AI Isn't A Bubble. That's How NVIDIA's $500 Billion Push Ends Up In Your Retirement.

If AI is a bubble, Nvidia and Wall If AI is a bubble, Nvidia and Wall Street are about to pour another $500 Street are about to pour another $500 Street are about to pour another $500 billion into it, and then your billion into it, and then your billion into it, and then your retirement goes with it. If AI is not a retirement goes with it. If AI is not a retirement goes with it. If AI is not a bubble, and then it works, intelligence bubble, and then it works, intelligence bubble, and then it works, intelligence gets cheap, companies need fewer people, gets cheap, companies need fewer people, gets cheap, companies need fewer people, and your job goes instead. Your and your job goes instead. Your and your job goes instead. Your retirement or your job, right? Bad retirement or your job, right? Bad retirement or your job, right? Bad either way. But, I don't think those are either way. But, I don't think those are either way. But, I don't think those are the choices, and this video is about the choices, and this video is about the choices, and this video is about why. Every major tech needs two why. Every major tech needs two why. Every major tech needs two inventions. Someone has to invent the inventions. Someone has to invent the inventions. Someone has to invent the machine, and someone has to invent a way machine, and someone has to invent a way machine, and someone has to invent a way to pay for enough of those machines that to pay for enough of those machines that to pay for enough of those machines that they really change the economy. And this they really change the economy. And this they really change the economy. And this week is about the second one. Nvidia week is about the second one. Nvidia week is about the second one. Nvidia just signed agreements with Apollo, just signed agreements with Apollo, just signed agreements with Apollo, BlackRock, Blackstone, Brookfield, BlackRock, Blackstone, Brookfield, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. Together, they Goldman Sachs, and KKR. Together, they Goldman Sachs, and KKR. Together, they want to create financing platforms that want to create financing platforms that want to create financing platforms that can mobilize more than $500 billion of can mobilize more than $500 billion of can mobilize more than $500 billion of third-party capital for the buildout of third-party capital for the buildout of third-party capital for the buildout of AI infrastructure over time. AI infrastructure over time. AI infrastructure over time. But, Nvidia didn't raise $500 billion. But, Nvidia didn't raise $500 billion. But, Nvidia didn't raise $500 billion. There's no half a trillion dollar bank There's no half a trillion dollar bank There's no half a trillion dollar bank account waiting to buy GPUs. These are account waiting to buy GPUs. These are account waiting to buy GPUs. These are memoranda of understanding. Nvidia's own memoranda of understanding. Nvidia's own memoranda of understanding. Nvidia's own announcement says it very, very plainly.

announcement says it very, very plainly. announcement says it very, very plainly. These partnerships remain subject to These partnerships remain subject to These partnerships remain subject to execution of the final agreements. The execution of the final agreements. The execution of the final agreements. The platforms still have to be formed, platforms still have to be formed, platforms still have to be formed, investors still have to commit, and each investors still have to commit, and each investors still have to commit, and each project, of course, still has to qualify project, of course, still has to qualify project, of course, still has to qualify for financing. But, that being said, six for financing. But, that being said, six for financing. But, that being said, six of the largest pools of capital in the of the largest pools of capital in the of the largest pools of capital in the world have decided that AI compute can world have decided that AI compute can world have decided that AI compute can become something they know how to become something they know how to become something they know how to finance. finance. finance. That has not been true before. Microsoft That has not been true before. Microsoft That has not been true before. Microsoft invests in OpenAI, OpenAI buys Microsoft invests in OpenAI, OpenAI buys Microsoft invests in OpenAI, OpenAI buys Microsoft compute. Nvidia invests in CoreWeave, compute. Nvidia invests in CoreWeave, compute. Nvidia invests in CoreWeave, CoreWeave borrows money to buy Nvidia CoreWeave borrows money to buy Nvidia CoreWeave borrows money to buy Nvidia chips, and OpenAI reserves capacity from chips, and OpenAI reserves capacity from chips, and OpenAI reserves capacity from CoreWeave. Nvidia has also agreed, CoreWeave. Nvidia has also agreed, CoreWeave. Nvidia has also agreed, under certain conditions, to buy unused under certain conditions, to buy unused under certain conditions, to buy unused CoreWeave capacity. It's all back and CoreWeave capacity. It's all back and CoreWeave capacity. It's all back and forth and back and forth, and here's how forth and back and forth, and here's how forth and back and forth, and here's how big that circle is inside one number. big that circle is inside one number. big that circle is inside one number. Microsoft booked $24 billion dollars Microsoft booked $24 billion dollars Microsoft booked $24 billion dollars revenue from open AI in fiscal 2026. Set revenue from open AI in fiscal 2026. Set revenue from open AI in fiscal 2026. Set that against the 37 billion dollar AI that against the 37 billion dollar AI that against the 37 billion dollar AI run rate that Microsoft advertises and run rate that Microsoft advertises and run rate that Microsoft advertises and you can see how much of that headline you can see how much of that headline you can see how much of that headline number is one customer that Microsoft number is one customer that Microsoft number is one customer that Microsoft also invests in. So, you can drive also invests in. So, you can drive also invests in. So, you can drive reinvestment and contract and loan and reinvestment and contract and loan and reinvestment and contract and loan and chip purchase on a page and you're chip purchase on a page and you're chip purchase on a page and you're looking at the same money moving around looking at the same money moving around looking at the same money moving around in a circle making every company look in a circle making every company look in a circle making every company look richer on the way through.

