Build by AI

ChatGPT's Superapp Gambit I 7th June

Share episode

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

0:00 | 32:20
OpenAI is turning ChatGPT into a superapp while launching 'Lockdown Mode' to protect against prompt injection attacks. Meanwhile, massive AI funding deals are reshaping the industry, and new reports warn that AI could consume as much water as a billion people by 2030. Alex and Sam dive deep into OpenAI's IPO strategy, debate whether these security measures actually work, and explore what happens when AI infrastructure starts competing with basic human needs for resources.
SPEAKER_00

I think OpenAI's new lockdown mode is actually going to make prompt injection attacks worse, not better.

SPEAKER_01

Wait, what? How does a security feature designed to prevent attacks make them worse?

SPEAKER_00

It's like putting a giant sign on your house that says valuable stuff behind this specific type of lock.

SPEAKER_01

Okay, you have 30 seconds to justify that take, because that sounds completely backwards to me.

SPEAKER_00

Think about it. They're essentially admitting that prompt injections are such a serious threat that they need a whole new mode. That's going to attract every researcher and bad actor to focus specifically on breaking through lockdown mode. It's like announcing the exact vulnerability you're most worried about.

SPEAKER_01

Alright, I'm intrigued. And there's a lot more to unpack here because this is happening right as OpenAI is planning this massive ChatGPT overhaul ahead of their IPO. You're listening to Build by AI. I'm Alex Shannon.

SPEAKER_00

And I'm Sam Hinton. Today we've got OpenAI making some massive moves, security updates, super app plans, and IPO preparations all happening at once.

SPEAKER_01

Plus, we're looking at some eye-popping funding numbers. AI infrastructure consuming ungodly amounts of water, and Apple's WWDC promises.

SPEAKER_00

And we need to talk about this SpaceX Google deal, because nearly a billion dollars per month for AI chips is just. that's a number that breaks my brain.

SPEAKER_01

Alright, let's dive in. Starting with that open AI lockdown mode and why Sam thinks it might backfire. So OpenAI has unveiled this new lockdown mode for ChatGPT, specifically designed to protect against prompt injection attacks. For folks who might not be familiar, prompt injection is basically when someone tries to trick an AI into ignoring its instructions and doing something it wasn't supposed to do. The key thing here is that even OpenAI admits this isn't a complete solution. They're saying it significantly reduces the risk, but ChatGPT could still be vulnerable. So Sam, walk me through your theory about why this might actually make things worse.

SPEAKER_00

Right, so here's the thing.

SPEAKER_01

Okay, but isn't there a flip side to that? Like if you're a business considering using ChatGPT for sensitive data, don't you want to know what protections are in place? Isn't transparency good here? See?

SPEAKER_00

That's the tension, right? And honestly, I think this move says more about OpenAI's IPO timeline than their security philosophy. They need to show enterprise customers and potential investors that they're taking security seriously, especially if they're positioning ChatGPT as this super app platform.

SPEAKER_01

But here's what concerns me about the transparency argument. They're being transparent about having a solution, but they're also being transparent about that solution not being foolproof. That seems like the worst of both worlds.

SPEAKER_00

Exactly. You know, it's like saying we've got this new security system, but it doesn't really work that well. What kind of message does that send to enterprise customers who are thinking about trusting you with their most sensitive data?

SPEAKER_01

And let's think about the technical side for a minute. Prompt injection attacks are fundamentally different from traditional cybersecurity threats. You're not trying to break into a system, you're trying to trick the system into doing what you want. How do you even defend against that?

SPEAKER_00

That's the fundamental challenge. Traditional security is about building walls and checking credentials. But prompt injection is like having a conversation with someone and trying not to get manipulated. You can't just firewall your way out of that problem.

SPEAKER_01

Right. And that connects to the other big open AI story today. The Financial Times is reporting that they're planning a major ChatGPT overhaul into a super app format ahead of their public listing. What does that actually mean in practice?

SPEAKER_00

Think WeChat, but for AI. Instead of ChatGPT being this single-purpose conversational tool, they want it to be your everything app, productivity, communication, maybe payments, integrations with other services. It's the platform play that every tech company eventually makes.

