Section Insights
Introduction to Anthropic IPO Day
What is the significance of Anthropic's IPO?
Anthropic's IPO is a major event in the tech industry, priced above expectations, and is expected to have significant impacts on the venture capital sector and the broader IPO market.
- Anthropic's IPO is priced at $83, above the expected range.
- The event is anticipated to influence the venture capital landscape.
- The day's agenda includes discussions on market reactions and Anthropic's business strategies.
Trends in Venture Capital Returns
How is the venture capital landscape changing?
The venture capital sector is seeing a concentration of returns among a few top companies, while the overall number of companies has surged, leading to a disparity in returns across the asset class.
- Venture capital investments have increased significantly, but returns have been uneven.
- A small number of companies are driving most of the returns, creating a bifurcation in the market.
- Investors need to be selective about which funds to support in the future.
Expectations for Anthropic's Growth
What are the growth expectations for Anthropic post-IPO?
Analysts expect Anthropic to face challenges in sustaining its rapid growth rate, but there is optimism about its potential for profitability and margin improvement in the coming years.
- Anthropic is projected to grow significantly but may not maintain its current growth rate.
- Analysts are optimistic about the company's future profitability and operational leverage.
- The upcoming earnings call will be crucial for addressing investor concerns about growth durability.
Anthropic's Product Strategy
What are Anthropic's key product offerings?
Anthropic is focusing on cloud managed agents as a significant product category, providing personalized AI solutions for various customer needs.
- Cloud managed agents are positioned as a major growth area for Anthropic.
- The company is incentivizing product adoption through discounts and usage credits.
- Personalized AI agents are expected to be a key differentiator in Anthropic's offerings.
Nvidia's Lobbying and Chip Sales
What is the current status of Nvidia's chip sales to China?
Nvidia's lobbying efforts have led to expedited approvals for some chip sales, but the company is now being more conservative with its sales to China following recent developments.
- Nvidia's lobbying has influenced the approval process for chip sales.
- The company has paused some sales to China, indicating a shift in strategy.
- Ongoing scrutiny and regulatory pressures are affecting the semiconductor market.
Transcript
0:22 Heat. Heat. Heat. Heat. >> >> Welcome everyone to the information ti.
1:16 My name is Akos Pascricha. It is Thursday, November 19th, and it is anthropic IPO day. We have got a lot of coverage for you all day long. A quick reminder, right after this stream, we have our big anthropic IPO live video summit starting at 11:30 a.m. Pacific, 2:30 p.m. Eastern. It will be available to subscribers of the information on our website. Anthropic priced its IPO at $83 a share above its marketed range of $70 to $78. Today at the summit, we're going to be talking through the market reaction in real time, how it will affect the IPO market at large, the company's cyber security playbook, and even its playbook in drug discovery. If you are a subscriber, you can register at the information.com/events.
2:11 If you're not yet a subscriber, check out our website for more details. On today's show, we are going to be talking about the impact that this IPO will have on the venture capital sector. We're going to once again revisit the fundamentals of the business as well. We also have exclusive reporting for you on how Anthropic is working to justify its massive valuation with the help of some discounts. And to close out the show, we have more exclusive reporting for you on the Commerce Department expediting approvals for certain chips from Nvidia that it has been selling to China. It's going to be a great show and a busy day.
2:49 So, let's get right on into it. Anthropic IPO is bound to have ripple effects throughout Silicon Valley and the venture capital sector. I want to bring on Darm Tacker, general partner at Battery Ventures for a conversation about that. Dish, welcome to the show. It's great to have you here. >> Thank you for having me, Akos. >> Okay, so I want to start with Anthropic's decision to go public in the first place. I mean, they decided that they needed to go this route to raise the capital that they need for their growth ambitions. And I mean, basically, this was an acknowledgement that they couldn't raise the money that they needed in the private markets, which has been so lucrative for latestage startups or latest stage tech companies altogether. So what does this mean here?
3:34 I mean private market funding has that entirely dried up now or what's the story there? >> I think there's a bifocation between the halves and have not Akash where first and foremost the cost of frontier intelligence is high. So this isn't about raising you know a few billion dollars from the private markets you know anthropics almost raising a hundred billion for that you do have to access capital markets public markets that is u secondly you know there there certainly is a market for that in public markets just look at the amount of money that is being invested in in Google and and meta and many other public companies that are out there raising money and third this really helps them beef up their balance sheet to lock up the long-term supply commitments they need from chip manufacturers, from Broadcom to give them chips as well as cloud operators to deliver on their historic revenue growth figures going forward. So, I think it's a it's a very good move by them to capture the markets while they can and distance themselves away from OpenAI and others in terms of locking up capacity and cloud capacity going forward.