richer on the way through. richer on the way through. Now, those contracts are real, but this Now, those contracts are real, but this Now, those contracts are real, but this isn't one closed loop where Nvidia sends isn't one closed loop where Nvidia sends isn't one closed loop where Nvidia sends out a dollar and gets the same dollar out a dollar and gets the same dollar out a dollar and gets the same dollar back. The concentration back. The concentration back. The concentration still raises a fair question. Are these still raises a fair question. Are these still raises a fair question. Are these companies financing genuine demand or companies financing genuine demand or companies financing genuine demand or financing one another so they can report financing one another so they can report financing one another so they can report demand? But, the financiers are not demand? But, the financiers are not demand? But, the financiers are not optional and railroads are the cleanest optional and railroads are the cleanest optional and railroads are the cleanest example of that actually. A railroad example of that actually. A railroad example of that actually. A railroad company had to buy land and build company had to buy land and build company had to buy land and build bridges and lay track and buy bridges and lay track and buy bridges and lay track and buy locomotives and hire workers years locomotives and hire workers years locomotives and hire workers years before ticket and freight revenue could before ticket and freight revenue could before ticket and freight revenue could repay any of it. The technology was repay any of it. The technology was repay any of it. The technology was super useful, but the company still super useful, but the company still super useful, but the company still couldn't finance a national network from couldn't finance a national network from couldn't finance a national network from the cash sitting in its bank account. the cash sitting in its bank account. the cash sitting in its bank account. So, America built land grants and So, America built land grants and So, America built land grants and railroad bonds and underwriting railroad bonds and underwriting railroad bonds and underwriting syndicates and investment banks and syndicates and investment banks and syndicates and investment banks and public markets that could collect money public markets that could collect money public markets that could collect money from investors in the US and Europe and from investors in the US and Europe and from investors in the US and Europe and finance railroad expansion. The country finance railroad expansion. The country finance railroad expansion. The country didn't just invent a better locomotive, didn't just invent a better locomotive, didn't just invent a better locomotive, it learned how to turn future railroad it learned how to turn future railroad it learned how to turn future railroad traffic often years away into money that traffic often years away into money that traffic often years away into money that could lay track today. Now, that system could lay track today. Now, that system could lay track today. Now, that system produced speculation. It produced fraud.

produced speculation. It produced fraud. produced speculation. It produced fraud. It produced defaults and the panic of It produced defaults and the panic of It produced defaults and the panic of 1873. It also produced the railroad 1873. It also produced the railroad 1873. It also produced the railroad network. The mechanism changes with the network. The mechanism changes with the network. The mechanism changes with the machine, but the purpose never does. You machine, but the purpose never does. You machine, but the purpose never does. You move money from people who have it to a move money from people who have it to a move money from people who have it to a productive asset that needs years to pay productive asset that needs years to pay productive asset that needs years to pay it back. You need land and power and it back. You need land and power and it back. You need land and power and substations and cooling and networking substations and cooling and networking substations and cooling and networking and buildings and racks of expensive and buildings and racks of expensive and buildings and racks of expensive accelerators before a customer receives accelerators before a customer receives accelerators before a customer receives token one. token one. token one. The capital goes out first, the revenue The capital goes out first, the revenue The capital goes out first, the revenue arrives later. You need to measure that arrives later. You need to measure that arrives later. You need to measure that demand without counting the same demand without counting the same demand without counting the same customer dollar four times. A company customer dollar four times. A company customer dollar four times. A company called Exponential View actually tried called Exponential View actually tried called Exponential View actually tried to do that. When a business pays an AI to do that. When a business pays an AI to do that. When a business pays an AI application, and that application pays a application, and that application pays a application, and that application pays a model company, and the model company model company, and the model company model company, and the model company pays a cloud provider, you can't count pays a cloud provider, you can't count pays a cloud provider, you can't count all three payments as three separate all three payments as three separate all three payments as three separate pieces of end demand. pieces of end demand. pieces of end demand. Exponential View counts the outside Exponential View counts the outside Exponential