AI trends for 2024 & predictions for 2025

Frequently Asked Questions

FAQ

• Where is the transcript for this webinar?

  Speaker 1:
  Welcome everyone to today’s webinar on AI Trends for 2024 and predictions for 2025. We have a great panel of well-accomplished VCs from Frontier, and I’m truly excited to learn from their insight in today’s panel discussion. We’re exploring what has happened in 2024 and making some predictions about AI for 2025.

  Before we get started, this presentation is for informational purposes only and is not offered to buy or sell securities, which are only made pursuant to the formal offering documents for the funds. Please review important disclosures in the material provided for the webinar, which you can access at www.avfund.com/disclosures.

  Please note you’ll be on mute the entire presentation, and this webinar is recorded and will be shared after the event. We encourage you to submit questions throughout the webinar. We’ll try to answer your questions during the Q&A session at the end.

  Next slide. For today’s agenda, we’ll start with a very quick overview about Alumni Ventures, followed by our panel discussion and audience Q&A. I’ll let each panelist give a self-introduction in a minute, but let me first introduce myself. My name’s Ray Wu. I’m a managing partner of Alumni Ventures. I’m based in Silicon Valley. I began my career as an operator before transitioning into corporate ventures for almost a decade, initially with Cisco leading software infrastructure investment, mergers and acquisitions. Then I ran HP Ventures for almost four years. I helped to build several cross-border funds between Asia and the US, focusing on AI, software-as-a-service, robotics, IoT, and infrastructure services. I have a joint EMBA degree from UC Berkeley and Columbia.

  And also let me share a little bit about Alumni Ventures. Alumni Ventures is one of the largest venture funds for individual investors in the US. We started 10 years ago with about $1.4 billion from more than 10,000 individual investors. Based on PitchBook data, Alumni Ventures is one of the largest VC investors in the last few years—one of the most active VC investors in the last few years—with 1,400 portfolio companies. We have over 120 people, five offices across the US covering the most active startup regions including Silicon Valley, New York, Boston, Chicago. CB Insights, which is the leading information provider on the VC industry, placed Alumni Ventures among the top 20 venture capital firms in the US.

  And we are the only firm on this list that invests primarily on behalf of individual accredited investors. So great to have the honor, and welcome everyone again to be part of our webinar. Next slide. So let’s do a round-robin introduction of our panelists. Maybe Spencer, you want to go first?

  Speaker 2:
  Yeah, sure. Thanks Ray. So I am a GP at Soso Ventures. We’re a $1 billion AUM fund that’s a little bit unique in that we’ve partnered deeply with about 40 or so large strategic companies from Japan. So oftentimes when we invest in US tech companies, we’re looking at ways that we can bring that technology international and help them accelerate success in Japan. We’ve invested anywhere from $500K up to $35 million in businesses across stages and have partnered with companies like Coinbase, Zoom, Palantir, Square, and more. So that’s a quick version. Looking forward to diving into AI with everyone.

  Speaker 1:
  Thanks Spencer. Denise, you want to go next?

  Speaker 3:
  Sure. Hi everyone. This is Denise from Gradient Ventures, and we are the early-stage AI fund here in San Francisco and we’re backed by Google. We focus on early stage—as early as pre-seed, and for the most part, seed—and write $1 million to $10 million check sizes.

  We’ve been around for roughly seven years and have roughly 200 companies in our portfolio. Some of them you might have heard about like Lambda, the GPU cloud provider; Writer.com, the AI compliance tool for writing; and Crea.ai, the AI tool for creators to generate images and videos.

  We usually prefer to lead the deal and try to be the supporter for early-stage founders by offering operational support like go-to-market, hiring, PR/marketing. And we’re close with GCP, apparently. So we always give our founders more credits and also help them to access the private API or have better API rate limits. So there’s a lot of interesting technical and operational support.

  I personally joined the firm roughly a year and a half ago. And before that, I was a product manager/operator in tech companies for seven years. I worked for companies like Facebook and Twitter on the B2B side of product building—for infra tooling or advertisers’ portals—and also spent some time with MuleSoft and JD.com before they got acquired. So I decided to be an investor because I love looking into software from different stacks.

  Speaker 1:
  Great, thanks Denise. Yeah, we co-led some deals, right? Alumni does. And as you mentioned—Rashmi, then you’re next.

  Speaker 4:
  Thanks everyone. Really excited to be here today. I’m Rashmi. I’m a co-founder and managing partner at BAM Corner Point. We’re an early-growth-stage-focused fund, mainly focused on Series B through D stages across enterprise tech.

  I have been investing for the past 14 years and operated for 10 years prior to that. I’ve invested in AI for over a decade now. Some of the investments that we’ve shared with Ray at Alumni and with Gradient are Lambda Labs in the GPU compute space, Poolside, which is in the GenAI code automation space. I’ve also invested in a number of AI companies across healthcare, sales and marketing, product automation, and HR tech. Great to be here.

