Build by AI

The New AI Arms Race: When $40 Billion Bets Meet Open Source Disruption I 25th April

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 32:34
Google reportedly plans to drop $40 billion on Anthropic while DeepSeek's open source models are closing the gap with frontier AI. Meta is laying off 10% of its workforce to double down on AI, and AI-designed drugs are heading to human trials for the first time. We break down what this massive reshuffling means for the future of artificial intelligence and why the next six months could determine who wins the AI race.
SPEAKER_01

There are two possible futures unfolding right now in AI. One where the biggest tech companies use their massive war chests to lock up the best AI talent and technology, creating an impenetrable moat around artificial intelligence. And another where open source models become so good that all that money and all those exclusive partnerships don't matter anymore.

SPEAKER_00

Yeah. And what's wild is we're seeing both of these futures play out simultaneously this week. The stakes here aren't just about which company wins, it's about whether AI remains accessible to everyone or becomes the exclusive playground of tech giants.

SPEAKER_01

Because when you're talking about $40 billion investments and models that can suddenly match the performance of the most advanced AI systems, you're not just looking at incremental progress. This isn't just about chatbots and image generators anymore.

SPEAKER_00

Right, we're at this inflection point where AI is becoming real medicine, real business infrastructure, real creative tools that people depend on. The decisions being made right now about how this technology gets developed and distributed, those decisions are going to affect everyone. And I'm Sam Hinton. We've got Google reportedly preparing to make one of the biggest AI investments ever, open source models that are suddenly competitive with the best proprietary systems, and some fascinating developments in AI drug discovery. This is going to be a packed episode.

SPEAKER_01

Alright, let's jump right in with what could be the biggest AI investment we've ever seen. Now, if confirmed, this would be absolutely massive. We're talking about Google essentially betting the farm on Claude and Anthropic's approach to AI safety and capabilities.

SPEAKER_00

Dude, forty billion dollars. Let me put that in perspective. That's more than the GDP of some countries. This isn't just an investment. This is Google saying we think Anthropic might be the future of AI, and we're we're willing to pay almost anything to make sure we're part of it.

SPEAKER_01

But here's what I find interesting. Google already has their own AI models with Gemini. They've got DeepMind, they've got all this internal AI capability. So why go all in on Anthropic? What does this tell us about their internal assessment of the competitive landscape?

SPEAKER_00

I think it's a hedge, but it's also an admission. Google is looking at the AI landscape and saying we can't afford to be wrong about which approach wins. Maybe they think Anthropic's constitutional AI approach is the key to solving alignment. Or maybe they just don't want OpenAI and Microsoft to have all the advantages.

SPEAKER_01

Wait, but doesn't this create some weird dynamics, though? I mean, Google is now potentially Anthropic's biggest investor while also competing directly with them in the AI space. How does that work exactly?

SPEAKER_00

You know, it's complicated, right? But think about it like this. You know, Google would rather have a seat at the table with every major AI player than risk being left out of the next breakthrough. They're basically saying, if we can't beat you, we'll buy a huge piece of you.

SPEAKER_01

And for Anthropic, this could be transformative. That level of investment gives them the compute resources and financial runway to compete directly with OpenAI and stay independent from Microsoft's ecosystem.

SPEAKER_00

Exactly. And here's the thing: people are missing. This this isn't just about the money. With Google's backing, Anthropic gets access to Google's infrastructure, their cloud platform, their distribution channels. This could be what finally gives Claude the scale to challenge ChatGPT's dominance.

SPEAKER_01

But I'm curious about the timing. Why now? Google has been investing in AI for years, but this level of commitment to an external partner feels different. What changed?

SPEAKER_00

I think it's the realization that the AI race isn't slowing down. It's accelerating. And every month we see new breakthroughs, new competitors, new approaches. Google probably looked at their portfolio and realized they needed more shots on goal, not fewer.

SPEAKER_01

And there's the regulatory angle too. Spreading their AI investments across multiple companies might look better to regulators than trying to build everything in-house. It's like, look, we're not monopolizing AI development. We're supporting the ecosystem.