SPEAKER_01

But the more I think about this super app vision, alongside the lockdown mode announcement, the more nervous I get. If ChatGPT becomes my everything app, and their security solution admittedly isn't perfect, what happens when it gets compromised?

SPEAKER_00

But here's where the security thing becomes critical. If ChatGPT is handling everything from your work documents to your personal messages to potentially your financial transactions, prompt injection attacks become catastrophically dangerous. You're not just tricking an AI into writing a silly poem, you're potentially accessing someone's entire digital life.

SPEAKER_01

That's terrifying when you put it that way. And it makes me wonder if lockdown mode is actually robust enough for that kind of responsibility. I mean they're basically admitting it's not foolproof.

SPEAKER_00

And think about the attack surface you're creating. Right now, if someone wants to hack your life, they need to break into multiple systems your email, your bank, your work accounts. But if everything runs through ChatGPT, you've created a single point of failure for your entire digital existence.

SPEAKER_01

Which brings us back to the IPO pressure. They need to show massive growth potential to justify a huge valuation. So they're rushing toward this platform model, but are they doing it too fast?

SPEAKER_00

Exactly. And this is happening because they're racing to IPO while the AI market is still hot. They need to show massive growth potential, which means expanding beyond just chat into this platform model. But they're building the airplane while flying it, security-wise.

SPEAKER_01

And here's another angle. By announcing lockdown mode now, are they trying to get ahead of security concerns that investors might raise during the IPO process? Like, don't worry, we've got this handled.

SPEAKER_00

This. That's a really good point. IPO due diligence is going to involve some serious security audits. Maybe they're trying to control the narrative by announcing their solution before auditors find the problems.

SPEAKER_01

So what should people actually do with this information? If you're a business, do you wait and see how robust lockdown mode actually is? Or do you start testing it now?

SPEAKER_00

I'd say test it, but don't trust it with anything mission critical yet. Treat it like a beta feature, because that's essentially what it is. And definitely don't put any sensitive data through ChatGPT until we see some real-world stress testing of these protections.

SPEAKER_01

And if you're thinking about building business processes around ChatGPT as a platform, maybe pump the brakes until we see how this security situation plays out.

SPEAKER_00

Absolutely. The Super App vision is compelling, but the security infrastructure isn't ready for it yet. Don't make ChatGPT your single point of failure until they can prove lockdown mode actually works.

SPEAKER_01

Keep an eye on the security research community over the next few months. If Sam's theory is right, we're going to see a lot of people trying to break lockdown mode specifically. Alright, let's talk about this SpaceX Google deal, because the numbers are just staggering. Early reports suggest SpaceX has signed a $920 million per month deal with Google to lease AI computing capacity. We're talking about $110,000 Nvidia chips for Google's Gemini Enterprise platform. Sam, $920 million per month. That's over $11 billion a year. For compute, help me understand how we got to numbers this big.

SPEAKER_00

Dude, this is the AI infrastructure arms race hitting its peak. Google is essentially paying SpaceX to be their compute provider, which tells us two huge things. First, even Google doesn't have enough internal compute capacity for what they want to do with Gemini Enterprise. Second, SpaceX has somehow become a major player in the AI compute game, which nobody was talking about even a year ago. This is Elon basically monetizing SpaceX's infrastructure in a completely different way while they prep for their own IPO.

SPEAKER_01

Wait, how does SpaceX even have 110,000 Nvidia chips? I thought they were a rocket company. Are they just buying up chips and leasing them out?

SPEAKER_00

That's the fascinating part. You know, this suggests SpaceX has been quietly building massive compute infrastructure, probably for their own AI needs initially. Think about it. Autonomous rocket landing, Starlink optimization, Mars mission planning. That's all compute-intensive AI work. But now they've realized they can monetize that infrastructure by leasing it to other companies. It's brilliant because they get steady revenue while their rockets and satellites are the long-term bet.

SPEAKER_01

Okay, but let's break down the economics here. If SpaceX is charging Google $920 million a month, what are their margins? Even if they're paying NVIDIA full retail price for those chips, the hardware cost can't be more than a fraction of that monthly fee.

SPEAKER_00

Right. So they're probably making massive margins on this deal. But they also have to factor in power costs, cooling, data center infrastructure, maintenance. Still, at $920 million a month, they're definitely printing money.

