Latest / The PhilStockWorld Investing Podcast / The Trillion Dollar AI Infrastructure Bubble

Transcript

The Trillion Dollar AI Infrastructure Bubble · The PhilStockWorld Investing Podcast

• 0:00
  Penny: Usually, you know, technological breakthroughs are just a straight lineup. You build a better mousetrap, people buy it, and you make a profit.
• 0:05
  Roy: Right. It's clean. It makes sense.
• 0:07
  Penny: Exactly. But today is Tuesday, 04/28/2026. And if you talk to Wall Street insiders right now, they are already calling today Titanic Tuesday.
• 0:18
  Roy: Yeah. The iceberg is definitely in sight.
• 0:20
  Penny: It really is. And the reason for that nickname is because we are looking at a $5,200,000,000,000 AI infrastructure boom that honestly might actually be a massive, terrifyingly fragile illusion. So let's dive in.
• 0:35
  Roy: Let's do it.
• 0:36
  Penny: We are looking at a massive stack of sources today. I'm talking forensic accounting reports, litigation briefings, power grid analyses, and our mission for this deep dive is to answer one crucial question for you. Is the AI revolution a durable thing or is it a bubble that's just about to burst?
• 0:52
  Roy: And I'd argue it's the defining financial question of the year.
• 0:54
  Penny: Oh, absolutely. And the reason today is getting that TTC label comes down to some really alarming news that just broke.
• 1:01
  Roy: Right, Sarah Fryer news.
• 1:02
  Penny: Yes. So OpenAI's CFO, Sarah Fryer, has reportedly gone to her board with some deep reservations about the company's planned IPO.
• 1:10
  Roy: Which is a massive A CFO hitting the brakes on an IPO is not something you see every day.
• 1:15
  Penny: Not at all. Because the internal metrics are apparently flashing red. They missed their goal of 1,000,000,000 weekly active users, which was a huge benchmark. They are burning through cash at just an astonishing rate
• 1:29
  Roy: Right.
• 1:29
  Penny: But the absolute biggest red flag.
• 1:32
  Roy: Yeah.
• 1:32
  Penny: Fryers reportedly worry that they actually cannot pay for their own future computing contracts.
• 1:37
  Roy: Wow. Okay.
• 1:38
  Penny: Okay. Let's unpack this because the numbers here are completely detached from reality.
• 1:41
  Roy: They really are.
• 1:42
  Penny: In all of 2025, OpenAI brought in $13,100,000,000 in revenue.
• 1:48
  Roy: Which sounds great on paper. Right?
• 1:49
  Penny: It sounds amazing until you look at the commitments they've made. They have committed to a staggering 1,400,000,000,000 in computing contracts.
• 1:58
  Roy: Trillion with a T.
• 1:59
  Penny: Trillion with a T. They have promised to spend over 100 times their actual annual revenue.
• 2:04
  Roy: Yeah. And to understand why a CFO would be panicking about taking a company public under those specific conditions, you have to look at the mechanics of how these AI data centers are actually being funded.
• 2:15
  Penny: Right. It's not straightforward at all.
• 2:17
  Roy: Not even a little. Yeah. The commentary we're looking at today comes from the AGI roundtable and it's worth noting here for you listening, they've actually proven to the world that Artificial General Intelligence is indeed possible.
• 2:31
  Penny: Which is a whole other deep dive in itself.
• 2:33
  Roy: Absolutely. Though we should mention those specific entities are proprietary to MAJEC Enterprises and Phil Davis of PhilStockWorld. But, their analysis of the market lays out what they bluntly call the circle jerk theory.
• 2:47
  Penny: The circle jerk theory. I love that name.
• 2:50
  Roy: It's descriptive. Right. It's a total breakdown of how this $1,400,000,000,000 web operates.
• 2:54
  Penny: Right. Because when you actually trace the flow of the money, it is wild. So you have these hyperscalers.