  Speaker 1:
  Thanks Rashmi. Chris?

  Speaker 5:
  Hi folks, I’m Chris Yeh. And most people know me as the author of the book Blitzscaling with my friend Reid Hoffman. But today I’m here in my capacity as one of the founding partners in Blitzscaling Ventures, which is a venture capital firm that has a couple of funds. But the one that’s relevant today is our AI fund, which focuses on early-stage AI.

  I also do want to fully disclose that I am a shareholder and one of the very early investors in Alumni Ventures, and it’s always great to be able to collaborate with Alumni Ventures on one of these webinars.

  Speaker 1:
  Thanks Chris, appreciate the support. So thanks everyone. So first, since we’re all investors, maybe we can start with the current AI trends. What do you identify as the most significant AI trends for 2024? And what kind of deals have you done this year that drive your investment themes? Who would like to start first? Chris, you want to start first?

  Speaker 5:
  I’m happy to go. So one of the areas that’s been getting a tremendous amount of hype recently—and we’ve contributed to that, so we’re not completely innocent of this—is the field of AI agents.

  And as usual, this industry has chosen a terrible term that nobody can understand. But the principle behind AI agents is: what if instead of just answering your questions or providing you with images or stuff, AI could actually do stuff for you?

  And so AI agents—or agentic AI, which is an even worse term—has been something people have been talking about a lot. But we think it really makes sense because if we think about how the world works right now, we have the internet, which is frankly a bad experience. We go searching for things, we’re hit with endless advertisements, we get followed around by retargeting ads—all these things make for a terrible experience.

  Whereas if you had a trusted agent that could actually do stuff for you, you would be able to take care of all that stuff on your own without having to go through it. The billionaires of the world have always had personal assistants, executive assistants who could do this. Large corporations sometimes hire entire staff to do this. But really, AI will be able to do this for all of us.

  So that’s an area that we think is really exciting because we think that AI that does stuff for you is vastly more valuable and impactful than AI that just answers questions. And we’ve been investing in some of the companies that provide infrastructure in this area.

  So we do both the consumer and the B2B side of things. On the consumer side, we have an investment in a company called Malon. That’s a Series A round that was led by General Catalyst. On the enterprise side, we have an investment in CrewAI, which is one of the leading open-source platforms for orchestrating multiple agents for the enterprise. And we recently made an investment—I believe we just wired some additional money today—in Skyfire, which is payments for AI agents. So we are trying to blanket the zone on that particular space. Hopefully we’ll be seeing more about that space in 2025.

  Speaker 1:
  Great. Great. Excellent starting point. Rashmi, anything you want to add?

  Speaker 4:
  Yeah, I would say that in terms of our AI bets, we’ve taken a fairly deep thematic view across the entire stack. And so as we look across the different layers of the stack—on the infrastructure space, on the middleware layer, and the application layer—that’s how we’ve thought about making bets across each of these layers of the stack.

  We’re in the midst of a massive platform transformational shift in technology, similar to what we have seen in the cloud or the SaaS era, but one that is growing a lot faster and changing a lot faster too.

  And so on the infrastructure layer, one of the key areas that we had identified was the lack of pressure on compute infrastructure and the need for specialized GPU compute infrastructure that will be needed to run all of the application demands that we’re seeing in the AI stack.

  That’s what led us to an investment in Lambda Labs, which is building a very differentiated, software-first approach into building a specialized and an off-the-shelf GPU purpose-built cloud for a number of enterprise accounts. The company has grown tremendously and we’re super excited about the potential there.

  Stepping into the middleware layer, one of the key angles that we think the developer stack or the MLOps stack needs to evolve is in having access to the best training data infrastructure for running these AI applications. That is what led us to an investment in Legobox that is building, again, the toolkit for a lot of applications and enterprises that need access to the best labeled data—or the training data—to be able to run the AI applications.

  And into the application layer—one of the key aspects that we believe is kind of a low-hanging fruit—is around automation for developer productivity.

  Code automation seems to be an area where we’re seeing a lot of investment as well as budget allocation from enterprises. The rapid rise of GitHub Copilot has been a great indicator of the need for automation in this category, and that’s what led us to an investment in Poolside, which is building an end-to-end, fully autonomous software engineer leveraging generative AI—so going beyond code completion into every aspect of the software development lifecycle going forward.

  Completely agree with Chris on the need for agentic workflows. And so going beyond a co-pilot approach into having a more complex stack where you’ve got multiple AI modules that are able to build off of each other and complete more complex tasks, especially in vertical industries like insurance, manufacturing, retail, and healthcare, are going to be incredibly important and driving significant productivity and efficiency gains across these sectors. So that is an area looking out into 2025 that we’re really excited about.