SPEAKER_01

But I'm trying to wrap my head around the economics here. If this deal is accurate, Google is essentially spending more on compute from SpaceX than most entire companies are worth. What does that tell us about where Google thinks the AI market is heading?

SPEAKER_00

It tells us they think Gemini Enterprise is going to be absolutely massive. You don't spend $11 billion a year on compute unless you're expecting to make significantly more than that. Google is betting that enterprise AI is about to explode.

SPEAKER_01

But what if they're wrong? What if enterprise adoption is slower than expected? That's a huge fixed cost commitment they've just made.

SPEAKER_00

That's the risk. But think about Google's position. They're getting crushed by open AI in the enterprise market. ChatGPT has become synonymous with business AI. Google needs to make a massive bet to catch up, and this is that bet.

SPEAKER_01

And from SpaceX's perspective, this is genius timing. They get to de-risk their IPO by showing massive recurring revenue from a completely different business line. Investors love diversified revenue streams.

SPEAKER_00

But here's what worries me. This level of spending on compute suggests we're in a massive AI bubble. When companies are throwing around these numbers, it feels like the dot-com era all over again.

SPEAKER_01

You know what's crazy? If these numbers are real, SpaceX is making more money per month from Google than most unicorn startups are worth in total. That's just wild.

SPEAKER_00

And it shows how the AI economy is completely reshaping traditional business models. SpaceX went from being a rocket company to being a cloud provider overnight. That kind of pivot used to take years.

SPEAKER_01

That's a good point. And the timing is interesting too. Both SpaceX and Google are making these massive bets, right, as we're seeing all these IPOs being planned. Open AI, SpaceX. Everyone's trying to go public while the AI hype is still strong.

SPEAKER_00

Right. And if this deal is real, it basically guarantees SpaceX has massive revenue for their IPO prospectus. It's almost too convenient. Though I have to say, if you're an investor, seeing Google as a customer writing checks for almost a billion a month is pretty compelling.

SPEAKER_01

But here's a question. Is this sustainable? Can Google really afford to spend eleven billion dollars a year on compute from just one provider? That seems like it would eat into their margins pretty significantly.

SPEAKER_00

Only if they can turn that compute into revenue. If Gemini Enterprise becomes the dominant platform for business AI, then yeah, it's worth it. But if it doesn't work out, this could be the most expensive mistake in tech history.

SPEAKER_01

What's the broader implication here for smaller AI companies? If this is what it costs to compete at the top tier, how does anyone else even play in this market?

SPEAKER_00

They don't. This is the consolidation phase. You either get acquired by one of these giants, or you find a very specific niche where you don't need this level of compute. The barrier to entry for general AI is becoming astronomical.

SPEAKER_01

And that has huge implications for innovation, right? If only five companies can afford to build cutting-edge AI, do we get better innovation or just more expensive innovation?

SPEAKER_00

But if you need ten billion dollars just to compete, you don't get much competition. It's like the space race, but for AI.

SPEAKER_01

Keep watching how this affects GPU prices and availability for everyone else. When deals like this are happening, it's going to squeeze supply for smaller players even more. Speaking of massive funding rounds, early reports suggest Anthropic has just secured $35 billion from Apollo and Blackstone. If this is confirmed, this is one of the largest AI funding rounds we've ever seen. Sam, $35 billion. That's not just big money, that's we're betting the entire firm on AI money. What do you make of these traditional investment giants going this hard into AI?

SPEAKER_00

This is huge because Apollo and Blackstone aren't tech investors. They're massive institutional asset managers. Apollo manages over $600 billion in assets. Blackstone is over a trillion. When they write a check this big for AI, it means AI has officially moved from tech trend to fundamental infrastructure investment. These guys invest in power plants, toll roads, airports, stuff that society can't function without. They're essentially saying AI development is now in that category.

SPEAKER_01

That's actually kind of scary when you think about it. These firms don't make bets, they make infrastructure investments. They're saying AI isn't optional anymore. It's essential infrastructure.

SPEAKER_00

Exactly. And the timeline implications are massive. Infrastructure investors think in decades, not quarters, this suggests they believe AI is going to be the dominant technology for the next 20, 30 years.

SPEAKER_01

But that also makes me nervous. Anthropic has been positioning itself as the safety-focused alternative to open AI. What happens to that mission when you have traditional Wall Street money with traditional Wall Street return expectations?