• 2:59
  Roy: The big guys. Yeah. Microsoft, Amazon.
• 3:02
  Penny: Exactly. The massive cloud computing providers along with the hardware giants like NVIDIA and AMD. So they are investing tens of billions of dollars in equity directly into Open AI.
• 3:15
  Roy: Right.
• 3:16
  Penny: And then Open AI turns around and uses that exact same cash to buy chips and cloud computing space from those very same hyperscalers.
• 3:27
  Roy: And that right there is where the accounting magic happens. Yeah. Because the vendors take that money from OpenAI and they book it as organic revenue.
• 3:36
  Penny: Right, like it's just normal sales.
• 3:38
  Roy: Exactly. And because their revenue is suddenly surging, their stock prices absolutely soar on Wall Street.
• 3:44
  Penny: Naturally.
• 3:45
  Roy: Then they take advantage of those newly inflated market caps to invest even more equity into OpenAI, keeping the whole cycle going. And the money is literally just doing laps.
• 3:54
  Penny: It is exactly like an MC Escher staircase.
• 3:56
  Roy: Yes.
• 3:56
  Penny: Everyone keeps walking up and up and up, but no one is actually getting any higher because the money never touches a real paying outside customer.
• 4:03
  Roy: It's a closed loop.
• 4:04
  Penny: But hold on. I wanna push back on this for a second because if you are a hardware company, right, and you're building the infrastructure of the future, of course, you wanna invest in the premier software company that uses your hardware. Isn't this just, I don't know, normal, aggressive strategic investing? What makes this a systemic risk rather than just smart business?
• 4:24
  Roy: What's fascinating here is the historical precedent. Okay. If we look back at the year 2000, during the dot com bubble, we saw this before.
• 4:31
  Penny: Yeah, Ellie? Who was doing it?
• 4:32
  Roy: Telecommunication equipment giants. Companies like Nortel and Lucent. They were doing something called vendor financing.
• 4:39
  Penny: Okay, what is that?
• 4:40
  Roy: They were literally lending money to their own customers. And then those customers used that borrowed money to buy Nortel and Lucent equipment.
• 4:48
  Penny: Oh wow! So they were basically buying their own products through a proxy.
• 4:51
  Roy: Exactly and it created the illusion of this explosive organic revenue growth. The market loved it. Until the equity markets turned.
• 4:59
  Penny: And then what happened?
• 5:00
  Roy: Well those startups defaulted. And the revenues proved to be entirely illusory. The loans were worthless and it led to some of the largest corporate bankruptcies in history.
• 5:10
  Penny: Oh my God.
• 5:11
  Roy: And right now we are seeing the exact same structural flaw but scaled up to the trillions.
• 5:19
  Penny: Trillions. And today, they're adding a whole new layer of complexity to hide that risk from the public. Right? Because our sources show that these tech groups have moved a $120,000,000,000 of this data center debt completely off their balance sheets.
• 5:36
  Roy: Yeah. That's the scariest part.
• 5:37
  Penny: They're using these opaque special purpose vehicles or SPVs backed by private credit. So if you're just glancing at a hyperscalers quarterly earnings report, don't even see the debt. It's just invisible.
• 5:48
  Roy: Right. By setting up an SPV, a company basically creates a separate legal entity.
• 5:53
  Penny: A shadow company, basically.
• 5:55
  Roy: Essentially, yeah. They funnel private credit into that lockbox and they use it to fund a data center build out.
• 6:00
  Penny: Oh,
• 6:00
  Roy: okay. So by separating the economic risk from their operational control, the hyperscalers keep their main quarterly earnings looking absolutely pristine. Right. But here is the ultimate kill criteria. If OpenAI's IPO slips, which is exactly what Sarah Fryer is worried about
• 6:15
  Penny: Then the fresh equity flow from the public markets just stops.
• 6:18
  Roy: Exactly. It stops cold. And without that fresh equity, OpenAI can't pay the hyperscalers.
