How Open Source AI Models Are Creating New Business Opportunities for Startups and Enterprises

How Open Source AI Models Are Creating New Business Opportunities for Startups and Enterprises

Open source AI models are gaining traction as companies, governments, and startups seek lower costs, greater control, and flexibility. Hugging Face explains how businesses can actually profit when powerful AI models become widely available, and whether American players can maintain their competitive edge in this evolving landscape.

Open source AI is booming. Hugging Face explains who profits. | Equity Podcast. | Transcript:

Hello, this is the Equity podcast again- TechCrunch's flagship project about the business of startups. I'm Rebecca Balon, and this is an episode where we invite industry experts to take an in-depth look at current trends in the world of technology. Open source models are becoming increasingly popular as companies, governments, and startups rely on the promise of lower costs, greater control, and flexibility. But as increasingly powerful models become widely available , how do companies actually benefit from this? And will American players be able to keep up as Chinese models gain momentum? With us today is Irene Solaiman, Director of Policy and Senior Vice President of Strategy at Hugging Face. Irene

has dedicated her career to researching how artificial intelligence models are produced, evaluated, and potentially misused. We'll talk about what startups should know before building products based on open models: from economics and global competition to cyberthreats that threaten openness itself Irene, welcome to our studio. Thank you, Rebecca. I am very happy to be here. Yes, I'm incredibly excited to chat with you again after our recent meeting at Disrupt a few weeks ago. You were on stage talking about model evaluation, and I am delighted with our conversation, because you are a masterful improviser, as we know from

our many conversations. You can talk about so many interesting things when it comes to open source. So, to give our listeners a little context, you joined the company-stop me if I'm wrong. You joined Hugging Face as a researcher in April 2022. You became Director of Policy, Head of Global Policy, Director of Policy, and Senior Vice President of Strategy in February 2025. Quite a long list of positions. So, in this role, you manage strategy, geopolitics, research, and policy. You founded Evals, an open source research community focused on AI evaluation. And before that, you worked at OpenAI on issues of bias, social influence, and public policy. So, stop me, did I mix anything up?

You know me better than I know myself, Rebecca. Oh my god, that's right. Good. So, what I wanted to start the conversation with is: when we talk about open models, we remember that in July we had your CEO, Clem DeLang. He talked about how companies are abandoning expensive AI services in favor of open models that they can run themselves to cut costs. We also see investors investing billions in training closed models. But if these models become a commodity and publicly available, then who makes the money? How do we make money here? Where are the threats and where are the opportunities?

Well, first about the changes: that conversation you had with Clem-it's funny how insightful you always are. I realized that this was before the incident we experienced in July. And while we've already seen a trend toward open models, it's only accelerated since then. So, I think today, European time, I don't know if you saw the news: Revolut-which is kind of a neobank for our American listeners, based in London, operating mostly in Europe- announced a move to an open-source model, with Claude and Codex only as exceptions. We've seen this general trend, especially since July, when both large and startup companies have announced their move to open scales with the clear goal of reducing costs.

Regarding your question about what this means for model developers: there are several ways to make money, and it's harder for developers if everything is hosted on their own servers, but for enterprises it's a better option, both for costs and for data privacy and localization, especially if you want it stored in a specific jurisdiction. But, specifically, as far as earnings are concerned, it will largely be the costs of inference. It could be a kind of "cloud dependency" if people use inference, for example, through Alibaba Cloud, which is the parent company of Qwen, Alibaba's flagship model, or it could be a revenue sharing system, which is a rather boring area of licenses that we can

delve into if our listeners don't fall asleep. Yes, okay. So, revenue, for example, for open model providers-is it revenue sharing built into the model license? I know this is what you were talking about, as if striving to lead this direction in Hugging Face. Yes. Licenses are usually written in legalese, so they are not often read carefully. Right. There are a few standard licenses most commonly found with open source: MIT and Apache 2.0. They are quite liberal. Unlike the latest iterations of flagship open-ended models, which use special conditions requiring a certain level of income sharing if you reach a certain financial threshold. So, we are talking about income distribution.

