Do we need yet another new AI platform? Would new foundation models stand a chance in today’s market? Will OpenAI, Anthropic, etc be the new Tech winners … or will they just get acquired, at some point?

Navigation:

Our co-hosts:

Our show: Tech DECIPHERED brings you the Entrepreneur and Investor views on Big Tech, VC and Start-up news, opinion pieces and research. We decipher their meaning, and add inside knowledge and context. Being nerds, we also discuss the latest gadgets and pop culture news

Bertrand

Welcome to Tech Deciphered episode 59. Again, this will be an episode on AI. This time we make the case for not starting yet a new AI platform and not investing in one either. What’s happening in AI? I guess all of you have heard about all the massive investment made by some of the leading players, OpenAI, Microsoft, Google, Meta, XAI, Oracle, as well as even some smaller startups from Anthropic to Mistral.

Bertrand

Of course, Amazon has built its own frontier model as well as is providing infrastructure to a lot of players. But isn’t there already too much capital going into AI platforms and isn’t it going too fast?

Nuno

Well, we think probably yes. That’s going to be the thesis of our discussion today. Some of the startups you mentioned, by the way, they’ve raised a ton of money. I mean, Anthropic is well above 8 billion. Mistral is at a couple of billions as well, right, at this stage?

Bertrand

I think we are closer to a billion.

Nuno

It’s around a billion. Yeah, around a billion, several billion for Anthropic. Then there’s a couple of companies we will talk about later that raised a few billion that are no longer independent. It feels like too much. It feels like there’s a gold rush of sorts. It is probably a good analogy. A lot of investors putting capital into it.

Nuno

But at the same time, it’s being matched by the big players, Microsoft, Google, Meta by pushing LLM, XAI, OpenAI, we could argue, is it a big player already or not? But they just raised a ton of money as well. At 150 billion valuation, right? I mean, it’s like…

Bertrand

Yes, 150 billion valuation.

Nuno

It’s like, okay. They’ve already said, well, we are going to continue raising money. We’re going to continue spending a lot of money as well. It feels like a gold rush. It feels like there’s too much capital deployed into it. Something’s got to give, someone’s going to fail. Maybe that’s a good segue to the fact that we already start seeing some moments that are quite interesting.

Bertrand

Maybe to OpenAI, to jump back, to be fair, they’ve reached more than 3 billion of annualized revenues, and that was based on numbers from June. This is very high valuation for sure. But it’s also coming with one of the fastest growing business ever in terms of revenues.

Nuno

Yes. We will see if it’s sustainable, the business that they’re building, because there’s no competition. People are using other services out there in tandem. I see a lot of people, for example, using perplexity rather than ChatGPT. We’ll see if Gemini catches up at some point or not. I would argue they have grown very fast, but they have not faced as much competition as they’re probably starting to face right now. At least that would be my thesis.

Nuno

In any case, we start seeing signals of companies that go to the market and get bought out, and they get bought out for amazing amounts. If we look at Google and there, I would argue, acquihire of Character.ai. It seems to 2.5 billion to get the team and I guess the tech as well. But it feels a bit like acquihire. They’re trying to get the CEO back into the fold at Google. So interesting, I feel.

Nuno

The Inflection one was at least in the highest hundreds of millions. We heard 650 million for the IP licensing deal from Microsoft to get the founder and CEO to join Microsoft to head AI. Let’s say it’s around a billion and let’s call it even, but that’s like another gigantic acquihire.

Nuno

What gives? And it’s a little bit like it feels a little bit like free money. I’ve had a bit of a chance to think about this. It’s not just that it feels a little bit like free money. It’s some of these acquirers, like the Microsoft of the world, et cetera, might still be overvalued themselves, question mark, and therefore they’re basically putting capital because they have the capital to deploy at this stage.

Nuno

But it feels a little bit abusive. It raises all these concerns. If I’m an investor, and I’m backing a company in the AI space that is doing yet another foundational model, or that it’s playing at the limit between infrastructure and platform, and I know it’s going to be capital-intensive, but I’m one of the early investors and I hope that they’ll raise a bunch of money, well, there’s no guarantee that at some point they won’t just get bought out.

Nuno

Maybe the early investors will make a pretty decent return or a pretty good return, certainly a very fast return. But maybe those who are coming later won’t make a ton of money out of it because the company end up being just acquihired.