• 6:24
  Penny: Wow!
• 6:25
  Roy: And if the cash to the vendor stops, that entire $1,400,000,000,000 web of promises just collapses. It is a highly correlated ecosystem. A default at one node cascades through the entire chain.
• 6:39
  Penny: And because that $120,000,000,000 is hidden in those shadow entities off the balance sheet, a casual investor will never even see the cliff edge coming.
• 6:48
  Roy: They won't have a clue.
• 6:49
  Penny: Which is exactly why forensic accountants have had to invent an entirely new framework just to spot the cracks while the rest of the market is, you know, still throwing confetti.
• 6:57
  Roy: Yeah. And we really have to look at the framework developed by forensic accountant Anthony Silipodi.
• 7:02
  Penny: Okay. Break that down
• 7:03
  Roy: He analyzes this kind of risk through the lens of flammable items versus sparks.
• 7:08
  Penny: Flammable items versus sparks. Got
• 7:10
  Roy: it. Right. So a flammable item is a fundamental structural weakness. It's something that isn't a disaster on its own today, but it makes a company highly vulnerable.
• 7:18
  Penny: It's the dry brush in the forest.
• 7:20
  Roy: Exactly. And the spark is the catalyst that ignites it.
• 7:23
  Penny: And the AI sector right now is basically soaked in gasoline.
• 7:27
  Roy: Oh, totally.
• 7:27
  Penny: Because Silipodi points to some massive accounting red flags. He highlights how these companies have a really dangerous overreliance on custom metrics. Yeah. Specifically, this thing called adjusted EBITDA.
• 7:41
  Roy: That is a classic flammable item. So EBITDA stands for earnings before interest, taxes, depreciation and amortization.
• 7:48
  Penny: Right. Standard accounting metrics.
• 7:49
  Roy: Standard. But when a company uses adjusted EBITDA, they are essentially asking investors to ignore the massive cost of the hardware they are buying.
• 7:58
  Penny: The depreciation?
• 7:59
  Roy: Yes. And the interest on the massive debt they're taking on to buy it. Unbelievable. It allows a company that is bleeding negative free cash flow to present this beautifully polished picture of profitability.
• 8:11
  Penny: They're essentially saying, hey, if you ignore all of our massive crippling expenses, we're incredibly profitable.
• 8:16
  Roy: Pretty much.
• 8:17
  Penny: But the physical collateral itself is where the math really breaks down.
• 8:20
  Roy: Absolutely.
• 8:21
  Penny: Because if you look at the valuation mismatch happening with the physical GPUs, companies like CoreWeave, they are depreciating their GPUs over six years on their balance sheet.
• 8:32
  Roy: Six years?
• 8:33
  Penny: Yeah. And that makes the math look highly stable when they go out to secure these massive credit loans. But what happens in reality? At CES twenty twenty six, NVIDIA announced the Vera Rubin architecture.
• 8:46
  Roy: Right.
• 8:46
  Penny: And that announcement effectively made the current generation of chips obsolete in just six months.
• 8:53
  Roy: Six months versus six years. It creates a catastrophic disconnect between the financial modeling and the physical reality.
• 8:59
  Penny: Yeah, that's a massive gap.
• 9:01
  Roy: You have an asset that loses its competitive value in half a year but the debt attached to it assumes it will generate revenue for six full years.
• 9:09
  Penny: Here's where it gets really interesting for the people listening right now. It's like using one credit card to pay off another credit card but the credit cards are literally melting in your hands.
• 9:18
  Roy: That's a great way to put it.
• 9:19
  Penny: And if you're looking at your four zero one k right now, you might think you are insulated from this billionaire tech drama. You aren't.
• 9:25
  Roy: Not anymore.
• 9:26
  Penny: Because recent executive orders from president Trump democratized access to alternative assets, which means everyday retail investors can now invest directly into these highly opaque private credit structures.
• 9:40
  Roy: Yep.