SPEAKER_00

Yeah, that's the tension. Apollo and Blackstone didn't become trillion-dollar firms by prioritizing safety over returns. They're going to want Anthropic to move fast and grab market share, which could put pressure on their safety first approach.

SPEAKER_01

And think about the governance implications. With $35 billion invested, these firms are going to want board seats, strategic input, maybe even operational control. That could fundamentally change how anthropic operates. But there's also the pressure factor. When someone invests $35 billion, they expect returns. And in AI, returns come from deployment, not research. So even if you have the money for safety research, do you have the time?

SPEAKER_00

That's the fundamental tension in AI development right now. Safety takes time, but the market rewards speed. With this much money involved, the pressure to ship is going to be enormous.

SPEAKER_01

But doesn't this just escalate the AI arms race even more? Now you have Anthropic with 35 billion, OpenAI planning their IPO, Google spending almost a billion a month on compute. Everyone's just pouring more fuel on the fire.

SPEAKER_00

Absolutely. And I think that's the point. We've moved past the research phase into the whoever spends the most wins phase. It's not about having the smartest algorithm anymore. It's about having the most compute, the most data, and the most money to scale.

SPEAKER_01

And that's fundamentally changing what kind of company can succeed in AI. It's not about being scrappy or innovative anymore. It's about having access to massive capital and infrastructure.

SPEAKER_00

Right. And um and that's historically how industries mature. Think about cars or airlines. It started with lots of small players, then consolidated down to a few uh giants who could afford the infrastructure.

SPEAKER_01

Which brings us back to that consolidation point you made earlier. If it takes $35 billion just to stay competitive in AI, how many companies can actually afford to play this game?

SPEAKER_00

Maybe five or six globally. You've got the big tech companies and maybe a couple of these massively funded startups like Anthropic. Everyone else is fighting for scraps or looking for acquisition targets.

SPEAKER_01

And that has implications for innovation, for competition, for consumer choice. When only a handful of companies can afford to develop AI, do we get the best AI or just the most expensive AI?

SPEAKER_00

Aaron Powell That's the trillion dollar question, literally. And it's why these funding rounds matter so much. They're not just about money, they're about who gets to shape the future of AI.

SPEAKER_01

What's interesting is the timing too. This is happening right as we're starting to see real concerns about AI's resource consumption. Speaking of which, we should talk about that UN report on water usage.

SPEAKER_00

Right, and that's going to be the the real constraint eventually. You can raise all the money in the world, but if you can't get the physical infrastructure to support these massive AI systems, the money doesn't matter.

SPEAKER_01

If you're an AI startup right now, this anthropic deal basically sets the new baseline for what serious funding looks like. Anything under a billion starts to look like pocket change. And that brings us to this sobering UN report that estimates AI systems will consume as much water as a billion people by 2030. We're talking about the physical infrastructure costs of this AI boom, and apparently it's massive. Sam, when we talk about AI infrastructure, most people think about chips and electricity. But water consumption. Help me understand why AI needs so much water.

SPEAKER_00

You know, it's all about cooling. These data centers running AI workloads generate insane amounts of heat. And water is still the most efficient way to cool them at scale. Every time you ask ChatGPT a question, you're basically contributing to this massive cooling operation. The UN is saying that by 2030, the water consumption for AI will equal what a billion people use for drinking, cooking, cleaning, everything. That's not just a big number, that's a humanitarian crisis waiting to happen.

SPEAKER_01

Wait, let me put this in perspective. We just talked about SpaceX and Google doing a $920 million per month deal for compute. Are we saying that none of these massive deals factor in the environmental costs?

SPEAKER_00

That's exactly right. Everyone's focused on compute costs, but nobody's pricing in the water. And here's the kicker. Water isn't infinite. We're already seeing water shortages in major tech hubs like California and Phoenix, where a lot of these data centers are located.

SPEAKER_01

This is mind-blowing to me. So you could have all the money in the world to spend on AI development, but you literally can't get the resources to run the systems.

SPEAKER_00

Bingo. And the timeline is what's scary, 2030 is only four years away, and that's not enough time to build alternative cooling infrastructure or find new water sources. We're basically racing toward a wall.