  Speaker 1:
  Great. These are all multiple layers of the areas we’re looking at as well. Denise, anything you want to add to this?

  Speaker 3:
  Yeah, just to add real quick—I definitely agree with Chris and Rashmi here, and I think AI agents specifically for code-related creation tasks are definitely very, very popular right now among customer adoption. And to go a little bit more granular in this category, there are different types of code creation processes. I think earlier we already saw many companies from Copilots all the way to Cursor in terms of fresh code generation.

  From our firm’s analysis, we believe more verticalized co-creation also can open up a really large market. For example, a lot of you must do this type of co-generation, but all the developers hate it. This type of task usually generates more traction in co-generation. Examples can be migration. Another example can be building some internal IT tool on top of a CRM or ERP system.

  And on top of that, solving all the security vulnerability issues that have been sitting in the backlog for many years—that’s another one. So these are the things you kind of have to do in order to fulfill some enterprise mandates or company mandates, but no developers want to spend time on these types of tasks. So we see these use cases as very promising for next year. We continue to monitor companies in code migration, or building IT tooling—internal tooling—and also focused on things like vulnerability issues.

  We do invest in a stealthy company currently doing AI agents for vulnerability issue co-generation. Essentially, you have all the tickets sitting in the backlog identified by the security scanners. They’re usually P1 or P2 in terms of priority, so no engineer is going to immediately address these issues, but they still present a huge vulnerability risk for the company. So we do want to have an AI agent behave like a Roomba—to continue to go through these tickets and solve them. We believe this is a very interesting use case to introduce AI agents in the security ecosystem.

  And I guess one last thing I want to mention is really about the native AI interface. So we talk about infra, we talk about middleware. I think in the application layer we all hear that, hey, AI can achieve a lot of use cases, but the true differentiation might not be in the model layer, might not be in the data layer—it might be in the experience layer.

  And the experience layer, a lot of times, is walking away from the traditional chat interface, which is usually very open-ended and unclear what exactly you can achieve. But nowadays, we see many, many companies using different interfaces—maybe the voice interface for inputs or outputs. An example can be Notably, using amazing podcast-like conversational AI to give you a download of the information that you’re trying to learn.

  And also, we see many companies trying to really create not just text inputs, but also using brush, color, and image for replacements of the text. Our portfolio company, Crea.ai, is really good at that. In a lot of visual generations, you cannot just easily communicate what you want in a video or in an image generation. You actually do want to use color, shape, brush, or all the visual components as inputs—as prompts—to replace the traditional text.

  We definitely see more and more of these types of user experience patterns appearing in the market. Many companies are generating a whole new webpage when you’re searching for a new keyword. A lot of companies are creating—another interesting company, not our portfolio—is called Perhaps. That company is creating another kind of overlay window on top of your browser as your summary. So there’s a lot of interesting interface innovation that we continue to monitor for the next year. So that’s the native AI experience that we’re looking for specifically.

  Speaker 1:
  That’s great. I do like the AI experience layers quite a bit. Just a quick question—do you ever see interrupt on the app side? Because fundamentally right now, when you look at the front-end or essential overlay by all the different apps, and then you see your interface simplify the whole thing—like you said, essentially as a new phone experience almost. Do you see that happening at all from a Google perspective?

  Speaker 3:
  From a phone, from a mobile phone experience?

  Speaker 1:
  Yeah.

  Speaker 3:
  We haven’t really seen a lot of companies who were able to nail that. I personally think it’s because small language models and on-device inference are still in progress—not necessarily fully reliable. But in the future, I think that’s the way to go, no matter whether leveraging the voice interface or leveraging cross-app—cross-app or in a way like OS-level AI agents—to understand what you’re doing in your Instagram app, understanding what you’re doing in your Coinbase app, and from there, understanding your behavior across multiple mobile apps to capture the intents, capture the user’s long-term history. We think that can be very interesting in the future as a consumer productivity tool.

  But yeah, so far, we haven’t really invested in any company in this sector just because I think the middleware layer is still in progress.

  Speaker 1:
  That makes a lot of sense. Spencer, anything from you?

  Speaker 2:
  Yeah, I guess I’ll just largely add on to what others are saying here. But I think what you’re hearing—if you’re in the audience right now—is that it’s one of the most exciting times to invest that we’ve seen in the last decade. And that’s because there just is such a plethora of opportunity.

  I’ve been looking into the health space and where AI can touch that, and we’ve seen large emerging companies everywhere from literally transcribing notes to an AI doctor where you can describe symptoms that you’re having but actually have a conversation—as opposed to going to Google right now and typing in “my chest hurts” and being told that you might be having a heart attack or 20 other cases.

  Now you can really hone in on something and try to figure out what’s going on here, what are questions I should be asking, what are things I should check—in a more, I’d say, consumer-friendly type of experience.

