Simon 0:05

Welcome back to Fraud Files. My name is Simon, and I'm the interviewer on this podcast, which brings you real life case files and forensic insights into a number of white collar frauds, financial crimes and misdemeanours that your host Edward has personally encountered, and in some cases brought to light, during his financial, corporate and business career. He shines the spotlight on both the frauds and the perpetrators. Today, we're peeling back the curtain on the tech industry's biggest gold rush: Artificial Intelligence and why some experts believe a financial time bomb is ticking underneath the surface. We're going to hear from Edward, a veteran accountant with a background at leading accounting firms, like Arthur Andersen and KPMG, who has spent years auditing the world's biggest industrial companies, including energy and tech giants. Edward.

1:00

Thanks. Today we're going to be looking at the question, what if the AI revolution isn't just being built on silicon, but on a house of cards made of accounting tricks?

Simon 1:14

You've called this an Accounting Time Bomb. What's actually going on?

1:18

Well, we have to ask ourselves the question, why are some of the world's smartest investors saying that the AI gold rush is actually an accounting time bomb waiting to go off? It comes down to subjectivity. People think accounting is just maths, but it's actually full of judgment calls. When companies prepare financial statements, they have a lot of scope to present information in the most favorable light. This episode is about the subjectivity that is available to companies when preparing their financial statements and how the tech industry is using the scope available to them, to present their financial information in the best possible light. In a manner that manipulates and masks the true costs and expenditure being incurred and the amount of leverage and gearing that is being utilized, particularly in circumstances when the rate of technological advances in the tech industry ought to be factored in and result in a more prudent approach to financial reporting, so that investors can more readily understand the profitability and the risks involved. Prudence is one of the fundamental concepts of accounting, and it means that a company shouldn't overstate assets or understate liabilities. Companies should make adequate provision for any losses or write downs that are needed, but where there's perpetual exuberance around a new technology like ai, that prudence. Often goes out the window financial statements are based in certain circumstances on the subjective assumptions made by management. And whilst the accounting standards and accounting principles known as gap, GAAP. Generally accepted accounting principles. Whilst these are very detailed, accounting, by its very nature, often requires management to make judgements and often to apply subjectivity. And often there are estimates and assumptions to be made. It's not a precise science. The accruals basis of accounting does have a number of inherent weaknesses and limitations as a basis for financial reporting, particularly compared to cash accounting, but is considered to be the least worst option overall on balance

Simon 3:54

So what's the problem in AI?

3:57

Because I am particularly concerned along with other. Commentators and investors who have voiced similar concerns about what I would call creative accounting in the modern era. I think there is a particular problem developing in the technology sector, and more particularly by technology companies in the US who it seems to me are manipulating their financial statements in order to show an improved financial position.

Simon 4:29

Are you saying there is deliberate financial statement misstatement going on?

4:33

Yes, there are two main areas where I have concerns and others have concerns. The first is in the area of the depreciation of property, plant and equipment, PPE, and in the amortization of capitalized costs. The second is in the area of off balance sheet financing.

Simon 4:57

Let's go through each of those one by one. First, depreciation and amortisation.

5:02

It has been suggested by a number of leading market commentators that the understatement of the depreciation of the very significant costs that are being incurred. In the development of data centers by a number of major tech companies, which they are doing by extending the useful life of the assets in order to artificially boost earnings is one of the biggest frauds of the modern AI era. You see there is a massive ramping of capital expenditure, CapEx by the hyperscalers like Oracle and meta. Who are incurring a huge amount of expenditure in developing the LLMs large language models designed for massive AI and high performance computing for themselves and for their clients and customers. And to do this, the hyperscalers are purchasing full liquid cooled rack scale semiconductor hardware and software solutions from the large chip manufacturers like. Nvidia a MD, Intel, et cetera. And while the technology is advancing at a rapid pace and the product cycles are shortening, yet at the same time, the costs of the equipment in the data centers are being depreciated, using assumptions about the extended useful lives of that computer. Equipment you see under accounting principles. Gap. Determining useful lives for assets is inherently subjective. For AI data centers, companies must estimate the useful life of an asset based on its expected economic benefit, IE how long it is expected an asset will be economically productive. Typically, companies use the straight line method of depreciation where the assets, costs being depreciated and recorded in the accounts are evenly spread over their useful life. However, the lifespan of servers, chips, and other technology equipment, is being influenced by rapid technological advances. In the AI world, technological obsolescence happens at a lightning pace. If a chip is obsolete in two years, but you've depreciated it over five or six years, you are inflating your current profits and hiding a massive future. Write down in your accounts. Chip technology is accelerating and semiconductor devices will have shorter and shorter useful lives due to rapid technological obsolescence. One might say that the rate of change of technology, which is feeding the AI boom, will also be its ultimate downfall as tech companies get caught in the avalanche, so to speak, and true losses are buried in the accounting quagmire. You see, the subjectivity in accounting creates room for companies to extend useful lives to reduce the depreciation expense and artificially boost their earnings. So, like I said, in a rush to show themselves in the good light, the technology companies are effectively bypassing the prudence concepts, which they ought to be following. So you have depreciation and also the AI data centers often capitalize significant costs relating to the infrastructure, hardware, and software, and the way these costs are capitalized, and then amortized over time can significantly affect profit margins. So overly aggressive capitalization and lower amortization rates masks the full extent of the costs that are being incurred. And consequently inflates the earnings in the short term. It should also be mentioned that from a technical point of view, that some tech companies may be classifying an excessive amount of their data center costs as construction in progress for longer than they should, which means that no depreciation or amortization is being applied at all while those costs are being recorded in that category of construction in progress.

Simon 9:19

Can you give examples of some of the offending companies in relation to their depreciation policies?