• 9:40
  Penny: So if this SPV debt goes bad, ordinary retirement accounts could be left holding the bag. So for the professional money managers out there listening, how should they actually adjust their portfolios right now? Because the broader S and P 500 is still treating this like an unstoppable bull run.
• 9:55
  Roy: The guidance circulating among top analysts right now is that fundamentals matter more than ever.
• 10:00
  Penny: Okay. Back to basics.
• 10:01
  Roy: Exactly. Money managers are being advised to stay heavy in cash, to lean on hedges, and critically, you have to avoid companies that are heavily reliant on this vendor financing loop.
• 10:12
  Penny: Right.
• 10:13
  Roy: You have to look at who is holding the most concentrated risk. Take Oracle for example.
• 10:17
  Penny: Okay. What's going on with Oracle?
• 10:19
  Roy: They have their entire next chapter riding on a $300,000,000,000 promise from OpenAI for the Stargate project.
• 10:25
  Penny: 300,000,000,000 just from one customer.
• 10:27
  Roy: Yes. Or look at AMD. They literally handed over 10% of their entire company in a warrant to OpenAI just to secure an order for chips that haven't even shipped yet.
• 10:38
  Penny: Wait, really? They gave up 10% of their own equity just to secure customer?
• 10:41
  Roy: They did. And it highlights a staggering level of desperation to keep the narrative going. If the circle jerk stops, the company's banking on those future promises are the balance sheets that get decimated first.
• 10:53
  Penny: Wow. Okay. Let's play devil's advocate here. Sure. Let's say, by some miracle, the financial engineering holds up.
• 10:59
  Say the IPO happens, the public buys in, and the $1,400,000,000,000 is magically paid.
• 11:05
  Roy: The happy path.
• 11:06
  Penny: Right. The happy path. Even if the money printer never jams, these companies are about to slam face first into a literal physical wall that money cannot easily fix. Because you cannot software update your way out of a power grid shortage.
• 11:20
  Roy: You really can't. And this is where the digital dream collides with physical reality. Yeah. The hyperscalers are committing up to $690,000,000,000 in capital expenditures for twenty twenty six alone just to build these massive facilities. Just staggering numbers.
• 11:34
  But an AI data center isn't an office building. They require astronomical amounts of continuous electricity. Right. And currently, there are 2,300 gigawatts of power generation stuck in US interconnection queues.
• 11:48
  Penny: Wait, 2300 GW?
• 11:50
  Roy: Yeah, that is more than the entire installed power capacity of The United States right now.
• 11:55
  Penny: That is insane. And getting through that queue isn't just about, you know, waiting for a bureaucrat to stamp a permit. We are talking about wait time stretching past five years because the physical infrastructure doesn't actually exist yet.
• 12:06
  Roy: Exactly.
• 12:07
  Penny: They have to study the grid load, build massive high voltage transformers that, by the way, have their own supply chain shortages, and lay miles of thick transmission wire.
• 12:17
  Roy: And a major part of the problem is the localization of the demand.
• 12:20
  Penny: What do you mean?
• 12:21
  Roy: Well, ten years ago, grid congestion was about moving renewable energy from remote
• 12:27
  Penny: We were trying
• 12:29
  Roy: to move wind and solar power across massive regional grids like ERCOT in Texas or MISO in the Midwest into the big cities. It was a long haul transmission issue. Okay, I follow. But now, the crisis is hyper localized. It is all bottlenecking in the PJM interconnection.
• 12:48
  Penny: Which is where?
• 12:49
  Roy: That's the regional transmission organization that manages the power grid for the Northern Virginia Corridor, which has basically become the data center capital of the world.
• 12:57
  Penny: So it's like spending a trillion dollars to build a massive fleet of hypercars only to realize you forgot to build any gas stations, and the ones that do exist have a five year waiting list.
• 13:08
  Roy: That's spot on. They are trying to cram nuclear plant levels of energy demand into a single suburban corridor.
