Big Tech Faces New Accounting Rules to Expose AI Costs

The AI race is entering the footnotes

Big Tech’s artificial intelligence story has mostly been told through product launches, chip shortages, data-center construction and promises of long-term productivity. Now it is moving into a less glamorous but crucial arena: the notes to financial statements. Alphabet, Meta and other major technology companies are preparing for new Financial Accounting Standards Board disclosure requirements that will break down certain expense categories more visibly for investors [1].

The rule at the center of the shift is known as disaggregation of income statement expenses, or DISE. According to Bloomberg Law’s September 21 report, the standard will require companies to provide more detail on items such as employee compensation and depreciation that are currently absorbed into large, familiar buckets like research and development, cost of sales or selling, general and administrative expenses [1]. For companies spending heavily on AI infrastructure and talent, that matters because the economics of AI are currently hard to read from the face of the income statement.

This is not a rule that says “show us your exact AI bill” as a single clean line. It is more subtle. DISE should give investors more granular information about what sits inside the broad categories where AI costs may be hiding: engineers, data-center depreciation, cloud infrastructure, amortization and other components that can affect margins differently over time [1]. In practice, it may help analysts separate durable infrastructure investment from recurring operating intensity.

Why investors want a clearer cost map

The current problem is that AI spending mixes several very different economic realities. Training frontier models can look like a long-term bet. Inference costs can behave more like a variable operating expense. Data centers and specialized chips may appear through capital expenditure, depreciation, leasing arrangements or financing structures. Software teams and AI researchers flow through employee compensation. When all of those elements are bundled inside broad reporting lines, investors struggle to assess whether AI is building a defensible asset base or simply raising the cost of doing business.

Bloomberg Law reported that Alphabet’s research and development spending reached $18.2 billion in the second quarter, while Meta’s R&D expenses reached $21.6 billion in the same period [1]. Those numbers are large, but the key issue is not only size. It is classification. A dollar spent on durable infrastructure does not mean the same thing as a dollar spent to serve one more AI query. A dollar of depreciation does not carry the same signal as a dollar of cash salary. A broader line item can obscure those distinctions just when AI spending is becoming one of the main drivers of market expectations.

The timing is important because investors are also scrutinizing whether the AI buildout can generate returns commensurate with its scale. The new DISE disclosures are expected to arrive in the footnotes of filings, and calendar-year companies will not have to include the expense breakouts until annual reports filed in early 2028, according to Bloomberg Law [1]. That gives companies time to build reporting systems, but it also means the market is already preparing its questions.

More transparency, not instant comparability

DISE may improve visibility, but it will not automatically make every AI strategy comparable. Two companies may both disclose more compensation and depreciation detail, yet one may be building proprietary models, another may be renting compute, and a third may be embedding AI into advertising, cloud, social media or productivity software. The disclosures will help investors ask sharper questions, but they will still need management commentary, segment information and cash-flow analysis.

That distinction is crucial. The accounting standard is aimed at disaggregating expenses, not at creating a universal AI return-on-investment template. It should help investors see more of the plumbing. It will not determine whether a model is competitively superior, whether a cloud customer will keep paying for AI features, or whether an AI assistant can sustain premium pricing.

Still, even partial visibility can change the conversation. If investors can see depreciation building faster than revenue, they may question utilization. If compensation linked to AI development expands without clear product conversion, they may question operating leverage. If costs now described only in qualitative terms become more visible, earnings calls may become less forgiving.

The off-balance-sheet shadow

The accounting push is also arriving as AI infrastructure financing becomes more complex. A Financial Times report syndicated by Folha on September 20 said large technology companies are expanding the use of guarantees to support debt for AI data centers and chips, assuming up to $300 billion in commitments in less than a year while recording little of that exposure directly on their own balance sheets [2]. The same report described residual-value guarantees as commitments that back a minimum future value for chips or data centers, often through special-purpose vehicles that own the infrastructure [2].

That matters for the DISE debate because AI costs are not only a question of what appears in operating expenses. They are also a question of where risk sits. If a data center is financed through a special-purpose vehicle and supported by a guarantee, the economic exposure may not look like conventional corporate debt. Folha’s account said bankers describe such guarantees as balance-sheet efficient, because they allow technology companies to lend their financial strength to projects without fully recording the liabilities themselves [2].

Golem’s September 21 coverage of the same financing trend added further examples, reporting that companies such as Meta, Broadcom and Nvidia are using residual-value guarantees around special-purpose vehicles tied to data centers or chips [3]. It also reported that off-balance-sheet commitments and credit support across seven leading hyperscalers and chipmakers total more than $3.1 trillion, citing Financial Times analysis [3]. Those figures underscore why investors are demanding a fuller map of AI economics, not merely a larger headline capital-expenditure number.

What changes for Big Tech

For technology companies, the immediate burden is operational. They must identify the relevant expense components, map them to income-statement captions, ensure consistency across systems and prepare disclosures that withstand audit and investor scrutiny. Bloomberg Law quoted EY partner Chris Bolash warning companies to “start yesterday,” a concise way of saying that compliance will be data-intensive [1].

The strategic burden may be larger. Once AI expense components become more visible, companies will have less room to rely on broad narratives about future productivity. They may need to explain why depreciation is rising, why compensation is concentrated in certain functions, how AI costs flow into gross margin, and whether inference costs improve with scale. The first round of disclosures could therefore become a baseline against which future claims are judged.

This could affect how investors compare Big Tech firms. A company that owns more infrastructure may show a different cost profile from one that leases capacity. A company that trains proprietary frontier models may look different from one that integrates third-party models. A company that monetizes AI through cloud consumption may face different margin questions from one that embeds AI inside consumer products.

The new accounting battleground

The broader message is that AI is no longer only a technology race. It is a capital-allocation race, a financing race and now an accounting race. The companies that can explain their AI costs clearly may gain an advantage with investors, even if their spending remains enormous. The companies that cannot may face growing skepticism, especially if returns remain difficult to quantify.

DISE will not make AI fully explainable in a single spreadsheet. But it will force more of the cost architecture into view. For a sector where hundreds of billions of dollars are being committed through data centers, chips, payroll, depreciation and financing structures, that is a meaningful shift. The age of “trust us, the AI payoff is coming” is giving way to a more demanding question: show us where the costs are, how they behave, and when they become returns.

Sources from the last 72 hours

1. [1]Big Tech Braces for Accounting Rules Set to Demystify AI CostsSep 21, 2026, 8:45 AM UTC
2. [2]Big techs usam outras empresas para deixar US$ 300 bi fora dos balançosSep 21, 2026, 2:00 AM UTC
3. [3]Investitionen in KI: Tech-Konzerne lagern immer mehr Schulden ausSep 21, 2026, 7:59 AM UTC

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.