View counts the outside customer dollar exactly once, which is customer dollar exactly once, which is customer dollar exactly once, which is the correct way to find out how much the correct way to find out how much the correct way to find out how much customer demand you have in the system. customer demand you have in the system. customer demand you have in the system. Their June report puts it at $110 Their June report puts it at $110 Their June report puts it at $110 billion of generative AI revenue over billion of generative AI revenue over billion of generative AI revenue over the trailing 12 months, with the latest the trailing 12 months, with the latest the trailing 12 months, with the latest month running at an annualized pace month running at an annualized pace month running at an annualized pace above $175 billion. In other words, it's above $175 billion. In other words, it's above $175 billion. In other words, it's growing fast. Anthropic said its run growing fast. Anthropic said its run growing fast. Anthropic said its run rate crossed $47 billion in May, but the rate crossed $47 billion in May, but the rate crossed $47 billion in May, but the latest rumors from Anthropic insiders latest rumors from Anthropic insiders latest rumors from Anthropic insiders are that they are going to be hitting are that they are going to be hitting are that they are going to be hitting over $100 billion in run rate before over $100 billion in run rate before over $100 billion in run rate before their IPO later this year. And that's a their IPO later this year. And that's a their IPO later this year. And that's a rumor, and we'll find out what it rumor, and we'll find out what it rumor, and we'll find out what it actually says in the S-1, and the rumors actually says in the S-1, and the rumors actually says in the S-1, and the rumors will keep running to the IPO. But, the will keep running to the IPO. But, the will keep running to the IPO. But, the point is we continue to see rapid point is we continue to see rapid point is we continue to see rapid growth. The median firm spent about 12 growth. The median firm spent about 12 growth. The median firm spent about 12 bucks per employee, the top 1% companies

bucks per employee, the top 1% companies bucks per employee, the top 1% companies spent $7,400. spent $7,400. spent $7,400. And that looks like a market spreading And that looks like a market spreading And that looks like a market spreading at different speeds, not a market with at different speeds, not a market with at different speeds, not a market with zero outside customers. Exponential zero outside customers. Exponential zero outside customers. Exponential View's estimate is that every 10% cut in View's estimate is that every 10% cut in View's estimate is that every 10% cut in token prices leads to 12 to 18% more token prices leads to 12 to 18% more token prices leads to 12 to 18% more tokens used. But, we do need to be tokens used. But, we do need to be tokens used. But, we do need to be careful with that one. It's their careful with that one. It's their careful with that one. It's their estimate, it's not, you know, a result estimate, it's not, you know, a result estimate, it's not, you know, a result that's been measured in the market. It's that's been measured in the market. It's that's been measured in the market. It's basically an early estimate of the basically an early estimate of the basically an early estimate of the elasticity of tokens. And so, we're elasticity of tokens. And so, we're elasticity of tokens. And so, we're going to have to start to think about going to have to start to think about going to have to start to think about that more deeply as the token economy that more deeply as the token economy that more deeply as the token economy grows. In other words, generally grows. In other words, generally grows. In other words, generally speaking, as firms get lower token speaking, as firms get lower token speaking, as firms get lower token prices, they don't just pocket the prices, they don't just pocket the prices, they don't just pocket the savings. They run the model more often, savings. They run the model more often, savings. They run the model more often, they give an agent more steps, they let they give an agent more steps, they let they give an agent more steps, they let it check its work. One completed task it check its work. One completed task it check its work. One completed task that's on that advanced side may involve that's on that advanced side may involve that's on that advanced side may involve 50 or even 500 model calls. So, falling 50 or even 500 model calls. So, falling 50 or even 500 model calls. So, falling prices can increase total demand over prices can increase total demand over prices can increase total demand over time. Better models can widen the market time. Better models can widen the market time. Better models can widen the market again because they start to compete for again because they start to compete for again because they start to compete for the money you spend on work and not only the money you spend on work and not only the money you spend on work and not only the money you spend on software. the money you spend on software. the money you spend on software. CoreWeave shows the infrastructure side CoreWeave shows the infrastructure side CoreWeave shows the infrastructure side of things. It's backlog has passed of things. It's backlog has passed of things. It's backlog has passed roughly a hundred billion dollars and roughly a hundred billion dollars and roughly a hundred billion dollars and its quarterly revenue has more than its quarterly revenue has more than its quarterly revenue has more than doubled from a year earlier. Now, doubled from a year