD's latest accelerator and the server that is based on mi I four fifty, but this is basically sixty billion dollars of potential revenue for AMD over that time period.
The stock up more than seven percent. MEATA has traded flat, but I guess that it's hard to gauge if this makes any different Semara at all given its existing spending commitments.
Karrac Well, someone's been writing on just that.
We meg Intelligence senior analyst Manday saying says that Meta's six giga what IMD deal actually lowest chip spending is a share of capex and secure supply tailored to AI inference.
Also right, so the stock.
Link structure could give Meta added leverage against Nvidia, Mande saying joints us now, why does any leverage against Innvideo It struck a deal last week with them. Is this more like we've got other options?
Yeah?
And look, I mean the fact that they are planning to use Nvidia's CPUs, to me, that was a sign that, you know, if you had to get the GPU capacity from Nvidia, you pretty much had to use everything they had to offer. With this deal, I think AMD sort of comes across as the more desperate partner in terms of, you know, giving Meta their stock warrants and making sure that you know, they are at Meta, which is going
to spend one hundred and thirty five billion dollars. Out of that, half of that will be on chips approximately, So think of it this way. They currently are about a ten to twelve billion dollar rund rate. This could potentially double, you know, every year in terms of just by adding one customer at the scale. Right, you know they plan to buy the chips from AMD.
You know the way that that AMD puts say right, single double digit sorry, billions of dollars per gigawa, but the gigawatts need to get built, right, There are operational milestones. Every time I go and cee AMD or we speak to Lisasus she puts a lot of emphasis on inference, you know, m I for fifty in particular, and what those systems can do in the inference phase. Do you in this deal see a point of differentiation for AMD if they do go after the inference piece.
I do, and simply because all these companies are trying to emulate what Google has been doing with TPUs. And remember TPUs are used for both training and inferencing, not just for Gemini but also for Entropics. So the fact that you know they have signed this deal and are willing to, you know, go from being a merchant silicon provider to a custom silicon provider for Meta, and Meta's workloads are more skew towards you know video and you know the family of apps they have similar to Alphabet.
To me, this is a sign that's what Meta fields they can accomplish, you know with the AMD chips. Yes, they will continue to use in video for training, but for inferencing, Meta's workloads are different from you know, any other provider when it comes to the chat boards or anything else that they're doing.
You know, AMD was it pains to point out they'll kind of co engineer the roadmap to future server designs starting m I four fifteen and go beyond that man deep seeing a Boomberg intelligence with the React Thank you so much. Now coming up on the program, Rayna Pope from matt X joins us to discuss his chip startup's latest fundraise as it aims to take on in Nvidia that's coming up next.
This is Bloomberg Tech. AI startup Matt Eggs.
Has raised more than five hundred million dollars to build hardware to compete with Nvidia. The chip company was founded by two alumni of Google semiconductive business and its aim is to make a product specifically designed to run large language models. Co founder and CEO Ryana Pope joins us, Now, I think we should get into the specifics, but my goodness, five hundred million dollars is quite a large series B. What do you need that level of capital for? And what does it reflect?
Yeah, so, I mean I would say I'm very happy to be here and thank you one of the Really what it reflects is, on the one hand, very strong confidence from some of our lead investors in the product. This is from Jane Streets and situational awareness have have led our around very strong on Jane Street sides. They're absolute technical experts. They understand the kind of product we're doing. And then situational awareness. That's Leopold ashen Renner's fund. He
wrote the book on AGI brand. He really understands where this old space is going, what we like, what we are looking to do with with this product and with this money. Firstly, I would say the demand for l M computer is just insatiable. All of the frontier labs are looking at where this space is going, and they're all concerned I'm going to run out of silicon.
I won't be able to solve all the demand I've got.
So our goal overall has been to make the highest throughput per per square milorde of silicon that any product.
Right, this is about computational density, that's right. That's right. And so you guys are basically saying through in terms of flops per minimeter squared, like, this is something you can own. What was the breakthrough?
You know?
What is it that you're so good at to achieve this?
