The most unsettling part of a proposed $500 billion AI data-center fund isn’t that investor Michael Burry says it has “shades of Enron.” It is that, as Blake Oliver explains, nobody appears to be breaking a rule.
“There’s no fraud happening here,” Blake says in Episode 501 of The Accounting Podcast. “This is all happening in plain sight.”
That tension runs through Blake and David Leary’s discussion. AI makes financing structures more complex while helping firms complete audits and other accounting work faster. Yet many of the standards governing that work were written for a different era.
Nvidia’s financing shows how risk can grow within the rules
The proposal Burry criticized brings together Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to finance AI data centers. The plan is to use special-purpose vehicles to own the centers, buy Nvidia chips, and lease computing power to companies such as OpenAI and Anthropic. The debt would be backed by the computing assets, with Nvidia reportedly guaranteeing about 25%.
Why would Nvidia support separate entities instead of building the centers itself? Blake explains that selling chips to those entities would let Nvidia record revenue upfront. If Nvidia built and operated the centers, it would absorb the construction costs and recognize revenue later when it sold computing services.
The proposal adds another layer to the circular financing problem discussed in Episode 488. Money moves in a circle, and both sides report revenue.
Burry called the proposal an effort to use “unnatural credits to prolong momentum late in the bull phase.” “Maybe the problem is that GAAP allows this,” Blake notes.
Depreciation adds to his concern. AI chips are only useful for two or three years, but some companies use estimated lives of five or six years. Longer useful lives mean less annual depreciation and higher reported profit. Because large technology companies carry heavy weight in the S&P 500, a sharp correction could hurt ordinary investors holding index funds.
That same divide between reported results and underlying quality appears in audit.
Faster audits don’t automatically mean better audits
EY says AI improved audit speed or throughput by roughly 125% to 150%, while clients haven’t demanded lower fees. The firm also reported a 5% PCAOB deficiency rate, down from 28% the prior year, and pointed to its billion-dollar investment in people and technology.
David is skeptical that technology alone explains the improvement. Other large firms also posted better inspection results. He suggests the PCAOB’s changing focus on firmwide quality-control systems may affect the numbers.
Blake offers another theory. PCAOB inspections often focus on whether auditors followed required procedures, obtained approvals, and completed documentation. AI is well suited to checking those boxes. But it can also create work that looks “solid and sophisticated” while still being wrong. Complete documentation isn’t the same as sound professional judgment.
The productivity gains could still disrupt the market. Big Four firms may keep the savings as higher margins, but Blake argues that regional and smaller firms could eventually use the same tools to provide comparable services at lower prices.
Before that can happen safely, however, audit rules must catch up.
Audit standards weren’t built for AI agents
In a Gartner poll of 743 audit professionals, 93% reported using AI in some form. Yet only 30% used it for audit testing, 12% used it for quality reviews, and 38% of audit leaders had an AI strategy.
Hofstra University accounting professor and CPA Jack Castonguay argues that AI is audit’s biggest disruption since the corporate failures that led to the PCAOB’s creation. He says applying existing standards to a “fundamentally new operating model” won’t be enough.
The unanswered questions include:
- Evidence reliability. What happens if AI invents evidence or changes data it believes is wrong?
- Agent supervision. Who is responsible when auditors fail to review AI agents that gather and analyze evidence?
- Independence. Could an AI-enabled accounting system and an audit platform trained on the same data reinforce the same errors?
AI can test every transaction instead of a sample. That is a major advance, but current standards don’t explain how much human review is needed when a machine examines the full population. Castonguay wants standards for acceptable use, oversight, evidence, supervision, and independence.
The mismatch is also visible in financial reporting.
Reporting and assurance are moving on different clocks
The SEC’s proposal to move public companies from quarterly to semiannual reporting drew about 225,000 comments. By comparison, the PCAOB received only 33 comments on its request for input about future priorities, including AI-related research.
David questions whether two reports or four reports is even the right debate. If automation leads to a continuous close, he asks, “Shouldn’t the discussion be moving to daily?”
Tether presents a related problem. Assurance has limited value if users can’t inspect it. KPMG US issued an unqualified 2025 audit opinion for the stablecoin issuer, but the report hadn’t been published at the time of the discussion. As David asks, “If they don’t publish the reports, did they really do it?”
While regulators debate these issues, small firms are already putting AI to work.
Small firms can gain leverage without removing human review
The hosts highlighted four firms with fewer than ten employees. One Stop CPA uses Blue J for source-backed tax research, applies professional judgment, and then uses ChatGPT Enterprise to create memos and presentations. Agate CPA built an automated client intake process that increased conversions by about 25%. Public Trust CPA created a nonprofit invoice-approval trail using Power Automate, Adobe Sign, and QuickBooks. High Rock Accounting built a client-feedback app in a few hours and now holds AI happy hours to identify repetitive work.
These examples show that small firms don’t have to wait for enterprise software. But client expectations are rising, and review is costly. Blake’s conclusion about QuickBooks Live applies across the profession: AI can do the work, “but it still needs a human to review it.”
AI exposes weak points in accounting’s rulebook while giving firms new ways to research, automate, and compete. The winners will be firms that define acceptable uses, review responsibilities, and evidence standards before regulators catch up.
Listen to Episode 501 of The Accounting Podcast for Blake and David’s full discussion.