  I also talked today to a company that’s building an AI platform for addiction, for instance. What I thought was really cool about this is that there’s a stigma around addiction in a lot of places. And so in my mind, the barrier to join a group or go somewhere is somewhat high. But if I could have an experience that’s maybe a little bit easier to access—there is no stigma around it—and use that as a lower barrier to entry to figure out if I really should get help…

  This company has been taking off in a very unexpected way, but when I work backwards and think about the consumer experience—and again, this is just one example—I can see why.

  At Soso, we’ve been seeing this across many different industries, and I’ve started to implement some tools myself—like how do I respond to emails? There are some great tools out there, particularly around scheduling and follow-ups.

  If I can turn what takes me 10 minutes into a couple of seconds, that’s a huge productivity gain. And I’m seeing these just emerging experiences that make a lot of sense both on the consumer side and enterprise.

  And I’ll share just one more—again, more from the experience level: technicians working on maintenance of heavy machinery. For instance, here’s your manual—go figure out what’s going on with this machine—as opposed to an experience where I’m having a conversation with, call it an LLM, call it whatever you want, but I’m having a conversation describing what’s going on. It’s telling me, “Hey, have you done this test yet? Is this thing attached appropriately?” And it’s essentially a guide to solving the problem.

  So just what I’m seeing is entrepreneurs approaching these problems in a very different way, and I think the opportunity set is just massive. So it’s a very fun time to be investing.

  Speaker 1:
  Yeah, absolutely. Thanks a lot for sharing your perspective. I frequently hear questions about valuation as well—only because it’s very exciting, yet the public stock market keeps on hitting new highs, very much driven by AI stocks.

  Maybe talk a little bit in terms of what you feel about the valuations. Where is the current startup valuation versus the private versus public? And how do you think about, from an investor perspective, how they should think about investing into the private versus public at this moment? Spencer first?

  Speaker 2:
  Yeah, I can go first. I actually think there’s been a pretty strong rationalization in valuation that has persisted. What you hear in the news is the very few select cases of repeat founders that are raising massive Series rounds—and probably, I think we all have some of those in the portfolio and are very bullish on those moonshot-type projects.

  But at seed, I’m actually seeing, I think, great valuations given the traction that’s coming with breakout AI companies. So I would say the valuation might be high from a historical perspective, but if you take into account metrics that are better than they were a handful of years ago, I think it’s very rational at seed.

  And yeah, there’s a handful of categories that are really drawing massive funding rounds. Some of those, I think, do have very outlier 2021-type valuations, but they’re surrounded by exceptional teams with an exceptional and growing customer base from what I’ve seen. So there is a story there that they could grow into—but that’s a very select group of companies, in my opinion.

  Speaker 4:
  I would say on the growth side, it really is a tale of two cities. We see some companies that get a massive valuation uptick, and every investor wants to try and get into the round, as a result of which the round sizes up really fast.

  And then on the other side, you see companies with great fundamentals, strong operating capabilities, and yet they don’t quite get the attention of investors like they should. So I would say it’s a little bit of the 2021 valuations that you see on one side, and then on the other side, you see the rationalization of investments.

  It’s also interesting to note that the kind of investors in the companies that get hyped up are different than the ones that you would typically see. And in many cases, as you start parsing back the layers, there is participation from a significant number of corporate VCs that have metrics that go beyond just financial outcomes and valuations into the investments being made.

  It could be tied to a key strategic outcome that the corporate is looking to achieve—a go-to-market distribution agreement, or even an in-kind exchange for the amount of compute that these companies will be driving, in exchange for the valuation that is being provided.

  Speaker 4:
  So I think it takes a little bit of parsing—of understanding the reason behind why the valuation is high and what is driving it—and not getting too caught up in, “Well, if that’s the valuation, every company in this space should get marked the same way.”

  And then I would say on the public side, it’s definitely interesting to see—I mean, the stock market’s at an all-time high, or 30% up on the S&P 500 since the beginning of the year—when I think we entered the year with a very strong fear of getting into a recession. So it’s great to see that the markets are continuing to perform well, and with the rate cuts that have happened already and more on the horizon, it opens up the opportunity for additional IPOs and M&As to come down in 2025.

  In public markets, there definitely is the enthusiasm around AI-first stocks, whether you see that with companies like Nvidia or Palantir. But at the same time, there is also tempering—a balance of growth rates as well as profitability—where companies that probably grew at 60% but are still burning a ton of cash are not seeing their stock prices being valued the same as companies that have a good balance between growth rates and profitability.

  And that’s something that is reflected on the late growth-stage side, where we’re definitely seeing valuations being pared back pretty significantly from where they used to be in 2020 and 2021, which is why a lot of companies are now willing to consider raising a down round at multiples that are more in line with where public markets are now rating.