Nuno

It may also uncover a hidden secret, which is for you to scale, this is so capital-intensive in terms of consumption of compute, storage, other parts of infrastructure, etc, that at the end of the day, only a big, big guy can win. Therefore, either you raise tens of billions of dollars or you have no chance even to compete. What do you think, [inaudible 00:04:49] ?

Bertrand

Yeah, I think it’s a mix of things that are happening right now. One is, as we know, the FCC has been blocking acquisitions from the big players. What we are seeing is companies that might have been properly acquired in the past, but suddenly the bigger companies cannot really acquire any more. So they have to find ways around current regulations or at least how the regulators is seeing the regulations and is using its power to block acquisitions and therefore are going around by doing this type of very weird acquihires with separate licensing deals.

Bertrand

We’re in a pretty strange position. And it’s not just AI. In general, there has been much less acquisitions from the, can call them the Fabulous 7, if you want, because their ability to acquire has been very restricted. I don’t think, of course, it’s a good thing for startups. So it’s raising a lot of questions of how as a startup, can you keep ultimately providing liquidity to your business, especially when, as we know, there’s also very little IPO these days.

Bertrand

So it’s a very weird combination of you cannot really do an exit through the bigger guys, and only the bigger guys can afford to spend billions and billions in an acquisition, by the way.

Bertrand

If you don’t provide the IPO opportunity, then what do you do? Maybe that’s what we are seeing as well is what you end up doing is doing these weird acquihires, at least for the businesses that are not worthy of keep going at the rate they started.

Nuno

Yeah. Obviously, there’s two sides to this There’s the investment side of the play where if I’m an investor, I want to return, and therefore, I want the company to get acquired at some point, or for the company to IPO, that’s the ideal scenario. But obviously, I wanted it to be acquired at a valuation that is beneficial to me.

Nuno

If I came into a round where the company was valued last, I don’t know, 2 billion and the company gets bought for 2 billion, that’s not much of a gain, right? There’s that element of the piece of the puzzle. Then there’s the element you’re mentioning, which is really the M&A perspective of the big guys. Can I really acquire? Will I be allowed to acquire this asset and this team and this company?

Nuno

But in some ways, they’re doing really under the disguise of acquihire. I’m getting the talent, the core talent in. For example, the deal with inflection was that, they got the talent, and then they did a licensing agreement with the company, though. I don’t think formally they acquire the company. There’s obviously other natural activity going on. AMD bought a startup, Silo AI for 665 million.

Nuno

There’s activity going on not just on the platform foundational model side, but also around the chip capability side, infrastructure side. It’s a little bit truly like a gold rush. If you remember in the gold rush, there was the whole pics and shovels. Our friends from Levi Strauss, the jeans makers were attached to that because people needed more comfortable trousers to do work on or in, rather. That’s how Levi Strauss became a big name.

Nuno

In some ways, there’s a lot of things happening around the ecosystem, not just in the core platform foundational models side, but also around infrastructure that needs to change. Some would say it’s justified. There needs to be a ton of investment. My thesis is there’s too much money being poured into it, and there’s going to be a lot of people crying at some point, particularly on the investor side, because something does have to give.

Nuno

It’s impossible that we will have at scale, assuming very high capital intensity, 10, 15 players fighting for the same broad foundational models. I don’t see that happening.

Bertrand

I think there are different perspectives here. One is as consumers, all this investment in AI or businesses that are going to benefit from AI is going to be great because I see the performance of all these models keep improving. I see their capacities keep improving. I see their abilities. I mean, the way now it’s multimodal models. It’s not just text, it’s image, it’s not just image, it’s audio, it’s not just image and audio, it’s video.

Bertrand

So the ability of this model keep improving. I mean, let’s not forget the huge gap between ChatGPT 3 versus 4.0, and now we are even at 0.1. So there is a significant improvement in what you can do with AI, thanks to a totally unprecedented investment, investment in building out infrastructure. So GPU farms in order to build these foundational models.

Bertrand

Yes, to your point, will we get so many companies succeeding? I don’t know. My impression at this stage, we have an NVIDIA with some arms dealer of the AI world. It might be your third party. You talk about Levi in the old days selling tools to everyone. Nvidia might be that, selling tools or maybe even more aptly an arms dealer.