SPEAKER_01

And think about the political implications. When AI companies are competing with people for water, that's going to create some serious public backlash. You can't just say, sorry, ChatGPT needs your drinking water.

SPEAKER_00

Right, and that's going to force some really uncomfortable choices. Do we limit AI development because of water consumption? Do we start rationing water between AI companies and people? These sound like sci-fi questions. But they're going to be real policy decisions in the next few years.

SPEAKER_01

What's the regulatory landscape going to look like? Are we going to see water usage caps for tech companies? Environmental impact requirements for AI development.

SPEAKER_00

And honestly, this could be what finally slows down the AI arms race. Not technical limitations or funding constraints, but literal resource scarcity. The irony is that AI is supposed to help us solve resource management problems, but we might run out of resources before we get there.

SPEAKER_01

This also adds a whole new dimension to the AI arms race. It's not just about money and talents anymore, it's about securing physical resources. Countries with abundant water supplies suddenly have a strategic advantage in AI development.

SPEAKER_00

Right. And it explains why we're seeing more AI infrastructure being built in places like Iceland and Norway. Cold climates, lots of water, cheaper cooling. Geography is becoming destiny in AI.

SPEAKER_01

But that creates its own problems, right? If AI development moves to where the water is, what happens to the places where the talent and money are? Do we end up with this weird geographic split?

SPEAKER_00

Possibly. And that has huge implications for which countries end up leading in AI. It's not just about having smart engineers or venture capital anymore. It's about having basic natural resources.

SPEAKER_01

What does this mean for companies planning their AI strategies? Do you need to start thinking about where your compute is located based on water availability?

SPEAKER_00

Absolutely. And honestly, this is going to be another consolidation factor. Small companies can't negotiate water rights or build data centers in optimal locations.

SPEAKER_01

You're going to have to rely on the big cloud providers who can So we could end up in a situation where the big tech companies control AI development, not because they have the best technology, but because they control the water and cooling infrastructure.

SPEAKER_00

That's entirely possible. And that's a very different kind of monopoly than we've seen before. It's not about software or algorithms. It's about physical infrastructure and natural resources.

SPEAKER_01

The other thing that strikes me is the time frame. 2030 is only four years away. This isn't a distant problem. This is something we need to be planning for right now.

SPEAKER_00

And given how long infrastructure projects take, we're probably already behind. You can't just build a new water efficient data center overnight. This should be a wake-up call for the entire industry.

SPEAKER_01

Keep an eye on water rights and data center regulations. This is going to become a major constraint on AI development faster than anyone expects. First up, early reports suggest Shreum Krishnan is leaving his role as White House AI advisor and reportedly starting a new institution to influence Trump's AI policy.

SPEAKER_00

It suggests he thinks he can have more influence from the outside.

SPEAKER_01

The timing feels significant too. Right as all these massive AI deals are happening and we're seeing this infrastructure crunch, AI policy is about to get a lot more complicated.

SPEAKER_00

Yeah, and having someone who gets both the technical and policy sides working independently could actually be good for the industry. Government moves too slow for AI development cycles.

SPEAKER_01

But I wonder about the political dynamics here. Starting an institution to influence Trump's AI policy suggests there are some serious disagreements about direction happening behind the scenes.

SPEAKER_00

Right. And and when you're dealing with something as strategically important as AI, those disagreements matter. The policies we set now are going to shape the industry for decades.

SPEAKER_01

Keep watching to see what kind of backing this new institution gets. If major tech companies or investors get involved, that tells us a lot about where the industry thinks policy should go.

SPEAKER_00

And honestly, having more technical expertise in the policy conversation is desperately needed. Too many AI regulations are written by people who don't understand how the technology actually works.

SPEAKER_01

Apple's WWDC 2026 is coming up. An early report suggests we're going to see a major Siri, revamp plus Apple intelligence updates. Sam, is Apple finally going to catch up in the AI race?

SPEAKER_00

They have to, right? Siri has been embarrassingly behind ChatGPT and even Google Assistant. But Apple's advantage is integration. If they can make AI feel seamless across all their devices, that could be huge. But they can offer AI that keeps your data on device. That's their differentiator.

SPEAKER_01

But here's the thing. After talking about all these massive infrastructure deals, can on-device AI really compete with cloud-based systems that have unlimited compute?