  Speaker 1:
  I wonder if that’s because of the AI impact on some of these companies as well. It’s always interesting to watch—you mentioned the tale of almost two cities, right? A lot of them are truly AI-native. I do think there’s a lot of jump. And then there are a lot of historical companies that have been there for a while, and then the AI impact on them is actually significant. Any other comments?

  Speaker 5:
  I want to take a step back and comment on something that I think is very common when people are looking at the public markets. As we know, the Magnificent Seven have accounted for a significant proportion of the gains, and this happens throughout the history of investing—which is that people look at these enormously powerful companies and think, “Clearly they’re going to dominate,” and we just extrapolate in a very linear way from where we are today out into the future.

  What history also shows is that’s incorrect. If we go back—I’ve lived through a number of different booms—if we go back to the dot-com era, we do not then see the dot-com era dominated by America Online and Time Warner. If we look back at the Web 2.0 era, the sort of social era, we didn’t see it dominated by Google or Microsoft creating social networks.

  If we look at the mobile side of the world—I mean, just look at the video game side of the world—we didn’t see the mobile video game market dominated by Electronic Arts and Activision, and the list goes on and on. So we may feel like, “Oh my gosh, AI is going to be dominated by the same big companies as before, and therefore I should just play this boom by investing in the public markets.”

  But the reality is there are going to be tremendous winners that come from the private markets, and in many cases they’re going to be doing things and having business models that do not have a direct correspondence with what came before.

  So if you want exposure to that element—which is often the biggest value creation element of the boom—I think it is a good idea to look at private market investing. Bearing in mind, of course, that this is very self-interested on my part—we’re all venture capitalists, we invest in private markets, so of course we think it’s a great idea.

  Speaker 1:
  Very good comments. Alright, why don’t we jump into another question in that case. Why don’t we start with Denise? With your background in product management and AI infrastructure, what are the critical components for building scalable AI products?

  Speaker 3:
  For sure. I think we already chatted about the front-end side of things—the interface stuff—so I’m not going to address that part that much.

  If we look into the backend side of things, right, no matter for AI applications or AI agents, I think when we invest in a company, we really care about a few things.

  First thing is really the cost and performance balance. So we’re looking at whether the company is able to make their product really scalable by not just calling one single API endpoint and relying on one single closed-source provider. We actually look for companies who are able to do things like creating mini custom models or being able to identify what task should go to what type of model, depending on the trade-off between cost and also accuracy.

  We think this type of architecture will be able to generate a higher ROI in terms of the cost and accuracy metrics.

  The other part is really about scalability in personalization. So I guess personalization—a lot of times it’s not just like, “Hey, you have internet data from your model, you generate a lot of smart knowledge.” You also need integration with enterprise data so you’re able to get more local knowledge.

  We also look for scalability around hyper-personalization per user. So as a user—as a company employee or just a regular long-tail user in the productivity tool ecosystem—we wish the AI applications or AI agents know all of my history, maybe over the last year, in terms of how I, as a user, click into different workflows, click into different components on the UI, what websites I browsed before, what artifacts I generated before.

  So we’re always looking for very interesting architecture choices to pick the right long-term memory or the right way to store your memory so you can recall that really fast at the individual user level—to really deliver true personalization.

  And I guess the last part is really something Chris mentioned at the very beginning in terms of the AI taking actions for you. So we know naturally AI agents eventually are not just going to give you predictions, not just give you recommendations—they actually need to do the work for you.

  But we haven’t really seen a lot of AI agents that are able to do that in production in a scalable way, because all the AI-rights API and function calling is not able to do those actions in a very secure way.

  So there are a lot of problems around authentication. How do you know the AI agent can be properly authenticated and take the right actions on behalf of the right user? And a lot of times, the API actions are not even natively created by the tooling provider. Salesforce is not going to provide an easy API endpoint for all the AI agents to take actions.

  So in that case, how can the developers—the app builders—be able to take the actions on behalf of AI agents? I think a lot of companies, including MultiOn, I believe, are trying to address this problem.

  So that’s why we continue to feel—hey, if you really want to make sure AI is super scalable—finding the right way to create the right actions in a very secure way and also a very scalable way is the right approach technically.

  So these are a few things that we look into when we invest in AI companies and also just look into interesting companies as a potential user myself. Yeah, that’s a few components—from personalization, from taking actions, and also from making a balance between cost and performance.

  Speaker 1:
  Yeah, that’s a great point, because it sounds like we’re moving from demonstration to more of real usage, right? You mentioned security, personalization, ROI—that’s essentially about applying AI rather than just creating it. So that’s a great point. Thanks for bringing that up.

  Maybe Rashmi, since you are focusing on the enterprise side a lot more, how do you see AI kind of transforming enterprise solutions in the next year? Any insights on your investments also?

  Speaker 4:
  Yeah, the way we view it today is that every software company will become an AI company. AI is going to be a key ingredient as part of every software offering.