SPEAKER_00

That's a really good point. But here's what I keep coming back to. If this deal goes through, what does it mean for smaller AI startups? If Google is willing to drop $40 billion on Anthropic, how does anyone else compete for that level of partnership or investment?

SPEAKER_01

It could create a tier system, right? You've got the mega-funded players like Anthropic with Google backing, OpenAI with Microsoft, and then everyone else fighting for scraps. That's not necessarily healthy for innovation.

SPEAKER_00

Unless the open source models we'll talk about later really do level the playing field. But assuming this deal happens, what should people actually expect to see? Like what does $40 billion in AI investment actually buy you in practical terms?

SPEAKER_01

So for our listeners, what should they be watching for? How do we know if this investment is actually moving the needle?

SPEAKER_00

Keep an eye on Claude's capabilities over the next six months. If this deal goes through, we should see major improvements in Claude's performance, maybe better integration with Google services, and probably some aggressive pricing to grab market share. This could reshape the entire competitive landscape.

SPEAKER_01

And watch for talent movements too. Forty billion buys you access to the best researchers, the best engineers, the best product people. If anthropics suddenly starts hiring at an unprecedented rate, that's your confirmation that this deal is real and they're going all out.

SPEAKER_00

The other thing to watch is how OpenAI and Microsoft respond. You don't just sit there when your biggest competitor makes a move this big. I expect we'll see some announcements from them pretty quickly if this anthropic deal gets confirmed.

SPEAKER_01

Alright, so speaking of reshaping the competitive landscape, let's talk about Deep Seek. They just previewed their new V4 model, and the claims they're making are pretty bold. They say it has nearly closed the performance gap with current leading AI models on reasoning benchmarks, and it can process much longer prompts than previous generations.

SPEAKER_00

Okay. This is huge, and here's why. Deep Seek is open source. So while Google is spending forty billion dollars to get a piece of anthropic, DeepSeek is basically saying, here's a model that's almost as good as GPT-4 or Claude, and you can download it, modify it, and use it however you want.

SPEAKER_01

The long context processing is particularly interesting to me. Being able to handle larger amounts of text more efficiently, that's not just a nice-to-have feature. That's a fundamental capability that opens up entirely new use cases, right?

SPEAKER_00

Absolutely. Think about what you can do with really long context windows. You can analyze entire documents, have conversations that span hours without losing track, process legal contracts, research papers, code bases. It's the difference between having a smart assistant and having a smart assistant with perfect memory.

SPEAKER_01

But I'm a little skeptical of these benchmark claims. We've seen companies cherry pick benchmarks before to make their models look better than they actually are. How do we know DeepSeek is really closing the gap with frontier models?

SPEAKER_00

That's fair. But here's the thing. Deep Seek's previous models have been genuinely impressive for open source. The V3.2 was already competitive with some commercial models, so an architectural improvement that makes V4 significantly better isn't that hard to believe.

SPEAKER_01

And the fact that it's open source changes everything about how we should think about AI competition. Even if it's only 90% as good as GPT-4, but it's free and modifiable, that might be good enough for a lot of use cases.

SPEAKER_00

Exactly. And this is where that Google anthropic deal starts to look different. If open source models are getting this good, maybe spending forty billion dollars on proprietary AI isn't the winning strategy. Maybe the future is about services and applications built on open models.

SPEAKER_01

It's like the old Linux versus Windows debate all over again. But for AI, the question is whether open source can move fast enough to keep pace with well-funded proprietary research.

SPEAKER_00

And based on what we're seeing from DeepSeek, the answer might be yes. If they can keep this pace of improvement, we could see open source models matching or exceeding proprietary ones within the next year or two. That would completely change the economics of AI.

SPEAKER_01

But let's talk about the architectural improvements they mention. What does that actually mean? Are we talking about completely new approaches to how these models process information?