Bertrand

When you see about all these companies, these are massive companies that are generating billions and billions of quarterly positive cash flows. Who can afford to invest? You could argue, isn’t it a good investment from their perspective? Because when you generate so much cash, either you make acquisition, but right now you cannot, or you give money back to your shareholder, but then is it going to go to your competitor at the end of the day?

Bertrand

And instead, you might think, you know what? I have to preemptively build the best models out there. Because if you don’t, somebody else is going to do it, you might depend on your competitor to supply you key technology. I think if you take Apple, for instance, they are fine with that. They are fine to use Google for their search. Google pays them a lot of money. They are fine to use OpenAI. They don’t give them money, but at least they can outsource to OpenAI the more complex AI queries.

Bertrand

But other players like Google, like Meta, like X, like Oracle, might not want to outsource to anybody else. They want to control. Even Microsoft is controlling in some ways, thanks to their OpenAI partnership.

Bertrand

They have a lot of control about the infrastructure, about the models. I think these companies in some ways have no choice. They don’t want to depend on one of the other magnificent 7 for a core key part of technology that will cause them potentially to be competitive disadvantage if they don’t have the technology. Do we rationally need alpha dozen or so spending billions of CapEx? I don’t know.

Bertrand

But I can see how for some of these players, there is simply no other alternative. You cannot depend on core technology on your big competitors.

Nuno

Yeah, but indeed, that’s maybe the perfect segue to talk about the fact that the big guys are investing themselves, and they’re investing in all sorts of ways. We see Microsoft being the largest investor, to my knowledge, in OpenAI, but at the same time doing their own things in-house, from custom chipsets to their own large language models.

Nuno

Google obviously is trying to play a number of roles as well, to our knowledge, Amazon as well. We have Facebook with obviously Meta doing the LLM play, which seems to be more the open play. We’ve discussed it in previous episodes. So it’s not just that you have well-funded or very well-funded startups raising bunches of money at some amazing and ludicrous valuations like OpenAI, but you have the big guys also putting a ton of money into it.

Nuno

As we also mentioned in a previous episode, analysts kicking them a little bit only for them to then go back up their stock. But it feels really like an arms race. And in this arms race, the legacy guys, the guys who own these markets, who own these channels, the Microsoft, the Facebooks, the Googles of the world, are putting money directly themselves in their own product lines and their own platform development.

Nuno

So it’s not just they’re putting it into startups, they’re also putting it into their own. If all of that wasn’t enough, as we just mentioned, they’re also doing M&A whenever they feel they need more talent, or they need more.

Bertrand

Oh, acquihires.

Nuno

Correct. It’s acquihires that are, as we said, gigantic. So it feels like a market where in the end, the big guys win because they have a bunch of capital. There’s multiple companies that are over a trillion in valuation. They have a ton of resources and they win. Who wins besides them, I guess, is the core question at this stage.

Nuno

But it’s very difficult for me to see a world where we have three, four new pure players instead of that layer of AI platforms or platform as a service that will emerge and will compete head-on with the Googles of the world, et cetera.

Bertrand

Is it so surprising? I mean, if you take mobile industry, for instance, who ended up winning mobile, not Microsoft for sure, but it was Apple and Google. They were some of the big guys. I mean, Apple a bit relatively smaller, but my point is that no new players emerge, at least at the operating software layer. We had some, of course, on the hardware side, especially from China.

Bertrand

But Samsung was another winner. I’m not so surprised that you don’t have a lot of third-party winners. I mean, OpenAI seems to be Very successful coming, I can’t say out of nowhere, but definitely growing suddenly very fast. But yes, is there space for others? Maybe not.

Nuno

For me, the surprise is not that. The surprise is that there’s so much capital being thrown at companies that are likely going to be, as we just I mentioned, acquihired.

Bertrand

You talk about the smaller guys.

Nuno

Yeah. Well, it’s not just the smaller guys. I don’t know if Anthropic will have a future. I’m not even sure OpenAI will have a future. OpenAI has a ton of money from Microsoft in there. At some point, someone might say, we’ll see if they’re an independent company or not.

Nuno

Even OpenAI, who is, just to be clear, at the 157 billion valuation as of the last round, I’m not sure. For me, the surprising thing is not that the big guys are to win. The surprise is that a lot of people are betting that the big guys are not winning. If we’re seeing currently, Acquihires at 2.5 billion, 650 million licensing deals for Acquihires, etc, it’s not going to stay forever, these kinds of acquihires.