4:39 >> Okay. And we'll come back to that cloud capacity, but sticking with the the the private market fundraising here. I mean, you've also got names like Stripe, Anderil, Datab Bricks. I know you're a board observer for Data Bricks as well. I mean, how much appetite is there in the private markets for those companies? And what does this mean for those companies staying private? >> Yeah, I think look, there's there's two separate things. There's a need for capital and there's a desire to stay private. Companies like OpenAI Enthropic that spend 2030 billion for a training run need real capital while companies like Stripe and data bricks that are generally you know following you know 60 70% gross margins like traditional software companies don't need that much capital and so the their desire to go public is dictated by the their need for capital to deliver their growth forecast going forward. and then separately the the drive to go public or stay private is driven by a couple of other factors for companies like data bricks and I can't comment much given my my involvement there but Andrew and Stripe I can only speculate their desire to stay private is they can focus on big initiatives without worrying about quarterly reporting and and the market volatility that comes with it. On the other hand, I think employees have a desire to see liquidity and so as CEOs are balancing out the employees desire for liquidity while they can also make the big strategic decisions, they balance out staying public versus staying private. But I am hopeful that with the anthropic IPO and the reception that will prompt a number of other companies that have high quality revenues and big markets to possibly go public, you know, first half of next year. So in other words, this trend of of tech companies staying private longer, that story is is unchanged based based on these mega cap IPOs.
6:34 >> I think you have to look at it case by case, you know. So if I'm OpenAI and I need to raise, you know, 50 to 100 billion to compete with Anthropic, I pretty much have to go public because that's where you're going to access that kind of capital. If I'm an Anderil or Stripe, I have to balance two things. I have to look at employee retention and their desire for liquidity. And I have to look at how many big strategic initiatives I have that might that might make the near-term financials a bit more volatile and balance it too and and decide when to go public.
7:05 >> But I think the market should open up. This is showing you that there's almost a 100 billion capacity for or appetite for the market to absorb a high quality name like anthropic and that should benefit other atscale companies that have billions in revenue. The market's wide open when they decide to go public is is driven by other factors as I said, >> right? Okay. And then so you know I am curious about some of these younger AI companies which have reached some staggering valuations already. You know, I'm looking about this class of startups, you know, the Sierras, the Clays, the Harveys of the world. I mean, h that would seem to me like the next not the next crop of IPOs, but but a class of IPOs that may happen down the line. So, how far away do you think we are from those types of companies going public?
7:55 >> Yeah, I think going public is not just a matter of, you know, your valuation or your revenues. It also depends upon your addressable market and your predictability of revenues, right? The last thing you want to do is go public and have a quarter or two that you miss cuz then you're in the penalty box. So first all the companies you just talked about Sierra Harvey we have a company called Gong Jump and many others all in the AI application era the market is exploding for them as they're going after human labor augmentation or sometimes replacing right there's a $40 trillion tamper human labor that is being augmented or in some cases replaced by the likes of CRI and Harvey and others that expands the market and so these companies can grow into 1020 billion in revenue revenue relatively quickly and the growth projections are quite quite good for these companies. On the other hand, these companies have had to pay the the frontier model tax. You know, if you're using anthropic or open AI behind the scenes, you have to pay a high tax and thatffect affects your gross margin.
8:56 So, even if you are at a billion dollars in revenues but growing 100%. But at you know 20 30% gross margins that's not a great story to take yourself public. As you transition to open weight models and improve your gross margins while your revenues are growing I do believe the market will open up for many of these application names perhaps in the next you know two three years and you should see a number of vertical application companies in AI go public as well.
9:22 >> Right. Right. I want to ask you about, you know, the the power law and venture capital being at work here in magnificent scale, I guess you could say. I mean, with SpaceX, Anthropic, OpenAI will come up. I mean, these are big winners for many of these funds, but at the same time, I mean, you know, when we think about DPI and about exits and stuff like that, I mean, it's really concentrated in a few of these big names. Yeah.
9:50 >> And so I wondered what impact you think that concentration will have on the structure of the venture capital sector going forward. >> True. you are right. The private capital markets have also followed the trajectory of the magnificent 7 in public markets. You look at the S&P 500 and the top seven companies have generated almost 80 90% of the returns. And in the private markets you have two phenomena going on. You have the number of companies that have exploded to 40 50,000 is up four 5x and the number of winners has concentrated in the top 2030 names that are now worth 100 billion plus in in valuation. So it has concentrated. The positive side of all this is you know our LPs are very excited about companies trading in the trillion dollar range or hundreds of billions because the absolute dollar returns they get as an asset class are staggering. The dollars that have been going in venture over the last few years has expanded from a 100red billion to almost 250 300 billion annually without really seeing any DPI or returns the last few years. And so SpaceX and Anthropic and hopefully OpenAI data bricks and others going out will return you know 3 to 5x of those dollars back to the LP community which I think is a big positive. You see a second wave of application companies going public in the next couple of years which we just touched upon that should generate more liquidity. The flip side is that the the liquidity is dictated by 20 to 30 companies and you know the general partners that are in those companies will do very well while the rest of the asset class will see very difficult to generate meaningful returns. So you will see a wide dispersion in the venture asset class as well where a few a few funds will generate 90 95% of the returns and a big chunk of the long tail will not see it and so the next generation of venture funds are going to have to our investors have to be very picky about who they want to be involved with and who they have to stay.
