The Software Pricing War and AI Power Bottlenecks

0:13Welcome everyone to the Informations TI TV. My name is Akash Pasricha. It is Tuesday, September 22nd. I am here in our San Francisco studio this morning. Today on the show, we're going to discuss how software companies are turning to heavy discounts just to stay competitive with OpenAI and Anthropic. We're then bringing on another agentic AI founder who is coming out of stealth with$52 million in new funding from Redpoint and Lightspeed. We're then going to shift gears to a major power and energy spin-out coming out of Flex, which is a$40 billion manufacturing giant. I'm going to bring on the CEO.

0:51And to close out the show, we've got a conversation with our partner at Glance to look at how they are building the future of AI-driven shopping. It's going to be a great show, so let's get right on into it. In a world where it is open AI and Anthropic against the rest, software companies have to try everything they can to stay competitive. My colleagues Laura Bratton, Aaron Holmes, and Catherine Perloff published a piece today about how discounting is the latest tactic that has become a necessity for many of these companies. I want to bring on Laura to share more about what they found. Laura, welcome back to the show.

1:25It's great to have you here. Good to be here. So you wrote about the discounting that some of these software companies are having to pursue to stay competitive. talk about the context here and then we'll talk about the the specific companies that you guys wrote about in the story of course so since anthropic came out with its blockbuster coding tool clawed code investors have been concerned that its customers are going to use coding tools to build their own software and then similarly that they're going to start using ai agents to access their software tools rather than humans and that could impact the amount of seats that they need and their seed-based subscription costs.

2:07So there's been this broad shift toward usage-based pricing and outcome-based pricing, which basically charges based on how much AI you use and how much of the software is actually working for you. And I mean, this usage-based pricing conundrum threat, I guess, if you will, I mean, this has been going on for a long time. So how long ago did you start to hear about these discounts coming into effect? Yeah, so I reported a story earlier this year about how the pressure software firms are feeling as their customers shift some of their software budgets to these AI tools. Those software customers are starting to request shorter contracts.

2:51And that's resulted in just, you know, potentially an impact in which the software firms have less predictability in terms of their revenue streams. You know, ultimately, it could be a good thing. But as I was having those conversations with CIOs earlier this year, they had also started to mention some were seeing discounts and it wasn't necessarily quite as widespread as what I'm hearing now. But that's when I started first kind of hearing that this was a shift. And now it's something that I think a lot of software companies have been bringing up in earnings calls. right so who exactly is pursuing these discounts then talk about a couple of the examples that you guys wrote about so it comes in many different forms some software companies are offering longer free trials some are offering more ai credits some are offering straight up discounts so i'll walk you through a couple examples hubspot started offering free trials for its ai agents earlier this year.

3:50It's a 30-day trial, which doesn't sound like a lot, but its CEO admitted in a recent earnings call that this had weighed on growth in the most recent quarter. Workday offers free access to its Sauna Enterprise AI platform to top customers for one year. And then another example, Figma is offering more credits to use its AI tools for the saving cost, essentially. And just to be clear, these are discounts for the AI-related features that they're hoping to compete with the AI labs with, right? Effectively, yes. Another thing I heard from customers was that there are being discounts being offered on the entire contract or more AI products and capabilities are being bundled into existing software contracts.

4:49So effectively, this works as a discount on the AI products these software companies are offering. And their hope is that their customers aren't going to just use Claude and, you know, ChatGPT to access their software accounts. They're going to use the tools natively inside of their software apps. So discounting is such an interesting thing because it kind of depends on what the base is that you're discounting from. And pricing for AI services and products is kind of interesting because you have extraordinary costs also to cover with pricing. And so when you say discounting, was it at all a case that the base was very expensive to begin with?

5:34I mean, how much of a discount is this for customers, really? Yeah, I think it really depends which software provider you're talking about. And you're right to point out that a lot of this is occurring, right, as these software companies are trying to make the shift to more expensive usage-based pricing. And the hope is that they'll get customers hooked on these products that will soon become more expensive. In some cases, they actually shifted to usage-based pricing and then tried to pull back prices a bit in order to make it a little bit more palatable for customers. For example, Figma, after it switched to usage-based pricing in March this summer, decided to offer customers more credits to use its AI products for the same cost, effectively, you know, cutting prices for its AI capabilities to customers by as much as half, is what the CEO said in this recent buyer side chat.