There is potential for cloud providers to get a piece of this pie. Perhaps companies that build applications based on these open-source models. Could it be that we live in a world where open models create enormous value for the AI industry, while the companies developing them struggle to get anything out of it? It's hard to say without looking at the exact financial statements, and not all companies are public. So, the next few months are going to be quite interesting. The level of value for both employers and developers, in my opinion, also depends on the structure of the business organization. So, as far as Alibaba Cloud is concerned, it's such a huge company. This can be considered a kind of feature to

attract people to Alibaba Cloud, and Qwen is used really all over the world. So, both their smaller 27B model, which is Apache 2.0 licensed and runs on a laptop, and their larger frontier-level model -the latest Qwen 3.8 Max-is an open-scale model that can be used locally, or more easily used for inference. So a lot of people in these scenarios are earning significant amounts, but we can't see those exact numbers. Rightly. So, okay. So some of the big companies, particularly we're talking about China, like Alibaba, offer open source models, but they also offer cloud and computing power, right? So they can offer it as a kind of package deal. Whereas if we look at the

American ecosystem, that's not necessarily the case, is it? Many open model providers are simply model providers. They don't have the computing power that they could, I mean, maybe they have deals with Neo4j that they're trying to sell a package deal with, but they're not the ones getting the computing revenue. So I'm honestly a little more worried about them in this case. This is a difficult decision if you are releasing an open model that does not quite reach or surpass frontier level. Many people like to use frontier. I mean, these are great products, they're simple. Then how to get such users interested in deployment? And that's where these individual licenses for American suppliers of open- scale models come in. Some of the most notable

players who have done a great job are, for example, Meta, who has historically been the openweight champion. And RC, Reflection, release models with liberal licenses, but they also enter into corporate contracts. So you'll see that with Reflection, they have a corporate agreement with South Korea and have a partnership to create a sovereign AI cloud. So, there are other ways to implement such individual solutions. Of course. That is, almost like a premium service. Well, like, you know, you want to build your own sovereign or enterprise solution and you want to use open source models. Class, we'll create it for you, somehow Yes, despite the huge interest in AI, there are very few people who really know how to adapt it.

Yes. So, we see this with the trend on FDE, it's becoming the most in-demand profession in AI: deployment engineer. Because there are very few people who know both business and AI tools to create something customized for corporations or governments. That's right. You know, many of our students are startup founders. So why should this concern a founder? What should they pay attention to before creating a product on an open model? Before starting development, it is worth asking: what are their priorities for a specific task? And this is where we move on to routing. We don't have great model routers to pick the right model for the job.

The ability to create a stack where you have a potentially cheaper but effective model for certain tasks. Let's say I've been thinking a lot about cybersecurity, such as monitoring to get insights for low- risk infrastructure. You probably don't need the most powerful frontier model, but you should understand: what is the model capable of for my task Where can I integrate it into my infrastructure? And do I have the resources to run the model, or do I need an output, whether it's an open-scale model that's cheaper than a frontier model, or a frontier model itself? Yeah, okay, that's interesting. I keep hearing that not every company needs the most powerful model for every task, right? So, what tasks are open models, perhaps even

small ones, getting better at? And the second question: in what ways do more advanced models still have a significant advantage? This word is so important. This phrase is " good enough." I guess it comes down to a trade-off on how much you're willing to spend on tokens. I get scared thinking about the value of tokens without open-source models, and this is not a criticism of the labs that create advanced models. These are simply incentives in the market. For what people want to use open models for and what they are capable of, they can perform many, if not the same, agent tasks. I've been relying almost

exclusively on GLM 5.3 lately. Sometimes K3, using the Hermes agent for almost everything. And I think that's a compromise, because additional motivation may be needed. It doesn't give me the bounces that a frontier model would, but I guess it really depends on the situation, although I would say I use it for everything. Yes, and you know, it's not just a matter of there being a "good enough" model, it's increasingly about the strapping, right? This is a structure, an infrastructure that is built around the model itself. So I remembered you telling me earlier that less powerful models with better strapping can actually outperform the Frontier model. So how do you think this affects overall productivity?

Well said. For these more specialized startups or tasks, we've discussed this before, they often have cybersecurity issues. There are bindings that can focus on specific cybersecurity tasks and use models. And it will still use a fairly powerful model that can, for example, detect vulnerabilities. Such straps, which are increasingly appearing, are made to order. This could be a separate source of income for people who are able to develop the harnesses our economy needs now.