11:45 >> Let me ask you maybe a a philosophical question here. Is that good for venture capital that bifurcation is that good for well maybe not is that good for innovation is is really the question it's good at >> yeah there's three constituents here our LPs our investors it's really good for them >> for them to see you know 3x plus returns at a massive scale is very good because the DPI helps the the the ecosystem for venture managers like myself I think it's great in that it reduces the overall noise and competition the number of venture firms has exploded to over 5,000 in the last couple of years that creates more competition. It funds a lot of companies that don't deserve to be funded. And so I think cleaning up a little bit of the long tail would be good for for the highest quality managers that don't have to deal with 20 competitors for every good company they fund. and third for founders I think is going to be good in terms of you know the ones that have the right teams, the right ambition, the right product strategy, they will find the capital.
12:43 But for a lot of other wannabe founders that are just, you know, going along and riding the wave, it might get harder for them to to raise money. So I think it depends who you ask. But overall, just more liquidity is great and more quality is great. >> Okay. And last question for you, Dish. I know you you're not an investor in anthropic. I don't think you're an investor in OpenAI. Maybe maybe I'm mistaken about that. but you know talent retention after these companies go public. I mean there are employees that are going to make a massive amount of money here.
13:18 >> Life-changing amounts of money. The lockup periods for anthropic sort of follow the cadence of SpaceX. It's going to happen. Looks like they're going to expire a little bit quicker than than in classic IPOs. So how do you retain talent after an IPO like this? Like what's what's stopping a wave of exits? employees saying, "Look, I I wanted to start my own thing. I'm just going to go do that." >> Yeah, that would be a good thing for the ecosystem. You know, I started my >> But bad for anthropic. It was It's not good for Anthropic.
13:46 >> Totally fair. Totally fair. I mean, look, I think there's two things going on. I started my career back in the late '9s when we were at the cusp of the com era. And a lot of my friends who were at companies like Yahoo and Google back then were enormously rich, but it freed them up to go pursue their ambition and start companies. And that's how you saw a wave of cloud companies start a few years later. I think you might see the same thing here. You're going to have a few thousand employees that are going to have life-changing money and that will free them up to follow their ambition and start their companies or do more than philanthropy or do something more purposeful with their life, which would be great. It's not so great for anthropic and open AI and data bricks and others that see some of that talent shar. But on the other hand, they would rather have a happy employee that is contributing 200% than someone who's just counting their options being vested. So you know the employee has served their purpose and AI can augment what a lot of employees do. They can keep driving the innovation way while having employees that are that are motivated to stay there and the ones that want to pursue their ambition will do better for society and it's great for us overall.
14:49 >> Right. Well I guess I should I should know better than to ask a venture capitalist about people defecting from tech companies to start their own businesses. >> Well it does make our job harder because we now have to pick you know 20 companies from 20,000 of them. but it's great overall. It's good to have a bigger pie and a lot more innovation to to fund and and be a part of. >> Great. Okay. Well, D, I want to thank you for coming on. That is Daker, general partner at Battery Ventures here on TIV.
15:20 For more anthropic coverage, I want to bring on two more guests who have been following this story very closely. Arjun Bata is partner and co-head of tech equity research at William Blair and Paul Meeks is head of technology research at Freedom Capital Markets. Welcome to you both. It is great to have you here. Arjun, I want to start this discussion not so much on how the stock is going to trade, but more so on the fundamentals of the business, which you've written about at length. And I want to start with the multiple here.
15:50 So, you know, let's just take where it priced at $83 a share, 1.5 trillion. you know depending on the estimates next year's revenue looking I've seen 100 billion I've seen 130 billion run rates put out there by Goldman as well so when you look at the the multiple for anthropic and compare it to the other names in the tech sector Arjun how does it stack up and what do you think of it >> yeah I mean I think you know I I think this is quite an attractive multiple it's going to be on our estimates it's going to be about nine times forward revenue meaning 9 times 2027 revenue and it's going to be still a business that we're projecting is going to grow 150% next year right that's about in line with Microsoft which is certainly not growing 150% right so you're you're you're getting a pretty reasonable multiple for a very high growth rate you know in an industry that's that's just getting started right that's the other thing you know I think we're just scratching the surface on AI use cases and when I look at anthropic I think there are sort of years of durable growth ahead we're going to see this business continue to grow and we look out to 2028, it's trading at six times our 2028 revenue estimate. So that's what fast growth gets you, right? That multiple is going to compress as we as we look out further and further. Obviously the the range of estimates gets more uncertain as you look out into the future more just because this industry is still nent, it's still developing, right? competitive dynamics could certainly change. but we're pretty confident that you know Anthropic is going to stay at the frontier and to you know get a business that's growing 150% at nine times it's pretty attractive in my view.