earlier. Now, doubled from a year earlier. Now, backlog is not revenue and CoreWeave backlog is not revenue and CoreWeave backlog is not revenue and CoreWeave still has to build and deliver that still has to build and deliver that still has to build and deliver that capacity. Customers don't reserve a capacity. Customers don't reserve a capacity. Customers don't reserve a hundred billion dollars of compute hundred billion dollars of compute hundred billion dollars of compute though because nobody wants it. And this though because nobody wants it. And this though because nobody wants it. And this is why Nvidia needs Wall Street. The is why Nvidia needs Wall Street. The is why Nvidia needs Wall Street. The first wave came from hyperscaler cash, first wave came from hyperscaler cash, first wave came from hyperscaler cash, from venture capital, from debt raised from venture capital, from debt raised from venture capital, from debt raised by specialized clouds, but the next wave by specialized clouds, but the next wave by specialized clouds, but the next wave needs insurers and pensions and needs insurers and pensions and needs insurers and pensions and infrastructure funds and private credit

infrastructure funds and private credit infrastructure funds and private credit and sovereign capital. The same and sovereign capital. The same and sovereign capital. The same investors that already finance power investors that already finance power investors that already finance power plants and aircraft and fiber and plants and aircraft and fiber and plants and aircraft and fiber and warehouses and toll roads. Every major warehouses and toll roads. Every major warehouses and toll roads. Every major tech needs those two inventions. And tech needs those two inventions. And tech needs those two inventions. And this is Nvidia trying to build that this is Nvidia trying to build that this is Nvidia trying to build that second financing one in public on a second financing one in public on a second financing one in public on a deadline. Nobody covering this story is deadline. Nobody covering this story is deadline. Nobody covering this story is explaining that part. So, let me do it explaining that part. So, let me do it explaining that part. So, let me do it because it's I think it's the whole because it's I think it's the whole because it's I think it's the whole deal. I want you to imagine a separate deal. I want you to imagine a separate deal. I want you to imagine a separate company. They're created to own one company. They're created to own one company. They're created to own one block of AI infrastructure. It's got the block of AI infrastructure. It's got the block of AI infrastructure. It's got the building, the power connection, the building, the power connection, the building, the power connection, the cooling, the network and the GPUs. cooling, the network and the GPUs. cooling, the network and the GPUs. And it's got a customer who agrees to And it's got a customer who agrees to And it's got a customer who agrees to reserve that capacity or pay a minimum reserve that capacity or pay a minimum reserve that capacity or pay a minimum amount for several years. amount for several years. amount for several years. An equity investor puts in money that An equity investor puts in money that An equity investor puts in money that takes the first loss on that. A lender takes the first loss on that. A lender takes the first loss on that. A lender supplies debt and gets paid before the supplies debt and gets paid before the supplies debt and gets paid before the equity owner. In this situation, the equity owner. In this situation, the equity owner. In this situation, the equipment works as collateral and a equipment works as collateral and a equipment works as collateral and a reserve account covers a temporary reserve account covers a temporary reserve account covers a temporary shortfall. shortfall. shortfall. In some structures, Nvidia provides In some structures, Nvidia provides In some structures, Nvidia provides limited credit support. limited credit support. limited credit support. Uh Jensen Huang put a number on that, up Uh Jensen Huang put a number on that, up Uh Jensen Huang put a number on that, up to 25% of an opportunity. And he was to 25% of an opportunity. And he was to 25% of an opportunity. And he was careful to say on a project-by-project careful to say on a project-by-project careful to say on a project-by-project basis. So, Nvidia is taking on some of basis. So, Nvidia is taking on some of basis. So, Nvidia is taking on some of the residual risk in the system here, the residual risk in the system here, the residual risk in the system here, but not all of it. And then when the but not all of it. And then when the but not all of it. And then when the customer payments arrive, the project customer payments arrive, the project customer payments arrive, the project pays its operating costs and then its pays its operating costs and then its pays its operating costs and then its lenders and then its owners. The risk lenders and then its owners. The risk lenders and then its owners. The risk has not disappeared. Instead, it's been has not disappeared. Instead, it's been has not disappeared. Instead, it's been divided into pieces that different divided into pieces that different divided into pieces that different investors can evaluate and buy. Start investors can evaluate and buy. Start investors can evaluate and buy. Start with capital concentration. The same with capital concentration. The same with capital concentration. The same small group of chip companies and clouds small group of chip companies and clouds small group of chip companies and clouds and model labs and hyperscalers can be and model labs and hyperscalers can be and model labs and hyperscalers can be customers and suppliers and lenders and customers and suppliers and lenders and customers and suppliers and lenders and investors in one another. And if one