Yeah, So there's really a combination of two things. If you look at the products in the market. Previously, there's been the HBM based family, which is in video Google, Amazon, and then there's been the s RAM based family and you are we are both actually uniquely so it's kind of taking two good ideas and putting them together. It is possible to do both very high throughput as you get from HBM, but also very low latency as you get from ASTRAAM, and do that in the same product.
What it gives you is actually a product that is better than any other product in the market at through put. This is exactly the flop Spisco milimeter that you described, while also matching some of the best like the Cerebras and the groc at latency.
Let's talk about how quickly people can start deploying this miner, because what your.
Aim is to complete the final.
Design this year, you hope to start manufacturing shipping even in twenty twenty seven. Who do you need to partner with on that? How do you expect to be manufacturing him in the.
US or abroad?
Yeah?
So, I mean there's a few big parts of the supply chain, and this is common for US as well as many other semiconductor firms in this space. Really, you need logic wafers, memory wafers, which is HBM, and then you need rack buildouts, and so those are the big parts of.
Our supply chain.
TSMC is well recognized as a the best provider of logic wafers, and then the memory wafers. There's the big three which are Skhiinix, Samsung, and Micron, and then there's a whole range of providers across the rack and manufacturing side. One of the big things that if you want to manufacturing very large volumes we hear about these you know,
multi gig or what deals that are coming out. These require you know, billions of dollars of manufacturing and then actually setting up like hundreds of millions of dollars put into setting up supply chains in advance of delivering that. And so that's a big part of what we're excited.
To be able to do.
Now.
You left Google in twenty twenty two and the goal was creating a better chit from scratch, Ryanan, But have you been impressed by.
The leaps at TPU has taken. It seems to have impressed the market.
What is it that you felt wasn't at Google for you that you now can bild better?
Yeah, So I think what is really required is if you want to absolutely nail the LLM workload, you have to be willing to break compatibility with previous chips. And so one of the strong guarantees you see all of the existing players providing is you can take a program that was written on my previous generation ship or my generation of chips five years ago, and it will run
on my next generation chip. And so a lot of what that means is there are constraints on my chip has to support all of the previous number formats I supported. It has to support all of the different programming model the way I communicate between cores on the chip, but all of those have to be the same as each other.
We felt that it would be necessary, like if you really want to just absolutely nail this workload without regards for backwards compatibility or other workloads or anything like that, you need to something of a blank slate design is required. For us, this means very large matrices, very low precision support, and then in fact an ability to split your very large cystolic.
Ray into small pieces. You name checked Grok, I think with some admiration a Sesson Ago. I mean like when in video acquired Grock Jensen. Wong's view is that they were struggling to find their place in the world. In markere forswere Sarah Brasse you named it as well, filed confidentially for IPO yesterday. Why might you succeed where CROC had to go to Nvidia And I guess they're working on something you know, and the public markets you know are needed for capital going forward.
Yeah, so I would say this is like historically the market has been won by the HBM place based players, that's the Google Amazon in VideA, and not by Grock and SERIBRUSSTRAM only chips are very good for latency, but when you want to run very long context models you run out of memory capacity.
SRAM is too small.
It's fast, but too small. Really, the hybrid of doing weights INSTRAM so you get the low latency as well as having the HBM for for very long context support is we believe that's what enables the low latency without all of the compromises that you would get otherwise.
Minor Pope, thank you so much for joining us today, Matt x CEO, Look we gotta stit with funding. News Basis startup making AI tools for account unting, has raised one hundred million dollars and an over one million dollar valuation.
This, of course is investors we're putting.
Wary of disruption from agentic AI here with us is Matt Hart based as CEO. I go to the point, how is BASIS different from anthropics cowork plug in for an accounting business.
Yeah, well, first of all, great to be here, thanks for having me, Caroline. Look, we have an excellent relationship with Open Eye and with Anthropic, and I think Cowork cha attribut the whole range of tools that they have are absolutely fantastic. I think though you can only sort of serve so many masters, and when it comes to domain specific work specifically in our situation accounting, there's a need to build with that domain specificity in mind. I
think comes down to a couple of things. First of all, you want domain specific capabilities. So, for instance, we announced today the first example of a long running agent completing an entire business tax return workbook. That's something that you can't really do in any other AI tool that exists out there. You need domain specific accuracy. So we are able to guarantee to the firms that we work with that the I will meet the requisite level of accuracy for it to be used in a real manner and
we can sort of guarantee that performance. You need demain specific user experience so that it's fluid and it works well for people who are experts in their field as they go about doing their work. You can't just have a chat pot or you know, the variety of other sort of basic user experiences that you need need to build enterprise sort of grade features. You know, collaboration, audit trails, you know, audibility etc.