Nuno

At some point, it’s not going to be a couple of billion for the acquihire. At some point, it’s like, you’re going to die, I’m going to get you for tens of cents on the dollar, or cents on the dollar, or even less than that. I will get you at whatever because you’re going to die.

Bertrand

It’s interesting because look at OpenAI. They have had a lot of executives leaving the business the past few months since Sam Altman is back in power. And guess what? Everyone wants to finance these guys to start a new business. I mean, the co-founder, Ilia, of OpenAI just started a new business, very well financed. Money seems to be there for sure at the SECHECH.

Nuno

Yeah, but it’s there right now, Bertrand. We’ve seen these bubbles before, right? I mean, this is just a gigantic bubble at the speed of light.

Bertrand

Yeah, what’s interesting is that if you take OpenAI, I’m not sure we have had many businesses that went up going so fast, so quickly, get such a high valuation without being default alive. Because if you take Google when they went IPO at similar scale, I mean, they weren’t default alive. They had no problem running the business. Yes, they were investing in faster growth, but they were definitely default alive. Nobody could touch them at this stage.

Bertrand

Here it’s a very different story. OpenAI is not default alive. They need additional financing and there’s insane competition at their heels. It’s a different story. You’re right. Can OpenAI keep their lead? I don’t know. For the smaller one, at this stage, I’m more a believer that you have to have a very, very different value proposition as a smaller player on building AI platform.

Bertrand

If you take Mistral, for instance, the positioning is we are doing It’s a European AI platform. It’s made in Europe. We are following, we try to be an AI champion. We try to make sure that European governments and others give special funding, buy the product. They are playing the game that way. Might work for a bit. I don’t know how long it might work given the investments needed.

Nuno

At the end of the day, is OpenAI going to be the next alternative? Maybe, I don’t know. Maybe I’m just being overly negative on it. They have huge revenues, which is amazing. They have huge costs as well, which is also amazing.

Bertrand

Bigger cost.

Nuno

Profitable business it is not yet. I’m like, I don’t know. Is it the pass-through to our friends at NVIDIA and other areas of the infrastructure layer? I don’t know. What we know is NVIDIA is still making money. That we do.

Bertrand

Yes, but I must say I’m a big fan of NVIDIA because these guys have been behind so many revolutions from GPUs, I mean, for graphics to now, super computing, accelerated computing, as they like to call it. Now, big scale AI computing. I mean, they have built some of the core foundations that everyone is using in AI. I mean, we thought NVIDIA, I’m not sure we would be where we are. I’m certain we will not be where we are. We will have lost, I don’t know, five years plus.

Bertrand

So these guys have been very helpful behind the scene. So personally, I’m a big fan of NVIDIA, but you have also to admit that it took OpenAI with ChatGPT to really change to change the game in terms of LLM and beyond generative AI at scale.

Bertrand

Maybe another point to share some of how big all of this has become. I mean, we are right now in the past few weeks facing potentially a renewal of nuclear energy in the US. I mean, I’m very amazed because being myself a big fan of nuclear since I care about it, so maybe since the past 25 years, I’ve always been shocked at what the German did with their nuclear energy, what California did as well, and some other states.

Bertrand

To see that finally, nuclear is back on the table because the big guys who care and are making billions of investments suddenly realize that, you know what? There is no better way to power a data center because you don’t want your data center to depend on solar or wind. It’s amazing. It’s a very, very big positive, but it also shows the scale of the investment. I mean, these guys are restarting a closed nuclear reactor, are working on building new reactors.

Bertrand

I mean, we’re talking about Amazon just a few days ago, Google, Microsoft, restarting the Three Mile Island, Power Plant, Sam Altman pitching deals for dozens of nuclear reactors dedicated to AI. This is a huge, huge turnaround. Also, it’s showing the scale of the AI investment. You could argue this investment in energy will also benefit any way computing in general.

Bertrand

Even if the AI opportunity doesn’t scale as much as some will hope or expect, this will be used by data centers at large, not just in an AI context.