investors in one another. And if one investors in one another. And if one large customer cannot honor a contract, large customer cannot honor a contract, large customer cannot honor a contract, the loss moves through several companies the loss moves through several companies the loss moves through several companies at once. CoreWeave has fast growth. It at once. CoreWeave has fast growth. It at once. CoreWeave has fast growth. It has this great backlog, but it also has has this great backlog, but it also has has this great backlog, but it also has debts and interest costs and continuing debts and interest costs and continuing debts and interest costs and continuing losses while it builds out. Both of losses while it builds out. Both of losses while it builds out. Both of those facts matter. Demand can be real those facts matter. Demand can be real those facts matter. Demand can be real and the financing can still be too and the financing can still be too and the financing can still be too aggressive. In other words, the A100 aggressive. In other words, the A100 aggressive. In other words, the A100 chip, which was launched in 2020, will chip, which was launched in 2020, will chip, which was launched in 2020, will be 9 years old by 2029 and still be 9 years old by 2029 and still be 9 years old by 2029 and still generating value. So, you've got a chip generating value. So, you've got a chip generating value. So, you've got a chip generation from 2020 with a customer generation from 2020 with a customer generation from 2020 with a customer contract still running and generating contract still running and generating contract still running and generating revenue 9 years after launch and the revenue 9 years after launch and the revenue 9 years after launch and the conventional assumption of a GPU life is conventional assumption of a GPU life is conventional assumption of a GPU life is 3 to 5 years. 3 to 5 years. 3 to 5 years. That doesn't mean the chip stops That doesn't mean the chip stops That doesn't mean the chip stops depreciating, but it does mean that the depreciating, but it does mean that the depreciating, but it does mean that the strongest version of the stranded asset strongest version of the stranded asset strongest version of the stranded asset argument where an old GPU just quickly argument where an old GPU just quickly argument where an old GPU just quickly becomes worthless, becomes worthless, becomes worthless, it doesn't necessarily fit the evidence it doesn't necessarily fit the evidence it doesn't necessarily fit the evidence we're seeing. So, an old GPU can still we're seeing. So, an old GPU can still we're seeing. So, an old GPU can still keep earning money and that doesn't mean keep earning money and that doesn't mean keep earning money and that doesn't mean it's risk-free, right? It still might it's risk-free, right? It still might it's risk-free, right? It still might sell for less than the lender assumed, sell for less than the lender assumed, sell for less than the lender assumed, but it certainly is much less risky than but it certainly is much less risky than but it certainly is much less risky than the idea behind a lot of the bubble the idea behind a lot of the bubble the idea behind a lot of the bubble assumptions, which is this whole concept assumptions, which is this whole concept assumptions, which is this whole concept that people are assuming that GPUs will that people are assuming that GPUs will that people are assuming that GPUs will last longer and be more functional than last longer and be more functional than last longer and be more functional than they really will be. It's starting to they really will be. It's starting to they really will be. It's starting to look like GPUs have a longer life and look like GPUs have a longer life and look like GPUs have a longer life and that makes everybody's financing safer.

that makes everybody's financing safer. that makes everybody's financing safer. The third risk is incentives. Nvidia The third risk is incentives. Nvidia The third risk is incentives. Nvidia gets paid when the equipment sells. So, gets paid when the equipment sells. So, gets paid when the equipment sells. So, the cloud operator benefits from the cloud operator benefits from the cloud operator benefits from announcing a large backlog. The announcing a large backlog. The announcing a large backlog. The financial firm may collect fees before financial firm may collect fees before financial firm may collect fees before the project has performed for the full the project has performed for the full the project has performed for the full 10 years. So, if everybody gets paid 10 years. So, if everybody gets paid 10 years. So, if everybody gets paid early and somebody else owns the early and somebody else owns the early and somebody else owns the long-term risk, underwriting can long-term risk, underwriting can long-term risk, underwriting can deteriorate very, very fast. deteriorate very, very fast. deteriorate very, very fast. AI infrastructure has leverage and AI infrastructure has leverage and AI infrastructure has leverage and concentrated counterparties and fee concentrated counterparties and fee concentrated counterparties and fee incentives and uncertain collateral incentives and uncertain collateral incentives and uncertain collateral value, and those are real warning signs. value, and those are real warning signs. value, and those are real warning signs. Those are things we have to work to Those are things we have to work to Those are things we have to work to de-risk