Yes, and we are seeing this more and more often. And as a journalist, I find my inbox filled with startups trying to ride the latest wave. At first it was just AI app companies, then agent companies, they're still emerging, and now it's the binding companies. I guess the agent is like a tie-in, or, you know, they're very much related. You know, at Disrupt we talked a lot about model evaluation. A lot of work can be done to make sure that model evaluation is done correctly , but one thing that seems to be missing is the lack of good evaluation for bindings and tooling. Yes. What do you think about this?

Yes. I enjoy talking to you in general, but especially regarding evaluation. This is a niche area that has attracted attention because of its importance for deployment, but it is a separate branch of science. We simply have much more scientific data on natural language processing, the foundational field for large language models. We don't have much experience with the tools, as their current deployment is relatively new. We do not have clear sets of tools for evaluating tools that are standardized before their deployment. And I think the big barrier here is how damn expensive they are. Agent workloads, if anyone used them, are very expensive because of all these tools. So,

if you do an assessment, the amount could easily reach millions. Yes. So how can a startup handle this? How can they know if a cheaper model and accompanying tools are good enough or cheap enough for their product if there is no reliable way to measure costs , return on investment, and what you get? This will be a compromise between the short term and the long term. In the short term, it may be easier to just run the inference, turning to a cutting-edge lab if you have a high- intensity task, or if you don't have a lot of AI experts on hand. It's understandable to use whatever is easier. And often these will be closed models. If you want to scale, you know you'll be using it for the long term, which is why many people buy GPUs rather than rent them.

You can configure everything on your own. You can optimize the model over time without rent, and given the rising cost of computing, your long-term costs are likely to stabilize or decrease. This is desirable, depending on the business you run. And as for choices for intensive programming tasks, the latest open-scale models, like the Qwen 2.5 Coder 32B or Qwen 2.5 72B, are optimized for coding tasks. So, you can look at these types of benchmarks, the types of coding tasks that you're interested in, or, again, if you have these skills, you can further train this model. Hugging Face has many data sets for further training to optimize the model for specific tasks.

You mentioned it, I want to dive into it quickly. You mentioned that a lot of people are buying their own GPUs these days, right? Yes. I'm interested in learning more about this. Maybe you have some advice, because you know, we hear about GPU shortages all the time, and let's say I'm a startup founder, I need to experiment, try a lot of things. I use open models. I want to have my own, you know, GPU situation. Should I buy one, and how many GPUs do I need to buy? And is it okay to not get the newest, greatest Nvidia chip, because you probably won't be able to anyway.

I mean, yes, scarcity-it's not just, as you say, the presence, but the presence of what exactly? There is a huge demand for the latest Nvidia chips because many of the latest models run best on them. In terms of what you can get, there are many more libraries and resources, including on Hugging Face in general-that's the beauty of an open ecosystem-that will help you adapt the model to your chip. So, we see previous iterations of GPUs like Nvidia Blackwell, which many are using for frontier models, and here we'll delve a little deeper into the technical details. It's also a question of latency for what you're deploying, i.e. the level of response speed.

During our incident in the summer, we deployed Nvidia Quantize, or NV FP4, which is a four-bit quantized model. It won't have the power of full-speed inference, but it will get the job done and cost significantly less than we estimate the cost of an agent swarm. And for the listeners: if you don't know what incident Irene is referring to, it was, of course, when a swarm of OpenAI agents accidentally hacked Hugging Face, and Hugging Face was able to defend itself by using open models instead of closed ones, which I think set up some security barriers, which we'll talk about later, because I want to discuss the open source attack with you And one more note: Nvidia has acquired or is in the process of acquiring

Hugging Face, so just keep that in mind. One more thing about Nvidia, and I'll approach it from afar. I want to talk a little bit about China and the competition for dominance in the open model space. You know, the best open models-you listed a few of them. At this point, they are all Chinese. What happens if Chinese models become the standard for AI startups around the world? And actually not just for AI startups, but also for sovereign AI of governments; we see that the sovereign models of Rio de Janeiro and Japan are built on Qwen. So what, what's going on? Is this a problem? It's always a difficult balance when we talk about open models and geopolitics. Geopolitics has become much

more important even in the last few months. This is the question: a problem for whom? People love open models, no matter where they come from. I had to scale very quickly with our APAC division head. It's actually just me and Adina working with Open Frontier Labs, these Chinese labs that are so talented, that work so hard, and it's incredible to feel this ethos of believing in open scales. It would be even more exceptional to have a thriving global ecosystem and open leading models from the US. We work closely with the Muse team, with Reflection, with RC. It is really hard to train an open leading-edge model, but there is a reason why governments prefer models developed in the US, for example. There's a reason

why GPT-OSS still has a high number of downloads on Hugging Face. There is a demand for American models. How do we fuel this? I think this is a big, big question around narrative. There has been a demonization of open source, the claim that these models are more dangerous, although if you look at the explanations and warnings about cyberattacks in the last few months, the most famous cases have involved advanced models with closed scales. We're just looking for a balance to ensure that people have access and have the tools at their fingertips, and often that's what open scales are.