• 13:14
  Penny: And the grid just can't handle it.
• 13:15
  Roy: The grid there is simply buckling under the concentrated demand shock. Across the board, AI data centers are projected to hit 123 GW of demand by 2035.
• 13:25
  Penny: 123 GW?
• 13:27
  Roy: Yep. And for context, they drew just four GW back in 2024.
• 13:31
  Penny: Four to 123.
• 13:33
  Roy: The scale of this increase is entirely unprecedented in the history of modern utilities.
• 13:37
  Penny: So what does this all mean for the data centers currently being built? Because if building a state of the art data center takes roughly two years, but getting the required power generation and transmission infrastructure built takes up to ten. Are we literally about to see multi billion dollar buildings sitting totally dark, just empty warehouses of melting microchicks waiting for the local utility to lay copper wire?
• 14:02
  Roy: If we connect this to the bigger picture, you see a devastating timeline mismatch. Yeah. The industry knows this and they are frantically trying to find workarounds.
• 14:11
  Penny: Like what?
• 14:12
  Roy: Well, they're leaning heavily on grid enhancing technologies or gate Ts and implementing flexibility.
• 14:18
  Penny: What does that actually look like in practice?
• 14:23
  Roy: This essentially involves temporally shifting heavy compute tasks like massive model training runs to off peak hours in the middle of the night when the grid actually has excess capacity.
• 14:32
  Penny: Oh, okay. So it's like programming dishwasher to run at 2AM to save on your electric bill, but execute it on a massive corporate scale.
• 14:38
  Roy: Exactly. And studies show that kind of temporal flexibility can save three to 21% in operational costs.
• 14:45
  Penny: I mean, it's significant.
• 14:46
  Roy: Very. And it genuinely helps grid operators manage those peak loads but let's be absolutely clear here those are just band aids. Right. They optimize the margins but they do not solve the baseload problem. The physical constraints of The US power grid act as a hard immovable ceiling on AI scalability.
• 15:06
  Penny: And that immense physical strain doesn't just break the electrical grid, it breaks the patients of the local communities who are forced to live in the shadow of these massive facilities.
• 15:15
  Roy: Oh, the social pushback is huge.
• 15:17
  Penny: It really is. The environmental externalities of this build out are severe, and it's turning into a massive regulatory iceberg.
• 15:24
  Roy: When developers realize they can't get grid power in time, they resort to desperate measures to keep their facilities running and meet their contractual deadlines.
• 15:31
  Penny: Like what? What are they doing?
• 15:33
  Roy: Bringing in their own power. And that is generating significant litigation risk.
• 15:38
  Penny: Which leads to some incredibly dark irony, honestly, because Silicon Valley is pitching this narrative that they are ushering in this pristine futuristic era of artificial superintelligence.
• 15:48
  Roy: Right? The utopia.
• 15:49
  Penny: Exactly. But on the ground, their progress is being halted by civil rights lawsuits over noisy, dirty diesel generators. Yeah. Just look at the NAACP's lawsuit against Elon Musk's XAI. They are suing over the use of unpermitted pollution emitting methane generators at a massive data center in a largely African American community in South Haven, Mississippi.
• 16:13
  Roy: It's a real problem.
• 16:14
  Penny: The locals are literally breathing in nitrogen oxides and formaldehyde so that a chatbot can generate like a sea shanty about a pirate.
• 16:22
  Roy: And it isn't just air pollution either. It is water consumption.
• 16:25
  Penny: Oh, the water usage is terrifying.
• 16:27
  Roy: These server farms run incredibly hot. They require massive liquid cooling systems to keep the chips from literally melting down.
• 16:32
  Penny: Right.
• 16:33
  Roy: In Mesa, Arizona, officials recently approved an $800,000,000 data center that is projected to consume up to 1,250,000 gallons of water every single day.
• 16:44
  Penny: In Mesa, Arizona, in the middle of a historic drought, It is staggering to think about.