Nuno

You mentioned earlier before we switched to the infrastructure and hardware side, but you mentioned earlier, rest of the world, right? Obviously, mistral, playing the European angle, etc. We can’t forget, obviously, our friends in China, Baidu, having launched Ernie, Alibaba with Tongi Qiangwen, which is [inaudible 00:20:21], I think is how they say it over there. Tencent with Hun Yuan, ByteDance with Dobao, and Huawei with Pango.

Nuno

The world that we’re talking about is more the Western market US world. We’ll see if Europe somehow ends up in a different position, which is a little bit, I guess, the mistral angle in some ways. China is definitely going to end up in a different way. But in all of these scenarios, I have difficulty in seeing how new players will emerge that will take over the previous players, the incumbents in that.

Bertrand

China is an interesting question regarding Nvidia chips. As you know, US government put restrictions in term of selling the latest chips to Chinese companies. That one is also always in question. I mean, can China keep growing its AI industry? I would guess so somewhat, but can it be truly competitive to the West, that I’m less sure.

Nuno

Well, yeah, Huawei would want you to believe that they’re competitive, but they have some vested interest in saying that with their Ascent chips.

Bertrand

No, I mean, today, they’re absolutely not competitive without Nvidia chips. There would be no AI industry in China without Nvidia chips. If it keeps going worse at some point, it might truly delayed by years, if not a decade at some point, if Nvidia chips stop going to China.

Nuno

A bit more optimistic about the capacity of, let’s call it benchmarking that China has and being able to, at some point, emulate performance. But we’ll see. I’m sure there’s plenty of money being thrown at China as well to making it work.

Bertrand

I don’t think it has any chance without Nvidia chips. For me, the only question is, will Nvidia chips keep flowing to China? Will the spice keep flowing to China? I don’t know.

Nuno

We’ll see. To go into precisely hardware and infrastructure and how it will change. There’s a lot of discussion around the fact that what got us here is probably not what’s going to get us to the next level. That’s very heavy-duty training, the building of large language models, etc, will be now really much more focused on inference learning, and therefore, the type of compute we’ll need will obviously be different for inference learning. Do you have any thoughts, Bertrand?

Bertrand

Yeah, I mean, for me, what’s fascinating is a change from training, retraining that we have been seeing as a key of building LLMs and somewhat much thinner layer of inference that is very, I don’t want to say dumb, but quasi dumb and just outputting as fast as it can what looks like the best token.

Bertrand

To a new model where actually inference is going to do way more than a simple prediction. Some call that level one versus level two, level one being very instinctive answer that humans have, actually, you have very quick instinctive answer to some issues, either from looking at the road and seeing a truck coming at you or answering very, very basic questions like how much is 5 plus 5. You know that you don’t have to think about it.

Bertrand

It’s a level 2 type of approach to thinking where there is way more inference put to work. This has been best exemplified today by OpenAI 01, called before that internally, strawberries or Q-Star. This one, OpenAI 01, is showing that if you put more efforts in inference time, then there is a dramatic improvement in the quality of the answers because the system is truly thinking about answering the best way.

Bertrand

This requires a dramatic change in infrastructure because suddenly you might actually put more efforts in inference infrastructure versus training. So it might open the industry. So of course, NVIDIA is not holding still. They keep improving their inference capacity. Their latest chip have improved inference capacity dramatically. And you have seen a lot of new players like Grok that are being focused on inference only that might benefit from this shift. Some other players like AMD.

Bertrand

I think there is quite a lot of change coming in term of where we are investing in term of infrastructure with more potentially going to inference. Inference, interestingly enough, is pure Cox, because this is the stuff you need to run the service. So this one would be pure Cox.

Nuno

On the infrastructure side, we’ve historically not invested a lot in that space, but we recently have made two investments. One we can’t announce yet, but it’s in the networking part of the stack. Obviously, as you can imagine, networking has to change going forward for a variety of important reasons.

Nuno

The other one in a company called Vivum.ai, which sits at the intersection of platforms and systems on chip and a variety of other interesting trends that we see in the space, but really more around the next wave of what we believe are the methodologies of the future for artificial intelligence, in particular the use of evolutionary computation, dynamic neural networks, and a variety of other methodologies that we think are the future rather than the current present of where we sit.

Nuno

Definitely, we feel there’s the big gorillas at the table, the Broadcom of the world, the NVIDIA’s of the world, et cetera, are there. They’re still the incumbents very much in their respective spaces. But there’s obviously opportunities to go after really significant slivers of the market that might warrant some development, some innovation.