You said there is, you know, a demand for GPT-OSS. There is a demand for American open models. So many people, many companies and governments want to build on American models, not Chinese ones. What is the main reason? Is it just geopolitics? By the way, for context: GPT-OSS was released by OpenAI around August 2025. It's my fault for not explaining this earlier. This is because Sometimes I think: we're all geeks here, aren't we? We all know what we're talking about, right? But there were so many of them that I measure time by the latest generation of advanced models it was 2025. August 2025-this is GPT-OSS. When will the next open model be released? Well, because.

We here, you know, are also impatiently waiting. Yes. By AI standards, this is a very long time for a model. Since then, much better models with open scales have appeared. So the fact that people are still looking for this shows us that sometimes there is hesitation about deploying Chinese models because of geopolitics or some misunderstandings. I think part of this misunderstanding is the assumption that people are using APIs for Chinese models or models developed in China, and then their data will be sent to the mainland. We take great care to ensure that no data is sent to the mainland during our withdrawal. The partnerships you see between Chinese labs and AWS, for example, show that there are ways to deploy these models without their own GPUs,

but based in the US. Some of the reasons why people use them outside, using US chat rather than Chinese-developed models, could be due to this misperception, avoiding verification, but it could also just be a preference for certain values if you don't have the technical depth to be able to further train them. There are some differences, there are some opinions about what kind of rejections you get from models developed in China, for example Yeah, I mean, the main one I'm aware of, and I think it's probably more of a function of their API than the model itself, but I'm not sure. Well, you tell me. Well, you know, in the early stages of the whole LLM discussion, you know, everyone was testing, um, Deep Seek, or not even Deep Seek at that point, probably. It

was probably too early for Deep Seek, but they were testing some Chinese models, for example, asking about Tiananmen Square, and they didn't respond, right? So I'm not sure. Is this what you mean, or are there any other refusals? Yes, this is a classic example, and part of the reason you haven't seen in-depth analyses is that cultural assessments are very complex. We've done a few ourselves, and it takes PhD-level work to do it properly, but just with.This is where we went from "vibes" versus letters to a combination of both, it's, uh, there's, there's these sort of.I think it's like "Claude language."

There are different ways that different models talk. You can tweak, completely adapt, and make the types of cultures you want to emulate your own, but it's relatively resource-intensive, requiring GPUs and talent to make it work. Yes. Okay, so GPT-OSS, are there any other open models from the US that are currently being uploaded, used, or that are really popular on Hugging Face? We're seeing such huge demand not only for the new Petron, but also for the overall lineup coming out of Nvidia on the open-source side. So, the Nemotron family is a more advanced option in the field of language models.

What is usually overlooked in the open source community is biomodels, science, and robotics. You guys just released a bio model on Hugging Face. You created your own. This is really cool. Carbon X, and it's a fairly small model. You can run it on a single GPU. Here is the result. Yes, and if you run it, partly thanks to a single GPU, it is orders of magnitude cheaper than similar models, and this greatly helps biologists in their daily work. Yeah, I mean, you turned the conversation back to Nvidia for me. I promised I would get back to Nvidia, but you did it for me with Nemotron. So, I think, you know, one of the things we talked about earlier is who has the resources to support,

you know, if you're going to be at the forefront of open models, you need money. It's expensive to train such models, isn't it? And of course, there's distillation, which helps, although many people find the word almost a swear word, it's a very common practice in AI training. But does America need a giant like Nvidia to actually solve this and not only compete with China in the open source space, but also compete with OpenAI and Anthropic here in the States? This is a question of what the ecosystem as a whole needs, and we are currently going through the regulatory process of joining Nvidia, which I am very happy about, and since these are reflections that I have come to independently