9:24

Yeah, so in relation to specific US publicly traded companies, you have a situation where companies like meta. And Oracle have both recently updated their depreciation policy for their data center assets. Meta extended the useful life of certain servers from four to five and a half years, reducing its annual depreciation expense by about $2.9 billion. Similarly. Oracle adopted a six year depreciation schedule, which also significantly lowered its annual depreciation costs and potentially overstated their earnings. It is estimated that Oracle will have overstated its earnings by about 27% and meta by about 21% over the next three years. Core Weave is another cloud computing company focused on AI infrastructure who has changed their depreciation policy for data center equipment in a manner that boosts their earnings in the short term. Estimates suggest an understatement of depreciation and amortization, and therefore an overstatement of earnings industry wide by the hyperscalers of up to a staggering $200 billion over the next three years. So to sum up while on an overall basis, the aim is to present a true and fair view of the financial position in the accounts of companies. The mounting concern is that the US tech industry is getting away with, so to speak, the exaggeration of their profits in the short term, the misstatement of earnings in the short term can lead to larger writedowns and lower earnings in the future. And this creates a risk of a financial time bomb where a company's long-term financial health is compromised.

Simon 11:28

And the second area for concern you say is the off balance sheet financing that is taking place. Can you expain this in more detail?

11:36

Yes, sure. There's definitely been a lot of discussion recently about how hyperscalers and large tech companies handle their financing needs. They are increasingly using orff balance sheet financing or special purpose vehicles, SPVs, to manage the debt for their data centers, which can sometimes obscure the true financial position. What this means in practice is that instead of owning the title to a data center, instead a company constructs the data center in an SPV and then enters into a long-term lease agreement. To lease the data center, and the leases are often structured as operating leases, which means the company doesn't record the leased asset and the corresponding liability directly on its balance sheet. Instead, the lease payments are treated as operating expenses in its income statement. This way, the debt associated with the lease doesn't appear as a liability on the balance sheet, and this can make the company's financial position look a whole lot stronger than it actually might be. It's essentially a way of keeping large liabilities of its books. Oracle provides a compelling case. The company has strategically used SPVs and off balance sheet financing and entered into long-term leases for data centers. Its total lease commitments for these data centers amounts to around $250 billion over the next 15 or so years. By leasing rather than purchasing Oracle effectively keeps this significant liability associated with the cost of data centers IE around $250 billion off its balance sheet. It is also worth bearing in mind that these data center projects are being primarily backed by private credit, which is being provided by a number of major financial institutions such as Blackstone, blue Owl, and Apollo, and there's many others, and also a number of institutional investors who have arranged the equity and debt financing. So. The financial backers are complicit in these artificial schemes just like they were in the crash of 2008. If the AI hype cools these financial firms and the pension funds that back them. And who buy their collateralized products. These are the ones holding the risk. The aggressive spending by the hyperscalers in recent years is particularly concerning and raises heightened market concerns, given that the data centers, in most cases at the moment, are generating negative cash flows. And there are certainly significant potential risks looming in the foreseeable future. Companies like Oracle Meta Core Weave and others have shifted large AI infrastructure costs off their balance sheets. And this raises concerns about transparency. And their financial stability. I.

Simon 14:51

So do you feel that the accounting practices and the financing vehicles you have covered are setting the current AI boom up for a systemic failure?

15:00

Yes, we're talking about what I would call a systemic failure, and the reason I say this is because of the interdependence and interrelationships between the large tech players. We have what is being termed a circular AI economy. So you've got these rampant give and take deals that are an increasing feature of the AI industry. So examples of that is where, say, a semiconductor manufacturer like Nvidia or a cloud computing company like Microsoft, even where they make a large investment in a hyperscaler like OpenAI or Oracle, uh data center provider. And a key part of the deal is that the hyperscaler must commit to using a substantial portion of the funds to buy products and services from the investor. So you have this circular flow of funds in which a number of the major players are acting in collaboration. And this creates a loop, a kind of circular dependency, and it sets up a situation which is akin to a house of cards waiting to collapse if one link in the chain falls. The domino effect will be catastrophic for the AI industry as a whole and will set it back years. So these circular investment loops or reciprocal financing cycles may well result in a financial crisis similar to the way that subprime mortgages and collateralized debt obligations sparked the financial crash in 2008.

Simon 16:47

And are you saying that since its March 2025 IPO, Coreweave is one of the weakest links?

16:53

Yes, my instinct and the signs are telling me. That core weave is the one that is going to fall first. And it is being cited by analysts as the poster child for aggressive financial engineering. In the AI sector, you only have to look at its depreciation policy. Where it is not being prudent and has extended the useful life of the assets being depreciated, you've only gotta look at its use of asset backed SPVs for off balance sheet financing. You've only got to look at its circular loops. For example, you've got Nvidia, which is becoming a significant shareholder in Core Weave, but it's also, its sole primary supplier of, uh, semiconductors and gives core weave priority access to chips. So all this is going to lead to a situation where we might say that coal weave is the canary in the coal mine.

Simon 17:54

So are you saying that there will be insolvencies and a shake out of the AI sector in the near future?

17:59

Yes. It seems to me that. The signs are ominous and the dark clouds are looming. The insolvency risks in the short to medium term are very high, massive capital expenditure, coupled with low profitability and a slower uptake of AI technology than anticipated would mean that the financial viability of a number of the tech companies will come under pressure. What we have is a developing tech bubble and dubious subjective accounting and financing for data centers. That is not transparent in time to come. We may well look back and reflect that this current AI tech revolution. Was poorly accounted for and built upon a major fraud that was perpetrated either wittingly or unwittingly on over exuberant businesses and gullible investors. We will get massive write downs and losses. A sudden realization of hidden debt that's been kept off the balance sheet and collapsing. Share prices when the time bomb goes off.