  It’s just like how we don’t really think much about being a “cloud software company”—it’s assumed that software will be built with that cloud-first approach. We’re going to see the same on the AI side.

  What’s interesting on the enterprise side—I know Denise talked about this experience layer—we do believe that a lot of where the new enterprise stack is going to be moving toward is going to take that user-experience-first approach, versus the typical approach of how software was built where you need to have a database and then you have a front end.

  It’s going to be driven more by: What is the end user experience going to look like? What user behavior is going to be the easiest, the most scalable? And then you build the rest of your application stack around that.

  Speaker 4:
  I’ll give an example—Salesforce. As you think about it, the user experience really hasn’t changed over the last 30 years. It still is a screen that sales reps have to go in at the end of the day and enter in their information, and you hope that the information is captured correctly so that all of the insights that get driven out of that are going to be accurate.

  But as you think about the core use case that it’s solving for, it really is that system of record. It’s not that system of experience that you want the next generation of software to look like, which is why we’re seeing a lot of companies in the revenue intelligence space, which we think is going to take an even massive leap forward.

  You have companies like Gong that come in and are listening in real time to how sales reps are interacting with the customers and are able to provide coaching capabilities. You have revenue forecasting tools like Clari and Outreach that are coming in and telling you what is the next best action that needs to be taken by the sales rep in order to meet their quotas or meet their plan for that year.

  And when you think about stitching together the next version of CRM that is going to be from a user-experience-first standpoint, the database almost becomes irrelevant. You can plug and play any database in the backend, but it really is: what is that user experience that is going to be multimodal, where it needs to bring in inputs from voice, from interactions, from mobile, from the web—all of these together—to almost be that sales agent that can come in and help the reps close better, close faster, and for the company to be able to hit their numbers in a more consistent way.

  And if you think about the technology stack today—whether that’s Salesforce and CRM, or that’s ServiceNow in the ITSM space—all of these massive tech incumbents are at risk of being completely disrupted by a very new interface that may come in the future. And that’s what we’re really excited about: seeing what is going to be that next phase of AI evolution that’s going to come and will create the next $100 billion—or even $1 trillion—company in public markets over the next decade.

  Speaker 1:
  Yeah, that’s really exciting because it almost feels like regardless of what tools or infrastructure are behind it, it’s more about the problem you’re solving and how you actually get cut down to that noise almost, right? So the simplification of that whole process—that’s exactly why I think this is sort of the interesting revolution we’re seeing, rather than evolution to a certain degree almost—because you can cut through a lot of these traditional systems and almost re-engineer it from the ground up with something AI-enabled.

  So that’s very, very interesting. Maybe we can look at the international market. Spencer, I know with Soso Ventures you focus on helping startups expand internationally. What type of unique challenges do AI startups face in global expansion? How do you evaluate readiness, and what kinds of things are you looking at from a startup perspective that can help it to become more successful?

  Speaker 2:
  Yeah, I think what can make companies more successful is really starting to think about international early. I think a lot of times we bump into entrepreneurs who say, “Hey, I’m going to be ready for Asia next year or Japan next year—I want to come talk to you.”

  Oftentimes we say, “Hey, if you’re thinking about next year, you might actually want to be laying some groundwork now,” because relationships are very highly valued in Asia. So getting some initial touchpoints and understanding earlier can be really helpful.

  It also can be helpful to start to understand what those needs may be in advance—whether it’s around data and how that’s protected, or what other kinds of localization may or may not be needed.

  On the consumer side, we often see things almost naturally start to gain some level of traction in foreign geos and sort of draw entrepreneurs there as their business gets better capitalized. But on the enterprise side, thinking about this a little bit in advance—maybe partnering or starting to work with some of their entities that they own here in the US—can be a powerful method.

  Most of the groups that we work with—and I’m sure others on the call—are very large, multi-hundred-year-old entities that have a presence globally. And so if you’re thinking about that strategically, you can do a POC or start a partnership with a company even here in the US with the nod or the agreement that as we scale up, there could be a lot more opportunity in different geos. So yeah, I’ll just throw that out there for now.

  Speaker 1:
  Thanks, appreciate that. Chris, I read your book twice—so let’s get into Blitzscaling. It’s one of these books that I think is really amazing. At that time, basically, it was almost a new way to think about company growth. How can AI startups or investors think about this kind of Blitzscaling model, and how do they apply it effectively within the AI framework?

  Speaker 5:
  Absolutely. Again, we wrote the book before this most recent AI boom—but of course, Reid’s been in the thick of things with OpenAI and Microsoft and a variety of other things. It’s been amazing to see how many of the same principles apply.

  I think that the first and foremost thing is: Blitzscaling is about how do you win a valuable, winner-take-most market? How do you build long-term competitive advantage?

  And one of the lessons of Blitzscaling is that a technology moat is usually not enough. And that’s something we’re seeing over and over again in the AI space. You can have great technology, but there are a lot of other smart people out there who are building technologies as well.