Bertrand

For me, I remember being very impressed by NVIDIA move when they acquired the Melanox. It was in 2019, and Melanox was close to 7 billion. Mélanox was a big player in Superfast, Ethernet, and InfiniBand. I think NVIDIA really had a lot of foresight in term of how to keep investing in the chain. And for me, what’s interesting with NVIDIA is that they keep investing in the big picture. They are not just focused on the chip itself, but they are focused on the fabric between the chips. They are focused on building the full scale rack from A to Z.

Bertrand

So they are really focused more on selling new systems. It’s a software side as well with CUDA. From my perspective, very smart at playing that game of expanding the scope and making sure customers get a really good ROI and in a way a simpler, very well-integrated infrastructure. Of course, there will always be players investing in more point solution and actors who are interested in some best of point type of solution.

Bertrand

But in some ways, NVIDIA is becoming the apple of the AI space, providing you a very well-integrated hardware, software, different layers of the stack. If you an NVIDIA product, you know you have an unmatched level of integration.

Nuno

Until now, we’ve basically framed it as very unlikely that these broad foundational models, platforms, so to speak, platform as a service place will be fundamentally disrupted. The big guys are likely to win across the world, even in parts of the world, where we will have definitely some silos of offerings. If that’s the case, is there some hope? Is there a next wave?

Nuno

Is there a next wave of AI that we haven’t seen yet that will be best represented by other methodologies where the current winners might not be the winners in the future? We, certainly at Chameleon, we believe that there is a next wave, that there is a next piece of the puzzle that needs to be built.

Nuno

As I mentioned before, we have made investments around the infrastructure side and platform side. We believe there is a new order coming, evolutionary competition, which has been with us for several decades. It’s not a new thing. Just to describe a little bit what evolutionary competition actually is, it’s a little bit like thinking through algorithms like biological evolution. You have algorithms that spawn and hopefully become better than their predecessors, and there’s what would be effectively a natural selection process.

Nuno

The best algorithms will survive, and then we’ll spawn the next generation. We’ll have generations of algorithms coming out, etc. Some people mentioned that it’s more dynamic neural networks, and that’s a little bit where we’re at and where we would go to the next level. Neural networks are improving through time.

Nuno

So expect that there is a dynamic aspect to what we have in the world today. We believe there’s opportunities still in that space. Obviously, there’s tremendous question marks on capital intensity and whether this is finally the time for a revolutionary competition. We’ve had decades of saying it is, and it really hasn’t happened at scale.

Nuno

But we believe that’s one area that certainly is still a little bit ready for the picking and where we will see, and I have been seeing, at least in the last year or so, quite a lot of investment.

Bertrand

Yeah, no, that makes sense that there might be alternative approach to transformers system. I would say at this stage, me, where I’m more focused is innovations beyond the basics of how GPTs are trained. For me, the approach that is more focused is on inference time or agent-based system to compensate for the lack of thinking from GPTs is quite critical and transformational.

Bertrand

I think, truly ChatGPT01 is transformational, and we will see more in the coming three, six months. We will see more AI platforms moving with our own chain of thoughts. It looks like as well there has been some very good improvement in an agent-based solution that have been put in place in order to compensate for hallucinations, for issues with answers coming from regular GPTs.

Bertrand

I can see some layers above, either strictly inference or agentic, that are providing a next level of quality that is badly needed because let’s still remember ChatGPT 3, it was very poor. Hallucinations were everywhere. It was fun to try to use it, but practically for a lot of business answers it was just too limited. With for all, we saw really big improvement, but still and pretty easy to trick it.

Bertrand

So me, I’m quite excited with that next wave of inference layer, research through a chain of thoughts like a one or Atlantic-based solution. There might be startups going in that space and standing a chance. I think personally for startups, the opportunity is not in the AI platforms per se. I think we are reaching a stage where it’s a bit like databases. You will have a few big major database. You have a few major database.

Bertrand

Some have non-open code base, or he calls SQL Server. Some have open code base and business model from MySQL to Postgres and others. Some in between as a business, you rely on a multitude of database vendors, and having multiple in front of you give you the insurance that you can build a business on top of database and adjust your business and not be dependent on any one vendor.

Bertrand

I think we are reaching that stage with LLMs, with foundation models, where as a business, you can build with the insurance that you can pick with different models based on your needs. You can switch them out as well based on your needs, but also based on their evolution, the same way you could with databases.