, it is appropriate to make this caveat. I think we would thrive most in both open leading-edge and local models. So, right here on Hugging Face, you see the "golden mean " of what people are engaging with both through likes and downloads. We're looking at a lot of metrics in the 20 to 40 billion parameter range, and especially with quantized or much more powerful recent MacBooks, DGX, Spark, or AMD Halo, you can run them locally, or models with one to three trillion parameters, which we call "open frontier." For the ecosystem, we need smaller, open-scale models that are very flexible, but we probably need Nvidia-level resources to create a cutting-edge American model that people can explore, evaluate, deploy locally, and that would be a good

alternative to our closed frontier You know, Nvidia has recently become a big supporter of open models. I mean, my personal theory is that the more people use open models, the more GPUs are bought- either individually or people need to use the cloud, and this is happening at a time when a lot of labs, if they're not actively doing it, are threatening to create their own AI chips. So it's an interesting area, but certainly Jensen Huang, Nvidia CEO,'s commitment to open source has been supported by many other company executives.

We're getting to a point where using open models is no longer so stigmatized, right? Everyone seems to admit it. It's becoming more mainstream. But at the same time, it still faces threats, right? So you and I talked a little bit about the attack on OpenCode. I'm not sure, I hadn't actually heard about it at the time of our conversation, because our algorithms are just different. Yes. I'm sorry about that, because TechCrunch covered it, but you know, it was during Disrupt, so I was really busy. But in any case, in mid-October, OpenCode became the target of a hacker attack. So, OpenCode is an AI agent for

writing code that is part of the open model ecosystem. This was a roughly 48- hour agent attack targeting third-party user API keys. This compromised some accounts. Some financial damage was done. So, we see attacks targeting Hugging Face and OpenCode. Of course, they differed somewhat in motivation. One was motivated by a human, as in the case of OpenCode, as opposed to an attack by a swarm of agents. But still, these platforms are potentially becoming increasingly targeted. Are you concerned that security pressures could force more closure, closure. closure in the open model ecosystem?

This is why we need journalism, Rebecca. We all live in algorithmic prisons of our own making, and you are far from the only person in our industry who hasn't heard of this hack. Moreover, this happened just before the discovery of a vulnerability in PHP, and "Black October" reached a new level. Regarding the attack on Open-Code, we do not yet have any public information about who the attacker was. We have some assumptions about motivation. It was about 48 hours of intense attacks by agents looking for third-party API keys.

This is a stark contrast to the Hugging Face incident we experienced over the summer, where, to our credit, we were able to recover, and it appears Open-Code was able to as well, but we had no user or financial data compromised. And here-yes, financial damage was caused Yes. The things that bother me are that what makes an open community vulnerable is also what makes it special. We must be trusting. We want the average enthusiast in their basement, who has no institutional affiliation, to be able to contribute. I mean, they are part of the people who created the internet. It's like Wikipedia examples, it's just amazing how few people support Wikipedia's life. That's the beauty of open contribution. And we also

have.so it's the trustworthiness and the wealth of information and data that we have. For example, at Hugging Face we have these really interesting datasets to evaluate. Open-Code has these attractive third-party API keys. The combination of this makes us prime targets. Yeah, I mean, there's a gap in protection here, right? So where do you draw the line between protecting an open ecosystem and undermining the openness that makes it valuable? This is what we are thinking about now: where is this compromise? What kind of conversations do you have at Hugging Face? Is there anything you can share?

What exactly are you racking your brain over? I mean, the discussions have gotten to the point where a lot of people.we're responding to our community. We see such a huge interest in obliterated models. This is a Pandora's box, a very risky topic, and they really strike fear into a lot of people. Briefly explain to our listeners what a "destroyed" model is? Yes, that sounds like such a funny word. " Destroyed" models are models that have had all their safeguards removed, and while some of these open-source models are already quite liberal in their waivers and restrictions, you can pretty easily tweak them, remove those safeguards, and re-upload them to Hugging Face.

This is not a violation of our terms of service, and we keep it in part because we get a lot of feedback from cyber defenders and researchers that this is their only option for research or protection. Yes. Yes, I also changed my mind a bit about obliterated models after talking to cybersecurity experts, because, you know, I wrote about it. There's a company called obliteration.ai, I think that's what they're called, and they basically just host or give you the ability to " obliterate" any model. So you just don't have any protections, and you can imagine how terrible that is, right? Of course, you don't want to make it easier for attackers, but at the same time, it makes it easier for defenders,

and I guess the argument is that attackers will do it anyway. they do it anyway. It's important for defenders to know exactly how they're going to be attacked, right? In fact, it also works very well for red teaming. You know, it's really useful to have a "destroyed" model when you're trying to recreate the craziest version of a hack to test your systems. So yes, it's complicated, but if open model infrastructure becomes more expensive and risky to maintain, who, I wonder, will pay for security? Will this potentially give large funded platforms an advantage over smaller open source projects?