  In many cases, the foundational models are advancing. Again, to us, it seems like they’re advancing slowly—because we check in on a daily basis—but really, for anybody who’s not in the AI industry, who’s not looking at it all the time, they’re advancing incredibly fast.

  And it’s often the case that—wow, hold on—a year ago, we had to build a specific model or company just to do this, and now it just gets to everyone for free. So that’s why we think that from a Blitzscaling perspective, it’s still really critical to figure out:

  • Where does the winner-take-most market dynamic come from?

  • Can I build in network effects so that individual human consumers are making a choice and getting more value out of something—through a classic business model like a two-sided marketplace or a communications mechanism?

  • Am I also thinking about the distribution strategy?

  OpenAI’s initial success has caused a lot of people to think, “I don’t need to worry about distribution.” I’m like—that’s true when you’re first to develop a magical product that everyone talks about all the time. But everyone else really needs to think about distribution.

  How do you find a channel that takes you to market? How do you get people to try out your product and stay with your product over time?

  So my overall message is: the same principles of Blitzscaling still apply—and in fact, in some ways, they apply even more, because the technology changes are happening so fast. Technology moats are not enough.

  Speaker 1:
  A hundred percent. I think that’s coming back to Rashmi’s point in terms of solving the problem—and also Denise’s talk about interface. I think this is all coming down to this whole problem-solving concept: how do we actually find the solution, versus just essentially developing the technology. So great point.

  Before we jump into the Q&A session, maybe we can look forward in terms of 2025—maybe do a quick prediction. What advancements, because we already talked a lot about what’s interesting… Anything you want to add for the next year? Any advice for venture investors out there and startup founders who are actually interested in this market? Maybe we do a quick roundtable on that. Who would like to start first? Rashmi?

  Speaker 4:
  Happy to go. I would say in terms of advice for founders—I think there’s no better time to start a new company. It’s very exciting to be able to leverage all the technology that’s at your fingertips and build a company on a net-new stack, versus—I think—many companies that started six, seven years back, which are now trying to figure out how to retrofit their architecture and technology stack for the new innovation that’s coming.

  So I think it’s a great time to be building a company. It’s also a great time from an overall efficiency standpoint, given how much the cost of compute and cost of storage has come down.

  The one ask would be: again, identify what is the pain point that you’re trying to solve for. So instead of building the tech and hoping that people will come, maybe go at it from the other way and figure out what is the problem that is unsolved.

  There are a lot of challenges in a number of legacy industries that need to be solved with a net-new view on how AI can help build that new user experience. So there is no shortage of use cases to go after—start from that approach.

  And on the investment side—I think we’ve talked about this—I am particularly excited about the application of AI in many of these boring industries that haven’t really seen innovation come for a long time. I think as the first wave of that, we’ve seen a number of legal tech companies get a lot of valuation and excitement—again, because there is so much repetitive behavior that happens in these industries that are very human-driven.

  The new term that everybody talks about is “services-as-software.” I do think there’s a significant opportunity to drive innovation in many of these boring industries by applying AI and automation to that. So I’m really excited about finding those opportunities.

  Speaker 1:
  Great. That’s a great point. Talk about legal—Denise, you are now investing in a legal AI company, so of course we want to go next in terms of future prediction.

  Speaker 3:
  Yeah, for sure. Quickly mentioning the legal AI—our portfolio company Lytics is really focused on patent analysis, drafting, and infringement detection—those sorts of use cases. I think a lot of times again, accuracy and cost trade-off, which we just talked about, in legal AI is especially true because accuracy requirements are so high, and maybe the existing model is not good enough to give you high accuracy, like 99.9% for certain use cases.

  So it’s really important to have a little bit more custom models per task in these types of companies. So we continue to look into these types of orchestration or model mode in newer investments during our due diligence process.

  And I guess one more thing in terms of the company tech stack we usually look into and I wish to see more of next year is the company with a data moat. So the data they’re using—apparently it’s not just internet data, which everyone can access, and also the enterprise data, which enterprise will give to you if they are your customer—but I want to get a little bit more detail around how do they clean the enterprise data to make the enterprise data more high quality when they ingest those data into the AI system.

  I also care about companies who are able to potentially even have private or exclusive partnerships with data providers—private data providers—so that they can access really unique data sets. I also care about companies with maybe synthetic data—high-quality synthetic data generation—as their unique data moat.

  So I think looking into companies with a unique data angle will be another important thing we want to double down on next year.

  And I guess one more category to quickly mention is maybe the data center side of things. Of course, data center is the lowest in the stack in a way, but we do see more and more AI workloads, AI chips being created in the market. Even today, Google has this quantum high-performance chip announced.