17:34 >> Okay. So when does it start to get unattractive that nine times again depends on what revenue forecast you you you land with but let's say it goes to 15 16 20 like what what what's the number that you start to get skittish about? Yeah, it's it's really hard, you know, it's hard to say where where it could go. You know, I certainly think, you know, when you're looking at other fast growth companies, you look at security, right? Security is trading in the 30 times revenue range. it's a capital light model, right? This is slightly different. Snowflake is trading at 15 times. You know, I think if we're in the 15 to 20 range, you know, that's maybe when you start to think about, all right, you know, is there too much baked in too soon here? that's that's kind of the range that I'm I'm thinking at this point. I don't know that it's going to get to 30 times revenue like Crowd Strike and and and others. but that's sort of what we think about from a valuation perspective. That's more of a near-term game though, right? Like, you know, I do think this is a company that you probably want to own for the next decade because I think it's going to have a pretty profound impact on enterprise on society over over the next several years. So, it's going to keep compounding, but yeah, you you may want to there may be some near-term risk management that, you may want to do it 20 times.
18:50 >> Paul, what do you think? What's the number where you start to get skittish about it? >> Well, I take a look at a couple things. first of all, I don't disagree with 15 20 times. Maybe it's not a cell, but, you know, it's a hole no longer, you know, attractive for accumulation. But the other side that hasn't been mentioned yet that you have to consider in all this is the profitability. And the company believes and I have no reason to doubt with the trajectory that they're on now that they can get to an adjusted operating margin of 35%.
19:27 And that's probably when you get to beyond 2028 when I have about 190 billion to $200 billion in revenue. And one of the reasons I think it's legitimate is if you take a look at the even the cloud companies of yesterday year, you know, Microsoft, Azure, Amazon Web Services, Google Cloud, as they mature, they seem to have been driving their operating margins to about 35 to 40% handle. And so explosive revenue growth, right? It's going to be by my model up 4x between 26 and 28 for Anthropic. And we may not get to that 35% operating margin goal, but it looks to me in the cards long-term 85% gross margin, 35 40% operating margin.
20:20 And so then you have to take a look at the company differently because it's not what I call just a profitless prosperity company where it's grown like a a weed and there's no reash or even non-GAAP operating income. I think we'll look at it at that point and yeah I think the stock >> and and when Paul in your model when >> when do you have it getting to to those to those margin levels? I think has a chance for 2028 maybe 2029 >> even with the all the commit you know that the commitments and the vertical integration they're going for that that that's including all that >> yeah because I believe that the positive operating leverage has a lot of potential and even in the S1 when you think of where they've come from in their few years as a company at all even a private company it looks to me in the cards and we've seen that in the prior cloud era as I said of course these were CPU based not GPU based 35 40% margins I think it's doable so >> I'm I'm pretty bullish I'm much more bullish on Anthropic than I was on SpaceX which I you know wrote and appeared a lot on television batmounting that IPO I'm actually pretty optimistic about this one even though it looks like it's going to come out at 105 instead instead of the 83 price. and then and I'm a tech contrarian most of the time.
21:51 >> Okay. So Arjun, so look, Dario is going to he's going to have his first earnings call in a couple months and analysts like yourself are going to have the opportunity to ask questions. I mean, what do you think are the top questions he's going to have to answer on that call? Yeah, I think you know some of the biggest topics that we've been talking to investors about on the IPO is durability of growth, right? you know, they're they're growing 12x this year. Certainly that growth rate is not going to sustain, but when you look at the October AR disclosure, they are adding 120 billion in AR on a run rate basis on a annually, right? You want that dollar figure to at least sustain, right? I so I think that's one thing that investors will will pay a lot of attention to. capex as as you were just talking about right I think that's going to be another big topic of discussion because anthropic is now you know going down this route of building their own data centers right thus far they've pretty much outsourced that to AWS and Azure and the hyperscalers and now they're going to undertake this project to sort of be in control of their own destiny right in a sense of this is a computeheavy business and they want to vertically integrate but the question is going to How much are you going to spend?
23:13 and I think we don't have clarity on what what the capex figures are going to be. how long is that spending going to continue for? Do you need to raise more capital in the future? Right? I think that's going to be another another big question. And the last piece I think is going to be more from a competitive standpoint which is how does Enthropic ensure it stays at the frontier because I think you know if you listen to everything Dario says that is a critical dynamic in the business meaning they believe if they stay at the frontier they're going to acrue outsized economics in this market. which means you have to keep investing in R&D. I certainly agree with Paul that it is they are doing that very efficiently.
23:52 but I think questions about how you can continue to do that and stay at the frontier are going to be very important. >> Well, and Paul, what about the the switching costs here? Because I mean we we see this story of developers oscillating between co between claw between codeex. I mean they've got co-work out there I know but I mean there's dots there's Google just came out with their own agent. Wow. >> So I mean switching costs seem like very low here.
24:21 Yeah, actually I'll give you that one though. I'll tell you that the demand is so voracious that all of the you know large scale credible frontier models I think will do well. I imagine OpenAI has you know just a a bullish outlook when they come public but one of the things that Anthropic has done which I think is set themselves up nicely and so maybe I don't want to exaggerate that is a huge differentiation but I still think it's meaningful is their attack on the enterprise and I think that's important. Open AI started, you know, more consumer focus.