I mean, the advantage in the level of product support that you get from closed-loop labs is just going to exist. Most of them, except for really well-endowed resources like Alibaba, you know, we're still a 200-person company while we go through the acquisition process. So we won't be able to provide that level of support, but what people need for cyber security in a time when there's a lot to worry about - I'm pretty nervous about the state of the internet. Mhm. We need to have a certain level.It's a compromise. We need access to people who know their stuff, who are faced with rejections from the most modern, cutting-edge models.

Unfortunately, there is a dichotomy between cybersecurity professionals and AI security professionals, and both sides have good intentions. But many security professionals like open source because of privacy, because of considerations around local deployment, because of what they can do with it, and indeed have an ideological preference for open scales, which creates a new type of debate about "open" versus " closed." But the urgent need from our perspective, from where I work at Hugging Face, is that we need to arm the defenders yesterday; and if that means giving people a more open model, it seems like that's the balance we're going to have to find. Is there anything that could convince you otherwise?

For example, is there a type of attack that might make you think: really, I don't want attackers to have access to, you know, powerful open models? I mean, it's a matter of visibility of who's attacking and who's defending. I wish we had better methods. This is what we have been asking for at the policy level-better standards and incident reporting. We're very happy that we were able to be as open as possible, but it was quite difficult to figure out exactly how to report this and to whom. So as people on both the front end and those on the defense side who are victims of attacks go

through this, I would ask them to be as open and honest as possible about what they are going through, if they understand what model was used. As for your direct question about what would change my mind, I think it's clear evidence that the risks of using open weights in attacks outweigh the opportunities for defense. And in many ways this is limited by computing power. I don't think that non-state actors currently have enough computing resources for attacks of this scale, although that may change in the near future. But given our level of urgency, the need for protection-even if you've seen those European grocery stores, it may not be the highest priority for critical infrastructure

-but for businesses, protection from hundreds of thousands, if not millions, of losses seems important. Would you support any policy that would restrict, well, if not access to models, then access to computing power? I think we're catching up a bit here Token and computational efficiency has taken a huge leap forward in the last few years, especially after Deep Seek R1. I don't think this is the correct indicator. The general guidance we need is on what level of access control is appropriate, because I'm afraid we'll start implementing a certain level of access control for open-scale models, and then nobody will have access to anything.

Yes. Good. Um, you know, a lot of startups are building their solutions on open models and open source infrastructure. Um, we touched on this a little bit last week when we had Wiz visiting. Um, a significant portion of them don't have specialized security teams. I just want to hear your brief thoughts on this before we finish. How can small startups protect themselves now? For example, what are they perhaps underestimating in the current state of threats? It is clear that they are not yet ready to work with security teams, even with some level of familiarity with open sources.

My team is working on a plan for the next few weeks. I wish we had done this sooner, but in the meantime, there are great cybersecurity startups like Sam Graham that have openly available documentation. So use the resources you have to the maximum. This is the beauty of the open internet. And that's fine, it's fine to rely on what's convenient for you, but it's up to us in the open community to lower the barrier to using cybersecurity tools that you can rely on. Okay, great. Well, I think we've run out of time on this. Where can our listeners find you and follow your work online if they want to, you know, drop by, say hello, or write to you privately?

I'm not as skilled with private messaging as I might have been before ChatGPT. You can find us by creating an account on Hugging Face. We are everywhere on Hugging Face. I can most likely be reached by email at ire***@***.co. And continue to follow us on all your favorite social media platforms under the name Hugging Face Okay, thank you again, Irene, for joining our show. It was nice to see you. To our listeners: you can find me on Twitter and LinkedIn, and Equity on Threads and Twitter using the hashtag @EquityPod. See you next time. Equity is hosted by TechCrunch senior reporters, producer-Theresa Lo Concannon, editor-Kell. Subscribe on YouTube or wherever you listen to podcasts, and

find out about upcoming events at techcrunch.com/events. Thanks for listening, see you next time.

More Tech Transcript