Things of this nature.
And finally you need demain specific deployment approaches. It's going to you know, the capacity of a lot of these AI tools to be extremely useful in workflows probably far exceeds the adoption today and there's going to be a great challenge over the next decade of figuring out how to get AI into all the places it needs to get into. And so we think for all those reasons, being demained specific is very important. I think that's pretty evident to this.
Stat And so is that one hundred million to ensure the adoption curve is where you need it to be. Where does that money get deployed first and.
Foremost, Yeah, it's a combination of things. I think for us first and foremers who are always focused on building the most capable and the most accurate AI for accounting, So we are growing our engineering and mL teams dramatically. We're a fully New York based company, and we essentially want to be the home for applied AI work in
New York City, so that's always number one priority. But the other aspect of it is we are now starting to serve accounting firms across all of their practices CAS and core accounting being the initial segment, but now also tax audit, broader advisory, and over time. Figuring out how we can get AI into all these various places is extremely important. Demand has exceeded our capacity to serve it, and so we felt it was important to bring additional capital to make that puzsible.
So now I'm going through my taxes right now, and through the poor too of my accountant, who I won't name.
They do a good job.
I upload all of the documents W two, ten ninety nine, ten ninety eight whatever, but there is an element of observability because I still have to go in and double check all of the automated part of it. I know that's not the same, but the case study you gave of a business tax filing start to finish. The observability piece must be critical if you're handling that for businesses of scale.
Yeah, absolutely, good question. You know, we think that the ability for an accountant to understand what's happening at a granular level as the AI is helping them get work done is extremely important. That goes back to a little bit about what Caroline was referring to in terms of the you know, what makes us different than generic tools. Having demain specific user experiences and demain specific auditibility functionality
is extremely important. And then we also think that accounting firms have a very important role to play at the end of the day. Where people like yourself and what businesses around America probably want is a human that they have a relationship with, that they trust to do this
in chreatly important work. And so having the AI collaborate with humans to make that possible in an even better fashion maybe will improve the experience that you have on your taxes in future years as firms figure out how to bring these two things together is extremely important.
So here in the central question is is it an aid to an existing job or does it displace an existing role in accounting going forward, we just have thirty seconds' not.
A thirty second question, but I'll give you my best answer on it. So I think, you know, if you look at what's happened with software engineering over the course of the last year, and you can look at this internal into basis, pretty much today, no engineer at the company should be writing any meaningful amount of code, and yet our engineering team, our machine learning team is as
busy as its ever been. We're hiring as aggressively as we possibly can be in those areas, and that's because there's so much additional engineering work that we want to be able to do. That when you free up time to focus on more things, you can do more ambitious things. You can build things you never thought you'd want, you never thought you'd be able to do. And the same thing is true for accounting. There's tons of accounting work in the world that doesn't get done today. The Pentagon
just failed its eighth consecutive audit. You know, companies are misstating financials because they don't have enough accounting resources. Beyond that, you know the accounting. You know, if you go to any hospital system in America, they can't tell you how much it costs them to provide a simple procedure like a knee surgery. These are all things that accounting can
make possible. It is sort of the fundamental way that we understand economic activity that goes on in and around our organizations, and so we think there's a huge opportunity to do more accounting work. Most accounting firms the opportunity as well, and need the capacity to be able to
do that. And so from our perspective, this is going to allow firms and the accountants at those firms to take on even more work and get things more done and get more things done in the same way that sort of software engineering has been able to take off over the course of the last twelve months.
Matthew Hart, CEO and co founder of Basis. That was a pretty good, pretty good summary. Actually a little more than a pay seconds, so we'll.
Give you it. Thank you very much.
Now, coming up, we've got to look forward to what to expect from President Trump's State of the Union address, which is later tonight.