24:59 It bit them in the ass and of course, they're now hustling to get into the enterprise as well. But I think that's important. The other thing that you mentioned just now that I think is important is out of the 858 billion in promise commitments, 120 billion of that is to your buddy Elon Musk and SpaceX. And under those contract terms, they can dump that contract within 90 days. Either side can. And so I imagine that yes, as they continue to more aggressively develop their own data centers, that right now SpaceX is a bandaid just because capacity is so heavy. wouldn't necessarily be a blow to Anthropic, but it could really be a blow to SpaceX and the revenue estimate for that. And and and do you think that they dump that contract?
25:52 >> I think their plan is and that's why they sign at a pretty high price, you know, a a clause that they could get out in only 90 days because that's the plan. Now, one thing, >> refundable refundable business class ticket. >> Well, and then the other thing you do have to know is it may not be anytime soon, right? because you know particularly now with the legislative scrutiny and the scarcity of power and even labor to put up data centers you know a data center from announcement to full buildout is typically around 3 years you know if you're lucky and you use you know modular manufacturing techniques maybe it's shorter but that is tough >> three years three years to up scale the Well, and and I got to tell you that I mean that that that's why the I mean you suggested the that they get to that margin profile quicker than I would have anticipated. And so it was the timeline there that that sort of concerned me.
26:58 But I know that these are all questions that Daria will have to answer on his first call, which I know we're eagerly awaiting can't come soon enough in a couple months. In the meantime, I want to thank you both for joining us. That is Paul Meeks and Arjun Batya here on TIVv. The information published exclusive reporting on the discounts Anthropic is making to try to gain traction in many of its products outside of Claude Code. My colleagues Kevin McGloin and Laura Bratton wrote a story about that this week. I want to bring on Laura to share more about what they learned. Laura, welcome back to the show. It's great to have you here.
27:36 >> Good to be here. Okay. So, we know that Claude Code has been a huge success for Anthropic. What you and Kevin really got into in your story were the other products that it's going to have to make as stunning a success to really diversify its business a bit. And, you know, I I want to go through each of these bets are making. I mean, Co-work is, I think, maybe the other well-known product that people know very well. There's been a lot of competition since coowork first came out from other companies. So you've talked to customers. How do they find co-work compared to the other options?
28:16 >> Yeah. co-work appears to be really popular. Anthropic itself said in its S1 filing ahead of its IPO that you know co-work has seen a higher level of adoption in its first nine months since being released in research preview than Claude Code saw at a similar point following its debut. So I think that that's a clear sign of co-work success. And then you know among the consultants that I talked to they said that their customer spending was roughly split probably about 60% to engineering use cases you know cloud code and then 40% to co-work for knowledge workers to use cloud co-work essentially as a personal AI agent to help them build slide decks conduct research build out dashboards that can help them with things like accounting recruit new folks. So it seems to be catching on.
29:09 >> Okay. So cowork is catching on. cyber is another application they're pushing into. I mean this was an interesting point in the story. You said that customers are shifting up to 20% of their anthropic spending in some cases to cyber. That sounds pretty good. Are there caveats here? >> Yeah, definitely. I mean, the consultant I spoke to who told me that said that, you know, they're using the underlying models with their own harnesses for cyber use cases and that that's what he sees as the next biggest use case beyond coding. and I think especially what we saw with the capabilities of Mythos and Fable, that definitely seems like something I think could could be true.
29:52 one thing that I'll note that I found interesting is that customers are using their own software to use the models for cyber capabilities rather than using Anthropic's own tool called cloud security. and I think it just speaks to the tension in which customers are developing their own software that can call Anthropics models for different use cases rather than going to the applications that Anthropic is developing. And I think that whether Anthropic can get customers to use those applications directly as it has with co-work will be a key tell of whether it can succeed beyond coding use cases.
30:32 >> And the the last product category I want to ask you about are clawed managed agents. What are these managed agents? How big of a business is that for for Anthropic right now? >> Totally. So my colleague Kevin got a lot of the reporting on claw managed agents. he really led the charge there. but what I can say is that cloud managed agents is essentially a service for building and running AI agents. and that seems to be another massive use case for customers that are trying to expand beyond cloud code. So really, you know, I think that you can think of cloud managed agents as more personalized than cloud co-work, which is more of a a generalized agent.
31:13 and you know, Anthropic's biggest use case is going to be agents and and providing agents, personalized agents. whether that's to knowledge workers through cloud co-work or specific customized agents through cloud manage agents to its customers. And that's really what we see as its next biggest use case beyond coding. >> Okay. So, these these are these are three categories of products that that you know they're they're growing quickly. They're a big focus for the company. I I want to talk about now how Anthropic is incentivizing people to use all of these products because >> you outlined a pretty ambitious strategy from from the company.
31:54 >> How does it plan to meet it? What is it giving customers here? Yeah. So, we've seen Anthropic offering higher discounts than previously. and in some cases offering usage credits to try new products, usage credits when customers shift from individual or team accounts to enterprise plans with annual spend commitments. and the discounts can range up to 30% or higher. in some cases, we've heard of a Fortune50 customer that was offered a much higher discount than 30%. and to get the 30% or higher discounts. We've heard of some customers sort of being told that they can only get that discount if they promise that a certain percentage of their AI budget will go toward Anthropics products rather than rivals.