in the system. But, there's much de-risk in the system. But, there's much de-risk in the system. But, there's much more of a market here than most people more of a market here than most people more of a market here than most people assume. In March, CoreWeave closed an 8 assume. In March, CoreWeave closed an 8 assume. In March, CoreWeave closed an 8 and 1/2 billion-dollar loan facility and 1/2 billion-dollar loan facility and 1/2 billion-dollar loan facility rated A3 by Moody's and A low by DBRS. rated A3 by Moody's and A low by DBRS. rated A3 by Moody's and A low by DBRS. CoreWeave called it the first CoreWeave called it the first CoreWeave called it the first investment-grade financing secured by investment-grade financing secured by investment-grade financing secured by high-performance computing high-performance computing high-performance computing infrastructure with a customer contract infrastructure with a customer contract infrastructure with a customer contract behind it. So, GPU-backed debt already behind it. So, GPU-backed debt already behind it. So, GPU-backed debt already is getting rated and institutions are is getting rated and institutions are is getting rated and institutions are already starting to buy it. And then, in already starting to buy it. And then, in already starting to buy it. And then, in July, staff at the SEC addressed the July, staff at the SEC addressed the July, staff at the SEC addressed the next step. They confirmed that data next step. They confirmed that data next step. They confirmed that data center securitizations are not center securitizations are not center securitizations are not asset-backed securities under the asset-backed securities under the asset-backed securities under the Exchange Act, which means the risk Exchange Act, which means the risk Exchange Act, which means the risk retention rule written after 2008 retention rule written after 2008 retention rule written after 2008 doesn't apply to that, which raised some doesn't apply to that, which raised some doesn't apply to that, which raised some flags for me. I lived through 2008. I flags for me. I lived through 2008. I flags for me. I lived through 2008. I remember that. But, I do want to be remember that. But, I do want to be remember that. But, I do want to be clear about what happened there. What it clear about what happened there. What it clear about what happened there. What it is and what it is not when the rule is and what it is not when the rule is and what it is not when the rule doesn't apply.

doesn't apply. doesn't apply. That is not a rule being repealed. It's That is not a rule being repealed. It's That is not a rule being repealed. It's the staff confirming the rule never the staff confirming the rule never the staff confirming the rule never anticipated this kind of deal. Rated GPU anticipated this kind of deal. Rated GPU anticipated this kind of deal. Rated GPU debt exists. The legal path to selling debt exists. The legal path to selling debt exists. The legal path to selling it far more widely is clearer than it it far more widely is clearer than it it far more widely is clearer than it was a month ago. was a month ago. was a month ago. And we expect securitization to follow And we expect securitization to follow And we expect securitization to follow from that, although the $500 billion from that, although the $500 billion from that, although the $500 billion deal I talked about at the top of the deal I talked about at the top of the deal I talked about at the top of the video doesn't necessarily imply video doesn't necessarily imply video doesn't necessarily imply securitization. The mortgage system securitization. The mortgage system securitization. The mortgage system broke when underwriting weakened and broke when underwriting weakened and broke when underwriting weakened and borrowers stopped paying. borrowers stopped paying. borrowers stopped paying. Now, that's not what we're seeing. The Now, that's not what we're seeing. The Now, that's not what we're seeing. The June data I mentioned earlier, end June data I mentioned earlier, end June data I mentioned earlier, end customer revenue is accelerating. That's customer revenue is accelerating. That's customer revenue is accelerating. That's what we see consistently from everything what we see consistently from everything what we see consistently from everything across the industry right now. So, in across the industry right now. So, in across the industry right now. So, in that world, a bad AI project is more that world, a bad AI project is more that world, a bad AI project is more likely to start as a bad loan against likely to start as a bad loan against likely to start as a bad loan against one concentrated customer than is proof one concentrated customer than is proof one concentrated customer than is proof that a larger industry is not paying for that a larger industry is not paying for that a larger industry is not paying for AI. But, a stock market correction, a AI. But, a stock market correction, a AI. But, a stock market correction, a failed cloud operator, and a 2008 style failed cloud operator, and a 2008 style failed cloud operator, and a 2008 style financial crisis are not the same financial crisis are not the same financial crisis are not the same things. And to get the last one, the things. And to get the last one, the things. And to get the last one, the risk has to become large and highly risk has to become large and highly risk has to become large and highly leveraged and widely distributed and leveraged and widely distributed and leveraged and widely distributed and badly rated and connected enough that badly rated and connected enough that badly rated and connected enough that many institutions fail together.