6:31And he was really explicit that, you know, customers are cost sensitive and we had to pull back prices in order to drive adoption. Is it working? Were they seeing the adoption? Are these companies actually getting more business because of it? I think it'll remain to be seen whether they can convert the early traction they're seeing into revenue and margin expansion. But for now, it is driving adoption of products. Adobe in particular said that its freemium strategy in which it lets customers test out basic features of its AI products for free before after a certain, you know, they hit a certain limit, then they have to start paying.

7:17They said that that helped drive adoption of its AI creative studio Firefly in the most recent quarter and drive up revenue for that. But that hasn't translated into overall revenue expansion. Right. And so let me ask you one last question then. Prices are coming down. Is it your understanding from the people you're talking to, is this really just a play to drive adoption in the short term? Can we expect prices to come back up, do you think, in the long run? Or how long do you think this price war could last here? I think it depends on whether we see open models gain adoption. One thing executives are mentioning is that the price of AI will eventually come down and LLMs will become commodities.

8:05We've seen an increasing reliance on open source models. And so it just depends on how much of a leadership position anthropic and open AI can maintain and how much pricing power they'll have long term. But I think if we're thinking in the next year, of course, these companies are going to have to start charging the actual cost to deliver these products in order to have any sort of margin expansion. And if they don't, they're going to be punished by investors for it. Right. Great. Well, Laura, I want to thank you for coming on. That is Laura Bratton, author of our Applied AI newsletter here at The Information.

8:41A new AI agents company coming out of stealth aims to address the headache of corporate events booking. Andy, a company founded by Lohit Sarma, raised more than$52 million in funding from Redpoint Ventures and Lightspeed Partners, among other investors. The information was first to report that funding round in our AI Agenda newsletter out today. I want to bring on Lohit for a conversation. Welcome to the show. It's great to have you here, Lohit. Thanks for having me, Akash. So tell me about Andy. What is this corporate events booking agents platform that you've decided to dedicate your life to?

9:17Well, let me restate it. It's the way we define Andy, it's a corporate entertainment network, network being the key operative word because we're building for both sides of the ecosystem. So enterprises experience it as an operating system for all forms of entertainment and on the supply side, essentially a direct corporate demand channel that they don't have access to. That's fundamentally the difference between what we're doing and what you would so call your traditional event management agents and stuff, right? So the goal is for enterprises to essentially have an operating system with a business context and multiplayer workflows because corporate entertainment is spread across organizations.

10:00It's not travelist. Corporate entertainment is like, I mean, are we talking about like corporate dinners? Are we talking about conferences, venues? Like what does that mean? Great question. It's essentially broken across five verticals. Think of corporate entertainment as anything from two-person dinners, large dining rooms, event spaces, sports, merchandising, catering. So if you start thinking about how work is done in organizations today, everybody uses different sets of tools. So the goal is essentially to bring that into one centralized operating system, right? And then have the multiplayer workflows built into it.

10:38So finance teams can essentially have a centralized ecosystem through which they actually understand who's spending what. It's spread across series of multiple platforms and they don't have real-time visibility. So they're chasing people for receipts and it takes time to reconcile the books. So it's like Airbnb. It's a marketplace where the venues and the vendors, I mean, they list all of their services and their inventory. and then you have the businesses on the other side and the events team that can meet them on like an Airbnb type of platform. Is that the idea? The easiest analogy is think about the concurs and the bonds of the world on the travel space, right?

11:20Because you could go to Google flights and book your flights, but when you're in an enterprise setting, you don't do that because there's enterprise policies and stuff. But think about corporate entertainment today. You need the enterprise workflows of policy, spend, budgets, reconciliations, legal workflows. So there's that on the enterprise side and on the supplier side, which is the vendor side, it's that direct channel. So you can think of it as like a listing where they could list it, but then the workflows are different. The workflows require you to sign contracts, right? But you don't because there's legal ramifications, the check sizes are considerable, the payment terms are different.