SPEAKER_00

From what I understand about the V4 preview, they've redesigned how the model handles large amounts of text. It's not just about cramming more tokens into the context window. It's about processing them more efficiently so you don't get that degradation in performance you see with really long inputs.

SPEAKER_01

That's actually a huge deal for practical applications. I mean, most people don't need their AI to be 5% better at reasoning. But they definitely need it to not forget the beginning of a conversation when they're 50 messages deep.

SPEAKER_00

Right. And this gets to something important about the open source versus closed source debate. Open source models often focus on solving practical problems that users actually face, while closed source models sometimes optimize for benchmark performance that looks good in research papers but doesn't translate to real-world use.

SPEAKER_01

So what does this mean for businesses that are trying to decide between using a service like ChatGPT or Claude versus running their own open source model? The calculus is getting more complicated.

SPEAKER_00

It really is. If DeepSeek V4 delivers on these promises, you're looking at potentially having 90 to 95% of the performance of Frontier models, but with complete control over your data, no API costs, and the ability to fine-tune for your specific use case. For a lot of businesses, that's going to be compelling.

SPEAKER_01

And there's the network effect consideration too. The more people who use and contribute to open source models, the faster that they improve. Meanwhile, proprietary models are limited by the resources of their parent companies, no matter how large those companies are.

SPEAKER_00

Exactly, though.

SPEAKER_01

Although to play devil's advocate, those proprietary models have access to massive compute resources and exclusive data sets that open source projects can't match. Money still matters in eye development.

SPEAKER_00

True. But we're also seeing more efficient training methods, better data utilization, and new architectures that reduce the compute requirements. The gap between what you can do with massive resources versus modest resources is shrinking.

SPEAKER_01

Now let's shift gears to Meta, because they're making some pretty dramatic moves too. They're laying off 10% of their workforce as part of what they're calling an AI push. This comes after we've seen them recruiting talent from Thinking Machines Lab and other AI organizations.

SPEAKER_00

Man, Meta is really going all in on AI, aren't they? But laying off 10% of your workforce to focus on AI, that's not just a strategic shift. That's basically saying everything we were doing before is less important than winning the AI race.

SPEAKER_01

What's interesting is the timing. This is happening right as we're seeing Google make massive investments in anthropic and open source models getting more competitive. It feels like Meta looked around and said, we need to move faster or we're going to get left behind.

SPEAKER_00

Right. And think about Meta's position. They've got Llama, which has been really successful as an open source model, but they're competing against OpenAI's Chat GPT dominance, Google's Gemini, and now potentially a supercharged anthropic. They need every advantage they can get.

SPEAKER_01

But here's what worries me about these kinds of mass layoffs. When you cut 10% of your workforce, you need You're not just cutting inefficiencies. You're cutting institutional knowledge, relationships, diverse perspectives. Is that really the best way to accelerate AI development?

SPEAKER_00

That's a great point. AI development isn't just about having more AI researchers. It's about understanding how AI fits into products, how users interact with it, how to build responsible systems. You need people from all disciplines, not just machine learning experts.

SPEAKER_01

And there's the human cost too. We're talking about thousands of people losing their jobs because their company decided to pivot harder toward AI. That's a real consequence of this AI arms race that doesn't get talked about enough.

SPEAKER_00

Absolutely. But from Meta's business perspective, I get it. You know, they're seeing AI transform everything from search to productivity to creative tools, and they know their future depends on being a major player in that transformation.

SPEAKER_01

The talent acquisition from Thinking Machines Lab makes more sense in this context, too. They're simultaneously cutting costs and bringing in specific AI expertise. It's like they're reshaping their entire company around AI capabilities.

SPEAKER_00

And honestly, this might be what we see across the tech industry. Companies that aren't AI native are going to have to make these kinds of dramatic pivots to stay competitive. Meta is just one of the first to be this explicit about it.

SPEAKER_01

But let me ask you this: do you think Meta is making the right bet here? They're doubling down on AI while their core business model around social media and advertising is still incredibly profitable. Is this pivot necessary or is it panic?