32:43 which you know is called a share of wallet provision and is pretty interesting. It's something we've also heard of OpenAI doing. So sorry that the companies do they have to literally open their books and and show their their over this their bills to at this seems kind of crazy. Yeah, it it is you know in in the opinion of many customers I talked to something crazy and that they hadn't seen but you know what Kevin found in and his you know research into this was that Anthropic in some cases has asked customers to show invoices or documents to prove that they're living up to the promise of you know diverting a certain percentage of their AI budget to Anthropics products. so yeah, >> is this is and is this is this is this at all standard in in software or is this pretty novel here?
33:42 >> you know, I hear mixed opinions on that. Some people have said they've heard of hyperscalers doing this before. Others have said there's no way that companies and regulated industries are going to open up their books and let an AI provider essentially audit them to ensure that they're living up to their share bullet provisions because those are essentially promises. >> right. And then they and then maybe they'd say, "Hey, just upload into Claude. And by the way, the data, you know, zero data retention anyway, so don't worry about it." And then they say, "Well, we're also concerned about that right now." I mean, this is this is this is like a an uncomfortable level of transparency. I don't know. I mean, but maybe it works.
34:20 >> It seems like it's it's unlikely this will become a a widespread practice because I don't think customers would put up with that. and I, you know, I've heard of many cases in which Anthropic will offer the higher discounts 30%. when customers up their spending levels and they don't necessarily have to guarantee these promises. So, I think that especially as customers begin to use their own coding harnesses or essentially software that lets them use different models like coding agents. Anthropic is going to face more pressure and you know not necessarily be able to get away with behavior like this.
34:59 >> Right. Okay. So then last question for you Laura with these discounts I mean how long do you think these discounts prevail for? I mean are people concerned that this is all going to really have a reckoning in a couple years and the company's going to raise prices back up? In my opinion, the discounts are mainly to shepherd customers into usage based enterprise contracts where, you know, they don't have usage included in a $200 or whatever per month account and and we've seen a widespread shift to customers in the last six months. Like yes, of course, Anthropic made the shift towards usage based pricing way earlier this year, but a lot of companies are just now shifting to those enterprise plans. And you know, it seems Anthropic is from my reporting mostly providing these discounts to try and help with that transition and to get customers to sign up for for that annual commit. And I think as customers begin to be able to predict how much they're going to spend on AI annually, sure, maybe they'll be able to get a discount. And I think we should know that hyperscalers have long offered discounts for enterprise contracts, which is why the so cost of software is so variable for different companies.
36:18 and you know, taken together, actually anthropic and openi's discounting efforts are pretty modest compared to that of the hyperscalers. but I think that what what I'm trying to get out here is that, you know, this is a transitional period and eventually we're going to get to a point where the cost of tokens comes down a lot. Anthropic has already said that, you know, Opus 55, its latest model is so much more costefficient than a previous version of Opus and I think that's going to continue to happen. which is why the frontier model providers need to shift into the application layer so that they can, you know, grow margins and sustain revenue growth.
36:55 >> Right. Okay. Well, Laura, I want to thank you for coming on. That is Laura Bratton, author of our applied AI newsletter here at the information. More details have emerged about Nvidia's interactions with the US government as it relates to chip China chip sales. My colleagues Chanderlu and Leo Schwarz have exclusive reporting that the Commerce Department fasttracked approval of Nvidia's H200 chip exports to China. I want to bring on Leo for a conversation about that. Leo, welcome back to the show. It's great to have you here.
37:27 >> Thanks for having me. >> So, what happened inside the Commerce Department in getting these chip exports to China approved? >> Yeah, there's obviously been a lot of attention on the relationship between the Commerce Department and Nvidia and how it was handling different issues related to chip smuggling in recent weeks after the fallout from the Nvidia scandal. This new story that we have is actually the opposite. A lot of these stories have been about how commerce held up different rules or actions around chip smuggling. In this case, we found that after President Trump approved these advanced H200 chips to be exported to China, commerce actually expedited the process and in some cases staff at the Bureau of Industry and Security, which is the small office within commerce tasked with approving these felt that they were rushed to the extent of not being able to do ample due diligence on the applications.
38:22 And what were the repercussions of this? I mean, if if if approvals are happening faster than people would like, presumably this might mean that companies are getting a hold of these chips that commerce wouldn't have otherwise wanted to get a hold. I I wouldn't go that far, but I I would say that there are at least two examples that we document in the story of end customers of these chips who either have ties to the Chinese military or state-owned enterprises or have ties to chip smuggling. So it does seem like in some cases at least if there had been more time for due diligence or maybe if different processes had been followed those companies might not have been approved.
39:01 >> Okay. So let's talk about why this happened. What did you find in your reporting about the the the backstory here that led to the this expedited approvals? who were the characters involved here? Well, I I would just say in my conversations the past couple of weeks, I think a lot of this does stem down from the president himself. there's been a lot of policy within this Trump administration of not wanting to have blowups with China amid ongoing trade agreements, trade wars. obviously Trump is on a China trip right now.