11:56So our aim is to essentially build this network to solve for these enterprise workflows and they don't have a home today. It's done through a bunch of manual processes and service forms. So that's the same. And then you have an agent here basically do the booking, organizing, and admin stuff. Yeah. Today, a lot of the work happens through phone calls and emails, believe me. Yeah. Right. So the goal is we have agents on the supply side. We have agents on the demand side. Demand side has a context, understands the multiplayer workflow. Supply side, you have the agents that can respond real time.

12:28So today, when you're trying to book any of this, it's you're waiting for an email response and you're waiting for days, if not weeks, right? So now you instantaneously get a response. You get live listings. These are hidden naturally behind. It's not as simple as going on an OpenAI and cloud and saying like, send me the programs because those programs are usually hidden behind because those are very private for these groups. So the network here is interesting because building the network is what makes this. That's really the key success factor here. I mean, that strikes me as a process that is hard to apply AI to.

13:11I mean, I see the usefulness of having an agent do the bookings and stuff like that once you have inventory. but getting these two-sided, the two sides of the market built, I mean, that's like a pretty manual recruiting exercise, right? So that, I mean, like you're just going out and how are you convincing people to join this network? Well, we've been at it for over two and a half years and we just start coming out of stealth. And one of the reasons is building out that supply ecosystem. So we've got 1 ,600 hospitality groups, 93 ,000 venues that are some of the biggest hospitality groups from like Tau, JKS, the Michael Mina group, right?

13:50MML. So some of these groups that we've literally gone city by city. Right. And digitizing their content, bringing it up and building for these workflows. And then once you've got sufficient density, the marketplace starts kicking in and then the virtual concierge is being trained to essentially solve for any venue that doesn't exist. And there is a service layer underneath as well, right? Because just given the fragmentation of the marketplace, we essentially do have a service layer, which then it gets automated right after once the service layer solves for those venues that are not on the marketplace.

14:24Right, right. How many customers do you have using this? We have over 60 large enterprises currently using us. And these are like big tech companies? Yeah, so we've got companies like Cloudflare, Navon, Netsco, Salesforce, Monday.com, McRawhill. The goal overarching, if you start thinking about just entertainment as a category, it's not managed like travel. That's basically the problem we're trying to solve is how do you start moving this into a managed category much like travel? And it's a subscription model? Yeah. The way we set, we have, well, it's not, we have a platform fee for larger enterprises and then it's basically usage-based pricing, right?

15:04Based on a success fee. So if we're successful in completing the booking, we essentially charge them a success fee on that. Right. And how do you then deal with like token, you know, like the, you know, all of these long running tasks in the background that end up eating up all these tokens? I mean, how does this, how does this subscription then balance that? So that's basically baked into our platform costs and essentially the number of users that are using it. And we actually have a rough sense of like what the number of usage is. And we also have an intelligent routing layer where we understand what, from a browsing standpoint and discovery standpoint, which is majority of the activity early days, like what the usage is and what is the right model to route to underneath it.

15:46So that's something that we're keeping an eye on. And then in terms of completion of tasks, we didn't hand it off to the venue side. and the menu site, a lot of the, in terms of just the token usage is not that high, just given that we've already built the programs into the agents. So that's essentially how we're looking at token costs. It's something we'll keep an eye on. It's not perfect right now, but it's something that we'll continuously monitor. But the way we're doing it off the gate is essentially a platform fee plus a percentage of a success fee once the booking is completed. Got it. Okay, so that success fee is then how you can tailor the pricing to however involved the agent is.

16:28The agent was. That's right. Let me ask you one question before you go. I mean, Lohit, were you in corporate entertainment events before this? This is a pretty niche issue to dedicate your life to solving. Why is it this problem above any other problem you've decided to really commit to? Well, my previous, my background is in enterprise software. And it was prior to this, I was running all payments and money movement for ADP. And before that, I built an HR tech company, which was a special venture on ADP, became its upmarket solution. And I saw the same problem occur in both iterations of my previous life, where I was approaching it from the finance and procurement lens, where we would always run the issue of pulling back on corporate expenses, and especially on the entertainment side of the house.