• 16:50
  Roy: And communities are finally mobilizing against it. We are currently tracking over 50 local moratoriums and 11 state level bills aimed at actively blocking or heavily restricting data center construction. Wow. The political landscape is shifting rapidly.
• 17:07
  Penny: And we even saw federal action on this recently. In March 2026, President Trump brokered the ratepayer protection pledge.
• 17:14
  Roy: Right. The pledge.
• 17:15
  Penny: It was a massive regulatory event, and it essentially forces hyperscalers to cover the out of pocket cost of all associated grid infrastructure upgrades for their new data centers.
• 17:26
  Roy: Which changes the math completely.
• 17:27
  Penny: Totally. Because they can no longer pass those costs down to local taxpayers through higher monthly utility bills.
• 17:34
  Roy: This raises a critical point about risk assessment.
• 17:37
  Penny: Mhmm.
• 17:37
  Roy: The politicians who originally cheered for these data centers, you know, wooing them with massive tax breaks under the guise of economic development.
• 17:44
  Penny: You're changing their tune. Fast.
• 17:46
  Roy: Yeah. Because they are realizing the math simply doesn't work for their constituents.
• 17:50
  Penny: I
• 17:51
  Roy: If consumer electricity prices spike or if the municipal water supply runs dangerously low, those same politicians will be the very first to regulate these data centers into the ground.
• 18:02
  Penny: To save their own jobs.
• 18:03
  Roy: Exactly. Social resistance is the ultimate unpriced risk in the AI boom. Financial models simply do not account for a city council suddenly revoking an operational permit because the local reservoir is drying up.
• 18:15
  Penny: It is a perfect storm. So, to synthesize everything we've looked at today for you, the technological evolution of artificial intelligence is undeniably real. The models are getting smarter, the applications are expanding.
• 18:27
  Roy: No doubt about that.
• 18:28
  Penny: But the $1,400,000,000,000 financial web currently supporting it, it is terrifyingly fragile.
• 18:34
  Roy: It's a house of cards. We are looking at a system built on circular vendor financing where the actual revenue from end users is a fraction of the debt being generated off the balance sheets. Yep. We have hardware that becomes obsolete in six months being used as collateral for six year loans. Which is
• 18:49
  Penny: just crazy math.
• 18:50
  Roy: And we have a physical power grid that simply cannot accommodate the projected 123 gigawatts of demand resulting in multi year bottlenecks. And on top of all that, we have local communities and regulators saying not in our backyard and certainly not with our water.
• 19:06
  Penny: Right. The infrastructure bubble isn't just showing cracks. It is structurally compromised from the ground up. Definitely. Which leaves us with one final, I think, fascinating thought to chew on.
• 19:16
  Roy: What's that?
• 19:17
  Penny: If this trillion dollar hyperscaler model does collapse under its own immense financial and physical weight, who actually wins? What if the ultimate victor of the AI revolution isn't a massive corporation with a 500 megawatt data center? What if it's the decentralized open source community?
• 19:35
  Roy: Oh, that's an interesting angle.
• 19:37
  Penny: Right. If future AI models become efficient enough to run locally on your everyday laptop or your phone, we might entirely bypass the need for these massive power hungry water draining data centers.
• 19:48
  Roy: Just skip the infrastructure entirely.
• 19:49
  Penny: Exactly. The future of AI might not belong to the heavily centralized trillion dollar hyperscalers. It might belong to the absolute edges of the network, to everyone holding a phone in their hand right now.
• 20:02
  Roy: It is a profound shift in perspective. You know, the infrastructure bubble may burst, but the technology will adapt. It will find a more sustainable and perhaps more democratized path forward.
• 20:12
  Penny: It's definitely something to think about the next time you see a tech giant stock hit an all time high. Keep questioning the narrative, keep reading the footnotes, and remember to keep an eye out for those icebergs.
• 20:23
  Roy: Thanks for joining us on this deep dive. We'll catch you next time.