many institutions fail together. many institutions fail together. I can see some of those early pieces, I can see some of those early pieces, I can see some of those early pieces, but I cannot see any intent to put that but I cannot see any intent to put that but I cannot see any intent to put that full machine together and that full full machine together and that full full machine together and that full machine certainly does not exist today. machine certainly does not exist today. machine certainly does not exist today. The idea of financing is fundamentally The idea of financing is fundamentally The idea of financing is fundamentally linked to the anticipation of end linked to the anticipation of end linked to the anticipation of end customer demand and unlike in 2008, customer demand and unlike in 2008, customer demand and unlike in 2008, where the fundamental issue was the end where the fundamental issue was the end where the fundamental issue was the end customer wasn't paying a mortgage, the customer wasn't paying a mortgage, the customer wasn't paying a mortgage, the end customer is buying more and more AI. end customer is buying more and more AI. end customer is buying more and more AI. And that's a very fundamental economic And that's a very fundamental economic And that's a very fundamental economic fact that we need to keep our eyes on if fact that we need to keep our eyes on if fact that we need to keep our eyes on if we want a healthy system. The end we want a healthy system. The end we want a healthy system. The end customer has to want to pay for AI. As customer has to want to pay for AI. As customer has to want to pay for AI. As soon as that stops being true, we're all soon as that stops being true, we're all soon as that stops being true, we're all in trouble. But right now, the end in trouble. But right now, the end in trouble. But right now, the end customer wants more and more AI. So, customer wants more and more AI. So, customer wants more and more AI. So, what about your job? If the financing what about your job? If the financing what about your job? If the financing works and capacity gets built and works and capacity gets built and works and capacity gets built and capacity, you know, competes in the capacity, you know, competes in the capacity, you know, competes in the market and firms start to buy more AI, market and firms start to buy more AI, market and firms start to buy more AI, aren't they going to start to fire more aren't they going to start to fire more aren't they going to start to fire more people? That's the core question. among people? That's the core question. among people? That's the core question. among workers age 22 to 25 in highly AI workers age 22 to 25 in highly AI workers age 22 to 25 in highly AI exposed occupations was about 19% below exposed occupations was about 19% below exposed occupations was about 19% below where it would have been if it had kept where it would have been if it had kept where it would have been if it had kept pace with less exposed work. That change pace with less exposed work. That change pace with less exposed work. That change did not come from firing. It actually did not come from firing. It actually did not come from firing. It actually came from reduced hiring. Now, you might came from reduced hiring. Now, you might came from reduced hiring. Now, you might think, "Okay, there's your answer. It's think, "Okay, there's your answer. It's think, "Okay, there's your answer. It's costing jobs. It's just costing jobs for costing jobs. It's just costing jobs for costing jobs. It's just costing jobs for young people and they're not being young people and they're not being young people and they're not being hired." You may be right. You may be hired." You may be right. You may be hired." You may be right. You may be wrong. That's the problem with economic wrong. That's the problem with economic wrong. That's the problem with economic data.

data. data. In this case, that same divergence In this case, that same divergence In this case, that same divergence started before generative AI. And so, is started before generative AI. And so, is started before generative AI. And so, is it a continuation of a trend that it a continuation of a trend that it a continuation of a trend that started pre-ChatGPT? Is it something started pre-ChatGPT? Is it something started pre-ChatGPT? Is it something that is causative? It's hard to know at that is causative? It's hard to know at that is causative? It's hard to know at scale. Gusto surveyed 1,051 people who scale. Gusto surveyed 1,051 people who scale. Gusto surveyed 1,051 people who started companies in 2025. 60% used AI started companies in 2025. 60% used AI started companies in 2025. 60% used AI during launch and half said it made during launch and half said it made during launch and half said it made starting significantly faster or starting significantly faster or starting significantly faster or cheaper, but only 3% said they likely cheaper, but only 3% said they likely cheaper, but only 3% said they likely wouldn't have started without it. wouldn't have started without it. wouldn't have started without it. Frankly, I think that we are past the Frankly, I think that we are past the Frankly, I think that we are past the point where Silicon Valley would say we point where Silicon Valley would say we point where Silicon Valley would say we have intelligence capable enough to do have intelligence capable enough to do have intelligence capable enough to do most people's jobs. And yet, and yet, we most people's jobs. And yet, and yet, we most people's jobs. And yet, and yet, we do not see that going in a straight line do not see that going in a straight line do not see that going in a straight line to job replacement. And I think the to job replacement. And I think the to job replacement. And I think the reason is that they what I've always reason is that they what I've always reason is that they what I've always been saying, jobs are complicated. Jobs been saying, jobs are complicated. Jobs been saying, jobs are complicated. Jobs are way more than some raw intelligence are way more than some raw intelligence are way more than some raw intelligence against a problems. Jobs take people to against a problems. Jobs take people to against a problems. Jobs take people to do well, and I think most firms are do well, and I think most firms are do well, and I think most firms are smart enough to see that they actually smart enough to see that they actually smart enough to see that they actually need people in a lot of these roles and need people in a lot of these roles and need people in a lot of these roles and they could augment them with AI. AI can they could augment them with AI. AI can they could augment them with AI. AI can be a tool that people are expected to be a tool that people are expected to be a tool that people are expected to use, but they cannot reliably with use, but they cannot reliably with use, but they cannot reliably with confidence with a dollar on the line confidence with a dollar on the line confidence with a dollar on the line replace people.