39:33 what what we do know is that late last year he had approved the sale of the H200 chips, which was a big pivot in policy for the administration. The government famously was going to take a 25% cut of the sales which Trump was very happy about. And what our reporting shows is after that happened, they made this a priority to basically expedite the approval process. You know, we were told by someone familiar with the process that they did follow the sufficient due diligence and the various license conditions they usually do.
40:05 but what does seem true is that they at least put this at the top of the queue of what priorities should be within the Bureau of Industry and Security. And I think now just given all of the reporting we're seeing come out around dysfunction within commerce and BIS and how that relates to Nvidia and chip smuggling and broader export control policy. Obviously, this is just another example of where some priorities might have led to issues within the Commerce Department and Trump's China chip strategy more broadly.
40:35 is do you have any sense for did Nvidia pressure commerce at all to expedite approvals? >> I mean I think you can say that any company reasonably would pressure commerce like they're trying to get an approval. I don't know any company in the world who wouldn't say hey we want this to be done f we want this to be done more. I mean we've had obviously a lot of reporting in the past few weeks about how intense Nvidia's lobbying efforts across DC have been from the Hill to the White House to think tanks.
41:03 Certainly that lobbying happened within commerce. I I don't think we can say that, you know, Nvidia was the reason that they were expedited. we do have some reporting that it seems like Nvidia applications were expedited over some of their rivals like AMD. but at the same time, this is within normal behavior for for how a company's going to act. >> And I'm sure they they gave you a gave you a statement when when you asked them about this in video.
41:27 >> Yeah, I mean they said this went through the the normal process. you know we we we went through the the front door and things proceeded as as they were supposed to. >> Right. Okay. So look I mean we know this week collect crash said on the earnings call that they basically have stopped some of these chip sales to China altogether the ones that were approved which sort of suggested this entire episode has really struck a nerve with them and they're they're being a little more conservative with this. I wonder as you think about the the story moving forward, what are the questions that you're trying to get answers to?
42:03 >> Yeah, I think that's that's the the huge point here and you know, funnily enough, that that happened in the course of reporting the story where they had the earnings call and and she said that I mean it's almost it almost doesn't matter because China actually the the Chinese government has even though the US had approved them has been blocking most of these sales from happening. there has been a bottleneck. Although the information has reported that there have been some limited imports of these chips. but now that Nvidia seems to be putting a kibos on actually exporting them to China, it does raise the question of what Nvidia's China policy is going to be moving forward on these H200 chips which aren't even their most advanced ones. and I mean obviously right now the the company's in a bit of a holding pattern. So we don't really know what's going to happen in 6 months, let alone in a week. you know, who's going to be the CEO, what their policy is going to be, what types of investigations they're under by the government. So, you know, this is just the probably the the biggest story in AI and and tech and maybe business right now and and certainly export controls are at the center of that.
43:05 >> It is funny to me that we we are going to continue to hear we're going to continue to find reporting, I think, trickle out about these H200's chips and meanwhile, I mean, you know, as you just said, you know, they're they're not the most advanced. They're not the Blackwells and they're not the Reubins. I mean that, you know, Nvidia's product suite is going to continue to get more and more advanced, but H200's I think the, you know, this is going to be the chip class that people really focus in on. And I and I think, you know, at some point there it it will start to be less relevant because, you know, the chips will get so much more advanced.
43:40 But it was it was some great reporting from you and Channer. I want to ask you about a different story that that you and our team published this week. it was not about Nvidia. It was about Microsoft and you reported that Microsoft is closing in on a deal that could sell a pretty big amount of cloud services to Bite Dance. What are the details here? >> Yeah. Yeah. Well, when you talk about H200's and current export control policy, I actually think that this might be the more pressing one at least right now, which is this idea of of remote access or how Chinese companies are able to access the most advanced chips. This is one of the main ones. I mean, this is M Microsoft allows access to open AI through Azure through their cloud services. and the story that we report and that this remains a gap for many China hawks people following this policy because while the commerce department just closed the loophole that allows Chinese companies to access advanced chips in data centers in countries like Malaysia and and Singapore and Thailand, they can still do it through US companies and through the US. and that's what Bite Dance has been doing. Bloomberg had reported earlier this year that they were set to spend about a billion dollars this year on cloud services and AI through Microsoft. And what we reported is that they're actually nearing a contract for an additional $500 million in spend again mostly to access OpenAI through Azure on Microsoft. And this would be a way for Bite Dance, which is famously the the developer of Tik Tok, to be able to access advanced AI models.