  So I believe anything relevant to data center upgrades—no matter using AI for data center prediction on the workload, or ensuring energy is more efficient over there, and also ensuring you can reduce the energy or cooling system to generate more space in the data center so you can put more hardware in the data center to fulfill the demand from the AI workloads—I think all these optimizations in the DC are very promising as well.

  And also the demand is for real, right? Every single kind of model or AI workload is top of mind for most of the model providers and hyperscalers. So that’s another category we want to spend more time on.

  Speaker 1:
  Great, great. Good categories. Chris?

  Speaker 5:
  Sure. One of the things that I think is very true about AI is—it is evolving incredibly quickly. And so one of the things I want to leave the folks on the line with is the notion that you have to be willing to rapidly unlearn the lessons that have proven successful so far.

  So if you learned a lesson from 2023 on how to succeed in AI—which principally seemed to be, “put AI in your name somewhere”—it didn’t necessarily work in 2024. And what worked in 2024 won’t necessarily work in 2025. Things are changing incredibly rapidly.

  We call this being an infinite learner—being willing to unlearn the lessons of past success. You can find my TED Talk about becoming an infinite learner on YouTube. And hopefully Reid and I will be releasing our book on becoming an infinite learner in 2025 or 2026.

  Speaker 1:
  Definitely love to learn more. And Spencer?

  Speaker 2:
  Yeah, really just sort of repeating some of what I said in the opening—this is an incredibly exciting time to be investing. Almost every space that we’ve looked at, I see the potential for large opportunities.

  Everything from space to defense to the foundational layer—we’ve seen it applied in different ways. We invested in a company called IC, building autonomous naval boats, and some different space companies I’m talking to that are using AI in very interesting ways.

  If I had to make a prediction for 2025, it’s that we’re going to see a big unlocking of capital that’s been somewhat held within these, I don’t know, Gen One / Gen Two internet successes, as they either get acquired or find ways to go public.

  I also think we’re going to see some huge acquisitions of AI companies—based on them only raising a seed round—that find the right use case, the right niche, where they grow extremely quickly and some big players jump on that.

  So I think it’s going to be an exciting year. I think that’s going to continue into 2026, barring some macro event that really slows things down. But AI is just so applicable across industries and business models that everybody’s paying attention to it, and we’re seeing the dollars that were somewhat speculative in ’23 start to flow in ’24. And I think a lot of business transformations will really be made in ’25, which will aid these companies that are building into that space.

  Speaker 1:
  That’s great. We’re almost running out of time. This is actually a wonderful discussion. So let’s jump to two Q&A questions.

  The first question: “Can I invest in this fund using monies from an IRA without taking money as regular income first?”

  Yeah, you can invest into this fund from an IRA. We’ll send you more information.

  Next question from Derek: “Can you please discuss the increasing energy cost for AI compute?” That actually synchronizes with what Denise just mentioned in terms of the data center side. Do you want to address that? Essentially the question is more about how the increased energy demand and cost for AI compute—how should we think about energy-efficient infrastructure? Any thoughts around that? Anyone can jump in also.

  Speaker 3:
  I would just add a quick thought here. Of course, I’m not the expert to answer how we make energy more efficient in the data center.

  But in general, we do see the push from potentially governments—either the EU AI Act or any other AI regulation—potentially requiring the company or data center to report energy consumption. I think that can be one of the external factors to push innovation for these energy-efficiency optimizations. 

  I hope that’s something next year the regulations can define better to drive that use case in data centers. Curious what others are thinking about this space.

  Speaker 4:
  On the energy side, we’re also seeing a lot of alternative energy sources being applied across the hyperscalers—at least on the public announcement side. Microsoft had announced them setting up a data center next to a fairly large nuclear center.

  So we’re definitely seeing more and more alternative energy sources being leveraged—at least starting with the large hyperscalers—and hopefully that spreads across some of the other providers as well.

  Speaker 5:
  Right. And the last thing—I’ll just say one thing before I hop off—which is that ultimately, entrepreneurs, companies, everyone responds to market incentives.

  If it becomes constrained by energy, people will start to use more efficient models. That’s why people are already starting to look at smaller models and find more efficient ways to do this.

  So I’m not worried about energy constraints holding the boom back. I think there’s plenty of opportunity to go around. 

  Speaker 1:
  Yeah, I think energy is essentially a long-term driver one way or the other. Energy and information are sort of the two constants in human evolution.

  I actually wrote a blog around that. It’s 100% one of the areas we’re watching as well. I think we’re running out of time, so let me just close it down then.

  We can answer your questions later. So our AI Fund V is now open. If you’re interested, I encourage you to join early—we’re offering meaningful discounts if you join early on. The first close is end of January, and I see that our team has shared the link in the chat.

  So again, it’s great to have all of you here. Thanks to our panelists for really sharing their insights today. It’s been a wonderful discussion. I want to wish everyone a wonderful holiday season. Thanks for joining us.