Bertrand

It can be painful, don’t get me wrong, to change database or to change LLM, but you can do it which increase your ability to invest because you don’t want to be dependent on one vendor. You want to know that there is some level of competition going forward, that you won’t be held hostage by your LLM vendors.

Bertrand

As a result, I believe that that plurality of solutions from some big vendors, that plurality of options from free private to open, gives finally opportunity for businesses to invest way more in AI, as well as startups to invest more on the middle layers, on the application layers. I think that startups, successful startups, are actually going to be a layer above the foundation layers.

Bertrand

Startups will be focused, from my perspective, on vertical opportunities. They will be focused on being that layer between your very generic foundation layers that can work with anything to fine-tuning that layer to very verticalize opportunities, very verticalized industries, business models. I think that’s where we see opportunities for startups, because the open AI of the world, the Google of the world, they don’t want to focus on one specific industry, one niche.

Bertrand

They want to focus on solutions available at scale, sell API access, sell tokens. But I truly believe that for this AI technologies to be truly successful in a consumer environment or in a business environment, we need more specializations. I think that’s where startups will flourish because they won’t need to invest this billions in infrastructures, in GPU trainings. It will be millions, thousands of millions of investments for sure. But I that they will be able to find their spot industry by industry.

Nuno

It brings us back to something that I think I referred in previous recordings and previous podcasts, which is effectively, then that looks a little bit like an app-based economy. You develop apps that have very specific usefulness for a specific audience, be it consumer, B2B enterprise. You use AI. It’s AI first, but it’s not like you’re developing your own foundational models.

Nuno

It’s basically you’re riding on top of existing infrastructure I mean, we certainly internally in our firm use different models for different things already. It’s not just one size fits all thing, but we see that there is an emergence already of what would be considered really an app economy that has very specific point solutions for specific audiences.

Bertrand

Yes, totally. I truly believe it’s going to be like an app-based economy. In a way, OpenAI has been an exception to this rule by having a very direct to consumer offer with ChatGPT. For focusing on that background layer. So there might be a bit of an exception providing horizontal enterprise-scale type of APIs and services, as well as a very direct to consumer type of offering.

Bertrand

But again, as we see, even their offering, it’s very generic. They don’t try to adjust to different needs or anything. They have a very smart approach to keep adding key core feature to the platform, like multimodal approach, of course, better model over time. But again, it’s very generic each time. It’s not vertical oriented, it’s not focused on your specific needs as a business.

Bertrand

So I think, again, there will be plenty of opportunities for startups to go vertical. To your questions around investor, I think the best investments will actually be at this level for investors because fighting the Microsoft of the world is not an easy one. They have near infinite cash they can put to this. While at the same time, I think there would be smart opportunities that might be too niche for some of these guys, actually, at least early on.

Nuno

So is it too late to come in to disrupt for the big guys? The answer is “sort of”, right? In broad foundational models, platforms, in very big pieces on infrastructure, probably it is too late. But in things like the app economy that we were just mentioning, it’s just the beginning.

Bertrand

It is.

Nuno

In areas like middleware, pieces of the next wave of development around platforms and AI, maybe there’s still some opportunities. The same is still true also at the infrastructure and hardware level, where there might be some slivers of opportunities still which are natural for a disruptor to come in rather than an incumbent to fulfil time. The answer is a bit nuanced. Yet another platform, probably not. Not for the broad aspect, but then there’s a couple of other opportunities that emerge out of it, which honestly are also less capital-intensive, which is wonderful.

Bertrand

Yes, indeed. I believe it’s very exciting times for the tech industry at large. It’s a really brand-new wave and opportunities for the big guys, for sure. But I think even the smaller guys, but you have to be smart about your value and positioning in that space.

Nuno

In conclusion, this is the end of episode 59 of Tech Decipher, the case for not starting yet another AI platform and not investing in one. We discuss the landscape, everything that’s been happening on the acquihire side, what the big players have actually been up to because they have not been sleeping for sure, what’s happening in the infrastructure hardware side, what will take us to the next level in the next wave. Finally, the conclusion around whether it’s too late to disrupt the big guys or not. As you guys just listened, the answer is nuanced. Thank you, Bertrand.

Bertrand

Thank you.