45:20 Although I will make the caveat this, you know, what what Microsoft and by danceance would say is this isn't to train models. This is basically to be able to access, you know, enterprise open AI the way that a company in the US would. >> And and sorry, can you just make clear for us how does this relate to the to the remote access piece? I'm I'm missing that link here. So the remote access idea is that while Chinese companies wouldn't be able to import chips into the country, they can still access them abroad and data centers abroad. Of course, here this this is different. I mean this is bite dance is saying we just want access to open AI for our team. Although what we report in the story is that OpenAI has has raised concerns that these their models are actually being used for distillation or this process basically of using a better model to train another model. so this this all goes with the idea of Chinese companies getting inventive when certain aspects of the AI the US AI supply chain are cut off to them like chip exports they're still able to access either models or chips and compute through different means. in this case, while presumably they're not using it to, you know, access chips to to train their models, they're still being able to access OpenAI, even though OpenAI itself doesn't allow its models to be sold in China, >> but okay. And and again, another clarification question here. So, so is Microsoft are they selling I guess I'm just getting confused with the remote my understanding of the remote access was that I mean basically the companies in China are accessing compute outside of China and but here we have Microsoft selling to bite dance which seems to be in China right so where is the remote access part of this coming in >> well they're accessing open AI models and yeah like you said it's not accessing chips. But I think what a lot of critics of export of the new export control policy are saying is that well when it's actually US companies who are giving the AI services or in some cases data centers and chips then that isn't covered by this new rule which only covered those those seven countries mostly in in Southeast Asia. I I know that this is like the the level >> and maybe it's you know I guess remote access is kind of an interest because I guess remote access can mean the geographic remote access but then in this case it looks like we're talking about it basically through you know through an intermediary company in in some way axing the model accessing the model through a company that you are permitted to do some business with. So, I I I do want to ask you, Leo, I mean drawing the eye of open AI was was language that you used in the story and you you touched on it here. So, they're they're quite upset about this. Do do you think something will change or or help us sort of push this story forward a bit?
48:24 >> Yeah, so OpenAI has expressed concerns to Microsoft. Frontier Labs have briefed lawmakers on this this issue of distillation with Chinese companies. Some have pointed the finger at at bite dance specifically although bite dance denies the idea that they they use models for distillation. and we we also have a statement that Microsoft didn't find any evidence that there was that by dance was doing a specific type of distillation with the models. but what what could change? This is a a big question here. you know Microsoft and OpenAI famously have this fairly restrictive contract for OpenAI where Microsoft is allowed to sell its its models to Chinese customers even though OpenAI doesn't want that to happen.
49:11 that deal is set to continue through 2032. at the same time there has been a lot more pressure especially in the past few weeks to more closely examine what the relationship looks like between Chinese companies like by by danceance and US AI services. So after our story came out yesterday the China select committee put out a statement basically saying we want to have a briefing from for Microsoft to understand this new deal. So it could be possible that even if this contract between OpenAI and Microsoft doesn't get renegotiated around this point that there could be more scrutiny from lawmakers on it.
49:46 >> Great. Okay. Well, Leo, I want to thank you for coming on. That is Leo Schwarz, our tech and politics reporter here at the information. That does it for today's show. A reminder, we have our live video summit in about 45 minutes starting. If you loved these conversations, you are going to love our real time coverage of anthropic trading on the NASDAQ. Again, it starts at 11:30 a.m. Pacific, 2:30 p.m. Eastern. If you're a subscriber to the information, the registration is at the information.com/events.
50:21 And if you're not a subscriber yet, subscribe to the information. it is well worth it. A reminder, we are on this stream Monday through Friday at 10 a.m. Pacific, 1 p.m. Eastern. And if you can't make it, then episodes are available on the information.com, on our YouTube channel or wherever you get your podcast. Make sure to follow us on social media on X, on Instagram, on Tik Tok, and on LinkedIn. I am already excited for our next show tomorrow. Have a great rest of your Thursday. I'll see you on the live stream. Bye-bye for now.
Summary
- Anthropic's IPO is seen as a response to challenges in raising capital in private markets, highlighting a shift in funding strategies for tech companies.
- The venture capital landscape is bifurcating, with a concentration of returns among a few high-value companies, impacting future investment strategies.
- Companies like Anthropic and OpenAI require substantial capital for growth, while others like Stripe may prefer to remain private to avoid market volatility.
- Anthropic's valuation is under scrutiny, with analysts noting that its growth potential justifies its current market multiple.
- The company is offering significant discounts to attract customers to its products, raising questions about long-term pricing strategies.
- Nvidia's expedited chip export approvals to China have sparked concerns over regulatory scrutiny and the implications for U.S.-China tech relations.
- Microsoft is reportedly nearing a significant cloud services deal with ByteDance, highlighting ongoing complexities in U.S. tech exports to China.
- The competitive landscape in AI is evolving, with companies needing to innovate continuously to maintain market leadership and profitability.
Questions Answered
What is the significance of Anthropic's IPO?
Anthropic's IPO is a major event in the tech industry, priced above expectations, and is expected to have significant impacts on the venture capital sector and the broader IPO market.
How is the venture capital landscape changing?
The venture capital sector is seeing a concentration of returns among a few top companies, while the overall number of companies has surged, leading to a disparity in returns across the asset class.
What are the growth expectations for Anthropic post-IPO?
Analysts expect Anthropic to face challenges in sustaining its rapid growth rate, but there is optimism about its potential for profitability and margin improvement in the coming years.
What are Anthropic's key product offerings?
Anthropic is focusing on cloud managed agents as a significant product category, providing personalized AI solutions for various customer needs.
What is the current status of Nvidia's chip sales to China?
Nvidia's lobbying efforts have led to expedited approvals for some chip sales, but the company is now being more conservative with its sales to China following recent developments.