The editorial argues that bears and bulls have been talking past each other on AI capex for 18 months, with skeptics pointing at Nvidia's equity investments in GPU-buying customers like CoreWeave, Lambda, and Mistral while bulls pointed at shipped revenue. A formal probe — even an inconclusive one — compels a specific question about how much revenue Nvidia can recognize when its own invested capital flows back as purchase orders, converting a rhetorical argument into an auditable one.
The editorial notes that SEC investigations are investigative rather than accusatory and frequently close without public action, but stresses that this one is different because of Nvidia's outsized role in index-level earnings expectations. The $2.5 trillion compound market-cap swing across Nvidia and AI-adjacent names demonstrates that the market prices the existence of the probe, not just its eventual conclusion.
By submitting the Yahoo Finance report to Hacker News where it reached 273 points, the submitter surfaced the specific concern analysts have circled for 18 months: Nvidia invests equity in AI companies (CoreWeave, Lambda, Mistral, Perplexity, Figure, Wayve) which then spend large fractions of that capital buying Nvidia GPUs. Cash-flow statements treat these as two transactions; critics see one company funding its own demand.
Yahoo Finance, citing sources, reported that the SEC has opened an investigation into Nvidia's accounting practices. The story landed on Hacker News at a score of 273 and sent the stock into what headlines are calling a "$2.5 trillion rout" — the compound market-cap swing across Nvidia and the AI-adjacent names that trade in sympathy with it.
The reported scope, per the sources cited, centers on revenue recognition timing and the treatment of related-party transactions. Nvidia has not been formally charged with anything. SEC probes at this stage are investigative, not accusatory, and a non-trivial fraction of them close without any public action. But the mere existence of a probe is, by itself, a repricing event when the company in question underwrites roughly a third of the S&P 500's forward earnings growth story.
The specific worry that analysts have been circling for the better part of 18 months is what's politely called "round-tripping" and less politely called circular revenue. Nvidia has made equity investments in a long list of AI companies — CoreWeave, Lambda, Mistral, Perplexity, Figure, Wayve, and others — many of which then turn around and spend a large fraction of that capital on Nvidia GPUs. On a cash-flow statement these are two separate transactions. On a skeptic's whiteboard they look like one company funding its own demand.
For most of 2024 and 2025, the pushback on the AI capex story was vibes-based. Bears pointed at the circular-funding diagrams; bulls pointed at the revenue line and said "the chips are shipping, the datacenters are being built, the electricity bills are being paid." Both were right in their own frame. What a formal SEC probe does — even an inconclusive one — is force a specific, testable question: when CoreWeave buys $N of Nvidia GPUs after Nvidia invests $M in CoreWeave, how much of $N should Nvidia be allowed to recognize as revenue in the current quarter, and on what terms?
This is not a new question in accounting. The telecom bubble ran on fiber-swap round-trips. Enron ran on special-purpose entities that bought Enron's own assets. Neither of those are a prediction — they're a reminder that the pattern the market is being asked to stop worrying about has a history, and that history is specifically why the SEC pays attention to it. The honest read is that we don't yet know whether Nvidia's treatment is aggressive, standard, or genuinely clean. Analysts at Morgan Stanley and Bernstein have published notes on both sides over the past year. The probe will produce an answer the market can actually price.
The second-order effect is where this gets interesting for anyone building on top of the AI stack. Nvidia's revenue number is the capex signal that hyperscalers, neoclouds, and enterprise CIOs all use to calibrate their own spend. If that number is directionally right but overstated by, say, 10–15% because a chunk of it represents chips financed back into the ecosystem Nvidia itself seeded, then the "GPU scarcity will persist through 2027" consensus has a soft foundation. If it's clean, the probe closes and we go back to arguing about whether Blackwell yields are tracking.
The Hacker News thread, with 273 points, is instructive in its tone. The top comments are not gleeful shorts or breathless doomers — they're engineers and finance-adjacent technologists pointing out that this is the first time the circular-revenue thesis has had institutional teeth behind it, and asking the obvious follow-up: what does an AI infrastructure market without a 90% Nvidia margin on training GPUs actually look like? That question has been theoretical. It's now a scenario worth modeling.
There's also a timing dimension worth naming: the probe reportedly predates the public reporting by months, which means whatever the SEC has been looking at, it's been looking at it through at least two Nvidia earnings cycles. Each of those cycles included forward guidance that moved trillions of dollars of market cap. That's not an accusation of anything — it's just the context in which to read every "blowout quarter" headline from here forward.
If you're an engineer or an engineering leader, the short-term operational impact is close to zero. Your H100s, H200s, and whatever Blackwell allocation you managed to get this cycle will keep working. CUDA will keep being the moat it was last week. The teams you're competing with for GPU-hours aren't going to suddenly have less GPU-hours. The stack-level question isn't "will Nvidia survive" — it's "should the single-vendor dependency at the bottom of your AI budget have a plan B that's more than a slide?"
Concretely, three things get more valuable this week than they were last week. First, any real engineering work you've done on portability — ROCm support, TPU abstraction layers, llama.cpp-style inference paths that run on consumer silicon — has more optionality than it did on Friday. It was insurance before; now it's cheap insurance. Second, your relationships with the neocloud providers (CoreWeave, Lambda, Crusoe, Nebius) deserve a second look: those companies are precisely the ones at the center of the circular-revenue question, and their pricing and capacity posture could change quickly in either direction depending on how this resolves. Third, if you've been quietly deferring a conversation with finance about committed GPU spend versus on-demand, the news cycle just handed you a pretext to have it.
The thing not to do is panic-migrate. The failure mode here is a CTO reading one Yahoo Finance headline and rewriting a working production stack to "de-risk from Nvidia." Nvidia's technical dominance and the accounting question around its revenue velocity are genuinely separate problems, and conflating them is how you end up on AMD six months before AMD's software story is actually ready.
The base case is that this probe takes 12 to 24 months, produces either a no-action letter or a settled restatement of some specific line items, and the stock price moves around a lot in the interim without any of the underlying compute economics changing materially. The tail case — a finding that materially restates past revenue — is low-probability but not zero, and it's the first time in the AI cycle that "low-probability but not zero" has been the honest answer. Either way, the free pass that Nvidia's accounting has gotten from the sell side for the last three years is over. For practitioners, the work doesn't change. For planners, the risk model just got a new variable. Price it accordingly.
If the board thought there were accounting issues why did they make the CFO the interim CEO?
For those thinking how could it be possible that CFO didn't know, well companies this large, public or not, will have in-house counsel they pay for, for advice (sure, in-house counsel will farm out to third parties). If the counsel tells the C-suite they shouldn't do a specific thing, the
I guess that guy on youtube, Ned Citron was right after all.
Colette Kress the acting CEO of nVidia has been with the company since 2013 as Chief Financial Officer. It's difficult to believe the CFO was unaware of what could be years of blackmarket chip sales. It would ultimately be the financial department that has to manage that revenue on the books.If
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
This is just a continuation of the circular financing that has plagued Nvidia for the past 5 years. If the last earnings call was any guidance, their plan to renegotiate their current usage deals with neoclouds is going to severely shoot themselves in the foot.Now, if you believe the entire market h