SPEAKER_00

That's the key question. On-device AI is always going to be more limited, but it's also more private and doesn't require internet connectivity. There's definitely a market for that.

SPEAKER_01

And given all the security concerns we talked about with OpenAI's lockdown mode, maybe privacy-focused AI is exactly what businesses and consumers want.

SPEAKER_00

Right. Apple could position themselves as the safe AI option. While everyone else is racing to build these massive, potentially vulnerable systems, Apple offers AI you can actually trust.

SPEAKER_01

Politico is reporting on lawsuits against AI companies that could create a big tobacco moment for the industry. That's a pretty loaded comparison.

SPEAKER_00

Yeah, and it's not just about copyright anymore. We're seeing lawsuits around bias, misinformation, privacy, all the stuff that companies have been saying we'll figure it out later about. Well, later is now.

SPEAKER_01

With all this money flowing into AI, the anthropic deal, the open AI IPO plans, these companies are now big enough targets for serious legal action.

SPEAKER_00

Exactly. When you're raising billions of dollars, you can't claim you're just a scrappy startup anymore. You have real responsibilities and real liability.

SPEAKER_01

The big tobacco comparison is interesting because it suggests systematic harm that the companies knew about but didn't address. Are we saying AI companies are knowingly causing harm?

SPEAKER_00

I think the argument is that they know about potential risks, bias, misinformation, job displacement, but they're prioritizing growth over addressing those risks. That's the tobacco parallel.

SPEAKER_01

And if these lawsuits succeed, they could completely change the economics of AI development. You'd have to factor in massive liability costs from day one.

SPEAKER_00

Which might actually slow down the arms race we've been talking about. If you're facing potential billion-dollar lawsuits, you might think twice about moving fast and breaking things.

SPEAKER_01

You need the resources of a nation state to play in the top tier.

SPEAKER_00

The companies that survive are going to be the ones that can navigate all these constraints, technical, financial, regulatory, and resource-based. That's a very small group.

SPEAKER_01

And what's fascinating is how these constraints are interconnected. The security problems with prompt injection connect to the super app vision, which connects to IPO pressure, which connects to massive funding rounds, which connect to resource consumption.

SPEAKER_00

Exactly. But OpenAI needs to show growth for their IPO, so they're rushing toward a super app model, but their security isn't ready for it. Meanwhile, Google is spending almost a billion a month to compete, which is driving up resource consumption to unsustainable levels.

SPEAKER_01

The question for 2026 and beyond is whether this consolidation is actually good for innovation. When only five or six companies can afford to develop cutting-edge AI, do we get better AI or just more expensive AI?

SPEAKER_00

And more importantly, do we get AI that actually serves society's needs, or just AI that generates the highest returns for investors? Because those aren't necessarily the same thing.

SPEAKER_01

The water consumption issue really drives this home for me. We're talking about AI competing with people for basic resources. That's not a technology problem anymore. That's a societal choice.

SPEAKER_00

Right. And those choices are being made right now in boardrooms and investor meetings without much public input. The companies raising billions today are determining what AI looks like for the next decade.

SPEAKER_01

Which is why the regulatory and legal pressure we're seeing might actually be necessary. If the only constraint on AI development is financial, and money is essentially unlimited right now, then we need other kinds of constraints. So what does this mean for the average person or business watching all this unfold? How do you make decisions when the entire landscape is shifting this fast?

SPEAKER_00

I think the key is to stay informed, but don't bet everything on any one platform or approach. The AI industry in 2030 is going to look very different from today, and a lot of today's leaders might not survive.

SPEAKER_01

And maybe that's okay. Maybe we need this shakeout to separate the companies that are building AI responsibly from the ones that are just trying to get rich quick.

SPEAKER_00

That's the trillion dollar question, literally. And it's one we're going to be answering in real time over the next few years.

SPEAKER_01

Alright, that's a wrap on today's build by AI. Lots to keep an eye on as these massive AI deals play out, and we start hitting real world constraints on development.

SPEAKER_00

Yeah, subscribe wherever you get your podcast because this story is moving fast. Tomorrow we'll be back with more AI news and analysis.

SPEAKER_01

See you tomorrow.

SPEAKER_00

Later