SPEAKER_00

I think it's both necessary and a bit of panic, honestly. Look at how quickly ChatGPT changed user behavior around information seeking. Imagine if AI assistants become the primary interface for everything we do online. Meta could get completely bypassed.

SPEAKER_01

That's a terrifying prospect if you're Meta. Instead of people scrolling through Facebook or Instagram, they're just asking their AI assistant for information, entertainment, social updates. Your entire platform becomes irrelevant.

SPEAKER_00

Exactly. So from that perspective, laying off 10% of your workforce to fund an AI transformation isn't just smart, it's survival. The question is whether they can execute on that transformation effectively.

SPEAKER_01

And here's another angle. Meta has been pretty successful with their open source approach to AI through Lama. Do you think this workforce restructuring is about doubling down on that strategy, or are they pivoting toward more proprietary development?

SPEAKER_00

That's a really interesting question. The open source approach has given them a lot of mind share and developer adoption, but it doesn't directly translate to revenue the way a proprietary model might. They might be trying to have it both ways.

SPEAKER_01

Right, like using open source to build their ecosystem and attract talent, but then building proprietary applications and services on top of that. It's similar to what Google does with Android. Open source the platform, monetize the services.

SPEAKER_00

And if that's the strategy, then these layoffs might be about streamlining everything that's not directly contributing to that AI ecosystem play, which again makes business sense but is rough for the people who get cut.

SPEAKER_01

The timing of this relative to the other stories we're covering is striking too. You've got Google making massive external investments, open source models getting more competitive, and Meta doing internal restructuring. Everyone's making big moves at the same time. If AI can compress that timeline while maintaining safety standards, we're talking about a fundamental transformation of how medicine gets developed.

SPEAKER_00

And think about the implications beyond just speed and cost. AI can explore chemical spaces that human chemists might never think to investigate. It can identify patterns in molecular structures and biological interactions that are too complex for traditional approaches.

SPEAKER_01

But I have to ask, when we say AI-designed drugs, what does that actually mean? Is the AI doing all the work, or is it more like a very sophisticated tool that human scientists are using to accelerate their research?

SPEAKER_00

Great question. From what we know about isomorphic labs approach, it's probably more like AI is handling the massive computational work of predicting how molecules will fold, how they'll interact with target proteins, and which structures are most likely to be effective. But humans are still making the strategic decisions about which diseases to target and how to design the trials.

SPEAKER_01

The safety aspect is crucial here too. Drug development has all these rigorous testing phases for a reason. We need to be absolutely sure these compounds are safe before they go into people. The fact that AI designed drugs are making it to human trials suggests the regulatory process is adapting to this new technology.

SPEAKER_00

Exactly. And that that regulatory acceptance is huge. It means we're moving past the experimental phase into real-world application. And if these trials are successful, it could open the floodgates for AI design therapeutics.

SPEAKER_01

It's not just about beating humans at games anymore, it's about solving real problems that affect millions of people.

SPEAKER_00

That's the kind of AI application that could genuinely change the world.

SPEAKER_01

But let's dig into this broad and exciting pipeline that Jarderberg mentioned. What kinds of diseases or conditions do you think AI-designed drugs are best positioned to tackle first?

SPEAKER_00

I'd guess they're starting with diseases where we have really good data on the molecular mechanisms, but traditional drug discovery has struggled. Um, things like rare genetic diseases, certain cancers, um, maybe neurological conditions where the biological pathways are well understood. But you know, finding the right therapeutic compounds has been challenging.

SPEAKER_01

That makes sense. AI excels at pattern recognition and optimization problems, which is basically what drug discovery is: finding molecules that fit specific biological targets while avoiding harmful side effects.

SPEAKER_00

And here's what's really exciting. If this approach works, it could democratize drug discovery. Instead of only the biggest pharmaceutical companies being able to afford the massive RD investments, smaller biotech companies could use AI to identify promising compounds more efficiently.

SPEAKER_01

Although there's still the question of clinical trials, regulatory approval, manufacturing, distribution, all the expensive parts of getting drugs to market. AI might solve the discovery problem, but there are still significant barriers to actually getting these medicines to market. Patience.