17:17and I understood why. And when you looked into it, we realized it wasn't travel, right? So T and E. So we are the and E of T and E. That's where and E comes in because it was the E that was always the problem and not the T. So that's essentially, and then if you start looking at the market spends, the$325 billion market spend, right globally, in terms of how much entertainment is, entertainment just occurs. So it's a sizable market. It's a huge problem that every company, every CFO you speak to will tell you is something that hasn't been solved and it's essentially what we're trying to do here.

17:50Great. Well, Lohit, I want to thank you for coming on. That is Lohit Sarma, founder and CEO of Andy here on TI TV. Flex is a manufacturing company whose operations span all over the world. The company has a market cap of more than$40 billion. The company announced it is spinning off a new power and infrastructure company called Axiom recently, which will trade on the NASDAQ. I want to bring on Revati Advaiti, CEO of Flex and the expected CEO for the new company Axiom for a conversation. Revati, welcome to the show. It's great to have you here. Thanks for having me, Akash. So I want to get to Axiom in a second here, but Flex is not a company that I think our audience will know very well.

18:32It's a big company. Can you just remind us of the scale and the extent of that company's operations first? Yeah, Akash. So Flex is one of the world's largest contract manufacturing companies. which means that we're the name behind the brands. We make things for OEMs and customers from every sector, like automotive, industrial, some consumer products. We're big in healthcare. So you think of something, we probably make it. And we're in the top three manufacturers in the world. I think the thing I'll point out is, you know, big employee base, around 150 ,000 employees, around 80 countries. And so very, very global also.

19:18And so where then does the overlap come with power and infrastructure? How did you ultimately get to the decision to spin off a power and infrastructure company that is very much at the center of the AI conversation right now? Yeah. So Akash, my background before I came to Flex a little over seven years ago is in the energy space. I led the electrical business for a company called Eaton. And so most of my career has been in the energy space. And from that, I came to run a contract manufacturing company, didn't know anything about this space, but I love running businesses and I love doing transformations.

19:58So I jumped in and really focused on the portfolio. I'm a big fan of when you're doing any transformation, should get the portfolio right. So we did a lot of things. We exited some companies, some businesses. We spun off another company called Next Tracker. Now it's called Next Power, which is around a$13 billion market cap company today, and really focused on what parts of our portfolio can be built out in a way that is longer term and more strategic. And the reason for that, Akash is the contract manufacturing world as we all knew it was changing pretty dramatically, right? 20 years ago, 30 years ago, it was all about globalization.

20:41Labor arbitrage was the thing, moving things around the world. And it was clear that the geopolitics was going to play a big role and we should be looking at a more regionalized presence, but also more importantly, to add more product and technology capability versus only the traditional contract manufacturing capability. It turned out that as I was digging through the portfolio, FlexHab did some amount of compute integration for data centers. And being a big contract manufacturer, we put together trays, PCR, racks, all that stuff for data centers. Those days, volume wasn't that big. And then we had a tiny little business that made power modules that powered the chip itself.

21:28So it was a throwback from a little business that Flex had bought a few years ago. We had the IP and we were a product making company in that. I was very intrigued coming from the electrical business. I thought, it's like the hidden gem. It's like, you're taking the GE playbook and it's like, why has no one been paying attention to this particular vertical? Absolutely. And I thought that traditional electrical guys really focused on everything that was outside of the rack. I distributed power all the way to utilities. And this was well before NVIDIA and the ChatGPT moment and all that. It was clear that compute was becoming more and more power dense.

22:12And these power modules was going to require to draw a lot of power moving forward. Right. And so as I started thinking about it, I thought, well, finally, the electrical infrastructure is going to have to change because you can no longer build out power from the rack all the way to the grid the way we were traditionally doing. So I started buying up a few power companies, bought up a little distributed power company, bought up another utility company, bought up a cooling company, and really built out this portfolio kind of end-to-end with the idea that compute cooling and power is going to have to go together, and people are going to have to design it in an integrated way in terms of the siloed way it was happening.