We have the preview.
This is Bloomberg Tech. All eyes are on President Trump's State of the Union speech tonight. This only days after the Supreme Court decided to strike down his tariff policies. Bloomberg TV's Washington correspondent Tyler Kendall joins us, what do we need to expect?
Via Hey Edwar.
President Trump is expected to tout his economic policies that are already enacted, but also push ahead some policy proposals that he would like to see enacted related to affordability, as the White House really plays defense on the issue
according to recent polling. In fact, the Wall Street Journal is now reporting that President Trump is set to announce a negotiated commitment from big tech companies to pay more when it comes to electricity costs related to AI data centers, in a bid to remove some of that burden from US consumers.
I was here at the White.
House just last month speaking to the US Energy Secretary, Christopher Wright, as the administration urged the nation's largest power grid to hold an emergency power auction specifically for these big tech companies. So perhaps we'll get some firmer details on that tonight, because, as you well know, this has been a White House trying to thread the needle between bolstering artificial intelligence but also trying to quell some of
those concerns from Americans. When it comes to jobs, and of course the cost of living related to electricity, because that is going to be the number one issue as we head into the MA in terms of a new poll out this week from Ipsos finds that fifty seven percent of US adults disapprove of the President's handling of
the economy. Front and centered, as you mentioned, is going to be that tariff policy, and sitting there in front of the presidents tonight is going to be those Supreme Court justices that struck down those IPA teriffs.
Last week, Mex Tyler Kendall with a rundown, thank you very much.
And look, the state of the Union.
Is also the target for bets on prediction markets like Cawshi, like poly markets. Look a market that was once a fringe obsession of economists.
And election longks.
Now traders are wagering on just about everything. Critics call it unregulated gambling, and its today's big take deep dive and please Save Blumberg contributor to Chris Beams here to talk us through it.
It is a wonderful deep dive.
And just as we think about the state of the Union, how has the idea of regulating this changed in the years.
So currently the status quo is that prediction markets are regulated by the CFTC, and the reason is that the technically offer what are called event contracts, and event contracts are derivatives, and so the CFTC argues that falls under their purview. Now, there's a lot of people who disagree
with the CFTC state gambling regulators. A lot of casinos and other members of the gambling industry argue, yes, that prediction markets should actually count as gambling and therefore be regulated by the states.
How good are prediction markets at predicting the future.
So it depends what kind of market you're talking about. A lot of scholars have looked at political markets and prediction markets to predict elections and have overall found that prediction markets are more accurate than polling, and certainly more accurate than you or me or any individual trying to prognosticate about elections. What's interesting, though, is that those markets
become less accurate the smaller they are. So, you know, a presidential election market could be quite accurate, but then once you get down to more you know, state and local races, when there's less liquidity, that's going to be less accurate.
I mean some of the bets that we see your extraordinary.
I mean, I don't know how much liquidity there is on whether Jesus Christ is going to return in the next couple of years, but that's literally something you can go and place a wager or at least a prediction on using these markets.
How have the companies themselves, CALCI.
And Polymarket navigated what has been them thrust into really now success but also backlash at the same time and worries about insider trading in.
Thus, Yeah, the companies have taken different approaches. CALSHI has really tried to position itself as the adult in the room. They emphasize the internal rules that they have around insider trading, around market manipulation. They have whole teams and software algorithms designed to detect trading patterns that might set off red flags,
which they then report to regulators. Polymarket is in a slightly different position because so much of their trading happens overseas and is not regulated by the CFTC, and they also just haven't talked about it as much. So I'd say CALSHI, particularly because it was the earlier entrant in the US, has emphasized cooperating with regulators, trying to be above board about the markets that they provide.
It's the latest Bloomberg, a big take, and it's a must read. Bloomberg contributes its crispy thank you very much. That does it for this edition of Bloomberg Tech. It's not as if there's a shortage of things to come. This week State of the Union and then in video on Wednesday, Oh.
And earnings after the Bell today as well. Look, just don't forget to check out our podcast. You can find it on the terminal as well as online on Apple or Spotify.
On iHeart. The tech news keeps coming as as a disruption. This is Bloomberg Tech