SPEAKER_00

True. But if you can reduce the discovery timeline from 10 to 15 years to maybe three to five years, that's still transformative. You free up resources that can be invested in better trials, more diverse research, addressing rare diseases that weren't economically viable before.

SPEAKER_01

And there's the global health angle too. AI drug discovery could be particularly impactful for diseases that primarily affect developing countries, where traditional pharmaceutical companies haven't invested heavily because the profit margins aren't there.

SPEAKER_00

That's a really important point. If the cost of drug discovery drops dramatically, suddenly it becomes feasible to develop treatments for conditions that affect millions of people but haven't been commercially attractive to research.

SPEAKER_01

So what should people be watching for as these trials progress? How do we know if this AI approach is actually working better than traditional methods?

SPEAKER_00

First, we'll want to see the safety profiles of these drugs. Are they causing unexpected side effects that the AI modeling didn't predict? Then efficacy. Are they actually working as well as the computer models suggested they would? And finally, timeline. Did this whole process really happen faster than traditional drug discovery?

SPEAKER_01

And if the answer to all those questions is yes, then we're looking at one of the most impactful applications of AI we've seen yet. This isn't just about making technology more convenient, it's about saving lives. Alright, let's hit some rapid fire stories. First up, early reports suggest Comf UI just hit a $500 million valuation after raising $30 million. They're focused on giving creators more granular control over AI image, the E. Video, audio generation.

SPEAKER_00

Half a billion dollars for a UI company. That tells you everything about how valuable control and customization are in the AI space. People don't just want AI tools. They want AI tools they can fine-tune and modify for their specific needs.

SPEAKER_01

It's like the difference between using Instagram filters and having access to Photoshop. Comfy UI is betting that serious creators want the Photoshop version of AI generation tools.

SPEAKER_00

And that valuation suggests they're right. This could be the future of creative AI, not one size fits all tools, but powerful platforms that let users build exactly what they need.

SPEAKER_01

What's interesting is the timing. This is happening just as AI generated content is becoming mainstream. Creators are realizing they need more sophisticated tools to stand out and maintain their creative vision.

SPEAKER_00

Right. And Comfy UI's approach of giving creators granular control over every aspect of the generation process, that's exactly what professional users have been asking for. Simple consumer tools are great for getting started, but pros need precision.

SPEAKER_01

The 500 to poor also suggests investors believe this isn't just a niche market. They're betting that sophisticated AI creation tools will become as essential as traditional creative software like Adobe's suite.

SPEAKER_00

And if that happens, Comfy UI could be positioning itself as the foundational platform that other creative tools get built on top of.

SPEAKER_01

Next, we're seeing some interesting talent movement between Meta and Thinking Machines Lab. Reports suggest there's actually bidirectional talent flow between the two organizations, even as Meta is doing those big layoffs we talked about earlier.

SPEAKER_00

This is fascinating because it shows the AI talent market is incredibly fluid right now. People are moving where they think they can have the biggest impact or work on the most interesting problems, not necessarily just following the biggest paychecks.

SPEAKER_01

It also suggests that smaller, more focused AI research organizations like thinking machines might be able to compete for talent with the big tech companies, at least for certain types of researchers.

SPEAKER_00

Yeah, sometimes the best AI researchers want to work somewhere they can publish openly, collaborate freely, and not worry about corporate politics. That's a real competitive advantage for research-focused organizations.

SPEAKER_01

The bi-directional flow is particularly interesting. It suggests this isn't just meta-poaching talent. There's genuine collaboration or at least mutual respect between these organizations.

SPEAKER_00

Right. And in a field that's moving as fast as AI, these kinds of cross-pollinations of ideas and approaches can be incredibly valuable. Maybe we're seeing the emergence of a more collaborative I research ecosystem.

SPEAKER_01

Although it does raise questions about how Meta's big layoffs are affecting morale and whether top talent is looking for alternatives. When 10% of your workforce gets cut, the remaining employees start thinking about their options.