22:55And then lo and behold, the chat GPD moment happens. Right. And it makes all the sense of the world to then make it a standalone company with all the trends coming up. Growth went crazy. We're one of the only companies that does compute pooling and power. Right. And then our capital allocation strategy had to change because this is a products business, not a contract manufacturing business. You put all that together and energy infrastructure is going through the biggest change ever right now. So let me ask you a question about the energy. And I want to come back to the manufacturing stuff in a second, actually, because you have an on the ground view here.

23:34But on the power side of this, so you'll spin out this company. You know, we just finished earnings season and all the chip companies have put out big projections. I mean, NVIDIA is saying they're expecting at least 70 % growth. Broadcom expected to double. And I mean, power is the bottleneck here. And the thing that I've been trying to get a sense of. is to what extent you think these chip companies will be able to meet their projections given the power bottlenecks? Or do you think that there is a world where the power bottlenecks start to create a scenario where maybe they can't meet their projections?

24:14Yeah. Akash, so this is what I think is happening and is going to happen. So the problem with power is, I always say since the Westinghouse Edison days, the electrical industry had no reason to change. So innovation has been slow. It's been really hard to bring in new products and innovate. Now, lo and behold, the data center thing comes along and it's clear that you can't run the way you were running. So the biggest issue in the power infrastructure is the way the power comes in from the grid and by the time it gets to the chip, there's a lot of step up and step downs that happen, which creates tremendous inefficiencies.

24:56And it was designed that way. It hasn't changed in a long time. So if you look at the current projections of the hyperscalers and all the silicon guys, it is clear that even 10 years down the road in 2035, we're going to have around 20 gigawatts of gap to the capacity that they need. So even with the current run rate that everybody's projecting, we're going to have a pretty significant gap 10 years down the road. That's based on your modeling, on their modeling? This is a public modeling. I mean, everybody has numbers out there. Center analysis has some great analysis out there. So if you take the compute projections people have put in, if you looked at the infrastructure build out that people have announced, and you look at how much power is available and you can bring in, that's around a 20 gigawatt gap in and now.

25:48Right. So, I mean, chances are that will certainly constrain their capacity going forward. So then Axiom, where does that company then fit into this? And I mean, just talk about the roadmap in terms of where you're at, in terms of the facilities that you're building and what the ambition is here. Yeah. So let me start from the technology side of it is what Axiom is doing is we are working with the silicon providers in the early stage of them designing their next generation chip about what their power needs are. And if you look at the chips being designed today, they're all going to move to something called 800 volt, which is a more efficient way of distributing power.

26:39And that is a pretty big step. Like for the layperson, they'll think, oh, what's the big deal? This is power stuff. You don't mess with it. There's safety issues. There's regulatory issues. But as the chip can move to that 800 volt, which is pretty significant, it reduces the amount of copper. You can put more power through it. It drives tons of efficiencies. Then everything all the way to the grid has to change with this 800 volt infrastructure that's happening. So Axiom, because we're almost a new electrical company that's being stood up, so we don't have any baggage from the past and history.

27:18So we're going to be able to build this end-to-end electrical infrastructure with this 800-volt DC idea in the most efficient way, which can drive around kind of 3 % to 5 % efficiencies, which is pretty big if you think about for a 1 gigawatt data center. And how do you do that? By bringing in new products. The products that is coming in to power the chip will be focused on 800 World DC. You bring what's called sidecars to power on 800 World DC. You go all the way to the utility. We just announced a$4.5 billion acquisition of a company called EPC Power that provides grid connectivity that drives to this efficiency I was talking about.

28:03So all of these products put together will drive an end-to-end technology architecture that is very significant from an efficiency standpoint. And then you put that with compute and cooling. You know, the big thing nowadays everybody's talking about is I just want everything on a skid. I just want everything on a pot. Like just stick everything in a big building and send it to me. And we are able to do that. We provide big IT buildings, all with cooling integrated. We provide big power buildings. The future will be, Akash, a compute pod, a cooling pod, a power pod, everything built together, integrated, deployed, so we don't have all these labor issues that we're driving with.