replace people. replace people. And this is something where I think the And this is something where I think the And this is something where I think the the naive assumption in the valley is in the naive assumption in the valley is in the naive assumption in the valley is in conflict with how people who run conflict with how people who run conflict with how people who run ordinary businesses actually see the ordinary businesses actually see the ordinary businesses actually see the world. They don't see the world as, world. They don't see the world as, world. They don't see the world as, "Okay, this model has gain of function. "Okay, this model has gain of function. "Okay, this model has gain of function. It's intelligent. I can trust it with It's intelligent. I can trust it with It's intelligent. I can trust it with everything." They see the world as, "Is everything." They see the world as, "Is everything." They see the world as, "Is someone going to show up to work and be someone going to show up to work and be someone going to show up to work and be accountable?" So, when you see the next accountable?" So, when you see the next accountable?" So, when you see the next giant AI financing announcement, giant AI financing announcement, giant AI financing announcement, probably tomorrow, ask yourself three probably tomorrow, ask yourself three probably tomorrow, ask yourself three things. Is there a firm customer things. Is there a firm customer things. Is there a firm customer contract? How concentrated is the contract? How concentrated is the contract? How concentrated is the revenue? Can the GPUs earn enough over revenue? Can the GPUs earn enough over revenue? Can the GPUs earn enough over the life of the debt after power costs the life of the debt after power costs the life of the debt after power costs and construction delays and falling and construction delays and falling and construction delays and falling token prices to be worth it? And who token prices to be worth it? And who token prices to be worth it? And who loses money first if the forecast is loses money first if the forecast is loses money first if the forecast is wrong? wrong? wrong? Those questions will tell you whether a Those questions will tell you whether a Those questions will tell you whether a project is well-financed. I don't think project is well-financed. I don't think project is well-financed. I don't think AI is a bubble in the sense that the AI is a bubble in the sense that the AI is a bubble in the sense that the underlying market is detached from real underlying market is detached from real underlying market is detached from real demand. In fact, the evidence shows that demand. In fact, the evidence shows that demand. In fact, the evidence shows that it's attached to very real, very rapidly it's attached to very real, very rapidly it's attached to very real, very rapidly growing external customer revenue and growing external customer revenue and growing external customer revenue and demand. The revenue, the adoption, the demand. The revenue, the adoption, the demand. The revenue, the adoption, the contracts, the useful work are growing contracts, the useful work are growing contracts, the useful work are growing way too quickly for that claim to fit way too quickly for that claim to fit way too quickly for that claim to fit neatly. But, I do think some companies neatly. But, I do think some companies neatly. But, I do think some companies are overvalued. Some capacity is going are overvalued. Some capacity is going are overvalued. Some capacity is going to be built in the wrong place. Some to be built in the wrong place. Some to be built in the wrong place. Some investors will lose a lot of money. And, investors will lose a lot of money. And, investors will lose a lot of money. And, that happened with railroads, too. And, that happened with railroads, too. And, that happened with railroads, too. And, in that situation, the economy keeps the in that situation, the economy keeps the in that situation, the economy keeps the asset. So, you're not trapped in that asset. So, you're not trapped in that asset. So, you're not trapped in that world between losing your retirement and world between losing your retirement and world between losing your retirement and losing your job. What you're watching is losing your job. What you're watching is losing your job. What you're watching is a technology that creates new financial a technology that creates new financial a technology that creates new financial engineering to enable demand to be engineering to enable demand to be engineering to enable demand to be served by an extraordinary amount of

served by an extraordinary amount of served by an extraordinary amount of built capacity. That will change what built capacity. That will change what built capacity. That will change what work companies hire for. It will lower work companies hire for. It will lower work companies hire for. It will lower the cost of starting a business. And, it the cost of starting a business. And, it the cost of starting a business. And, it will do all of that at the same time. will do all of that at the same time. will do all of that at the same time. And, the way that's possible is the And, the way that's possible is the And, the way that's possible is the financial engineering that Nvidia did financial engineering that Nvidia did financial engineering that Nvidia did this week. Every major tech needs two this week. Every major tech needs two this week. Every major tech needs two inventions. And, Nvidia is now trying to inventions. And, Nvidia is now trying to inventions. And, Nvidia is now trying to build that second one for AI.