SPEAKER_00

True. But it could also be that Meta is being strategic about which talent they let go and which they try to retain. The bidirectional flow might actually be part of a broader talent optimization strategy across the AI ecosystem.

SPEAKER_01

Okay, so if you zoom out and look at everything we covered today, there's a really interesting pattern emerging. We've got massive corporate investments like Google's potential $40 billion bet on anthropic, but we also have open source models like DeepSeek that are closing the performance gap.

SPEAKER_00

Right. And that tension is going to define the next phase of AI development. Are we heading toward a world where a few giant companies control the best AI models, or are we heading toward a more distributed ecosystem where open source models are just as capable?

SPEAKER_01

And then you have these practical applications. AI drugs going to human trials, sophisticated creative tools getting massive valuations, companies completely restructuring around AI capabilities. It feels like we're transitioning from the cool demos phase to the real economic impact phase of AI.

SPEAKER_00

Exactly.

SPEAKER_01

I think the next six months are going to be crucial. If DeepSeek and other open source models can maintain their pace of improvement, and if Google's anthropic investment doesn't immediately translate to market dominance, we could see a much more competitive and diverse AI ecosystem.

SPEAKER_00

And that would be better for everyone. More innovation, more choice, more access to powerful AI tools. But it's going to require these open source projects to keep pushing the boundaries and these big investments to not create insurmountable moats.

SPEAKER_01

The human element is crucial here too. Meta's layoffs, the talent movements we're seeing, the fact that AI is now designing actual medicines that go into people, we're not just talking about technology anymore. We're talking about how AI is reshaping work, healthcare, creativity.

SPEAKER_00

And that's where the stakes get really high. The decisions being made in boardrooms about AI investments and strategy, so those decisions affect whether people have jobs, whether they have access to life-saving medicines, whether they can afford to use the most powerful creative tools.

SPEAKER_01

Which brings us back to that fundamental tension between concentration and democratization. If AI capabilities get concentrated in the hands of a few companies, that gives those companies enormous power over society. But if open source can keep pace, maybe that power stays distributed.

SPEAKER_00

And what's wild is that we're seeing both trends simultaneously. Google drops $40 billion on Anthropic on the same week that DeepSeek releases a competitive open source model. Comfy UI gets a $500 or B000 valuation while Meta cuts thousands of jobs to focus on AI.

SPEAKER_01

It's like we're in this moment where multiple futures are still possible, and the actions taken in the next few months are going to determine which one we get. That's both exciting and terrifying.

SPEAKER_00

Right. And I think that's why it's so important for people to stay informed about these developments. You know, this isn't just tech industry news, these are decisions that are going to affect how AI integrates into everyone's life.

SPEAKER_01

The other pattern I'm seeing is the increasing pace of real-world deployment. AI drugs entering human trials, creative tools getting massive valuations, companies restructuring their entire workforce around AI, the experimental phase is ending.

SPEAKER_00

Which means the window for shaping how this technology develops is closing. The fundamental architectural decisions, the business models, the regulatory frameworks, they're all getting locked in now.

SPEAKER_01

And that makes stories like the Deep Seek release so important. Every viable open source alternative is a check on the power of proprietary models. Every successful AI application outside of the big tech ecosystem is proof that innovation doesn't have to be concentrated.

SPEAKER_00

Absolutely. And for our listeners, the takeaway is that this is a critical moment to be paying attention. The AI landscape that emerges from the next six months is probably going to be the AI landscape we live with for years to come.

SPEAKER_01

That's a wrap on today's Build by AI. Thanks for diving deep into these stories with us. From massive AI investments to open source breakthroughs to AI drugs entering human trials.

SPEAKER_00

Yeah, it's going to be fascinating to watch how all of this plays out. If you found today's discussion valuable, make sure to subscribe so you don't miss our daily take on what's happening in AI.

SPEAKER_01

We'll be back tomorrow with more stories, more analysis, and more attempts to figure out where this AI revolution is actually heading. See you then.