Hook
Bloomberg ran a warning this week that should have hit every trading desk like a cold front off the Atlantic. The gist: circular funding is inflating the AI boom. Vendors invest in customers. Customers spend the money back on the vendor. Revenue gets recognized. Sentiment compounds. The note reaches for 2008.
Here is what nobody on my feed is saying. This pattern is not new to crypto. It is crypto's native operating system.
I run market surveillance on a 7x24 rotation out of Lisbon. My screens are stablecoin mints, treasury wallets, bridge outflows, sequencer fee lines. The shape of the flow Bloomberg describes โ capital that leaves a balance sheet and comes back wearing a customer's face โ is the exact geometry I watch every day in token ecosystems. Pulse on the chain, breath in the market. The AI story is not a warning about someone else. It is a mirror held up to us.
Context
Circular funding, stripped of euphemism, is a closed loop. Company A invests in Company B. Company B commits to buy from Company A. Company A books the purchase as revenue. If the loop is large enough and disclosed poorly, the demand is partly self-generated. Vendor financing built Lucent and Nortel in the late 1990s, and it detonated when the telco capex cycle turned. Enron hid loops inside special-purpose entities. 2008 hid them inside mortgage securitization.
In AI the loop is reputedly this: hyperscalers fund model labs, model labs commit compute spend, chip vendors invest in cloud capacity, cloud buys more chips. The Bloomberg note flags two things โ distorted market signals and financial instability โ and stops short of naming parties or numbers. That silence is the story's weakness and its power. A warning without a size is a mood, not a metric. I have no timestamp on it either, which means I cannot tell whether this is a warning before the correction or a retelling after it. That gap matters enormously.
Now translate. In crypto, the loop is not a scandal. It is the default architecture. A foundation mints a treasury. The treasury funds grants. The grantees build, then buy the foundation's token for gas, governance, or incentives. An exchange incubates a chain, lists its token, and routes volume back through its own venue. A Layer2 launches an incentive program, pays users in a token it printed, and books the resulting activity as adoption. Nobody calls it circular funding. We call it ecosystem growth.
I covered the 2017 ICO sprint as a junior researcher, filing a 1,200-word exclusive 45 minutes after a token sale announcement. I learned speed. I also learned that in a bull market, nobody audits the loop while the price is up. That lesson cost me later. It is the lesson Bloomberg is now charging AI for.
Core
Let me give you the surveillance-grade version. Based on my audit experience across four cycles, circular funding shows up in five recognizable signatures. AI exhibits at least three. Crypto exhibits all five, and has for years.
Signature one: related-party revenue concentration. When a company's growth comes from entities it also funds, the top-customer disclosure becomes the single most important line in the filing. In AI, watch for model labs whose largest customers are their own investors. In crypto, watch for chains whose largest fee payer is a foundation-controlled market maker. I pulled this thread during the 2021 NFT mania, tracking whale wallets for Bored Ape accumulation. The wallets were real. The question I never asked fast enough was whether the wallets were independent.
Signature two: capital commitments that outrun external revenue. A loop needs a forward commitment to work โ a multi-year compute contract, a token lockup, a minimum purchase. The tell is a widening gap between announced capex and paid external demand. In AI, that is hyperscaler capex versus non-affiliated revenue. In crypto, it is emissions versus organic fees. I lived the ugly version in 2022, when I downplayed Celsius's liquidity problem because the room wanted optimism. The gap was in the data. I was not reading it.
Signature three: receivables that age. If a loop partner pays with its own future, receivables days stretch. In crypto this becomes vendor financing dressed as a strategic partnership โ a market maker fronts liquidity, the project pays in tokens, the token is marked at the last print. Mark-to-own-token is the crypto equivalent of marking receivables at par.
Signature four: treasury reflexivity. A treasury denominated in the asset it is meant to back is a loop wearing a balance sheet. This is crypto's deepest structural flaw, and the AI world is now building the equity-market version: corporate treasuries and funds whose value depends on the price their own spending supports.
Signature five: sequencer and infrastructure capture. Here is where my two hard opinions collide with the AI narrative. Layer2 sequencers are, for all the decentralized-sequencing decks I have sat through for two years, basically single operators. When a chain pays incentives in its own token and the sequencer operator is also the largest treasury holder, the revenue is circular by construction. And Bitcoin after the fourth halving: miner revenue collapsed, hash power is concentrating toward three pools. A concentrated mining base plus a concentrated capital loop equals a market where the marginal buyer and the marginal seller can be the same balance sheet. Caught in the flash, framed in fact.
The accounting layer matters more than the narrative. Under standard revenue rules, gross versus net presentation decides whether a loop looks like growth or like pass-through. If Company A resells Company B's product and books it gross, revenue inflates. If related-party status is buried in a footnote, the market sees demand that is actually self-dealing. Crypto has no GAAP, which means the loop is not hidden. It is simply unlabeled. That is worse. A hidden loop can be found. An unlabeled one is normalized.
How do I actually detect a loop? Wallet clustering first. I trace funding graphs โ who seeded the wallet, who pays gas, who receives on exit. Circular flows leave a fingerprint: the same cluster on both sides of a transaction. A stablecoin mint-and-burn inside a 24-hour window within one cluster is a classic loop marker. Then I look at timing. Does organic volume spike within minutes of a treasury transfer? That is the AI loop in miniature โ capital out, activity back, signal manufactured.
In 2024 I shifted toward institutional flow analysis after the ETF approvals, modeling capital movement with traditional market metrics. The discipline transferred. BlackRock's entry, the basis trade, creation-redemption mechanics โ all of it is a closed-loop machine by design, and that is fine, because it is disclosed. The problem is never the loop. The problem is the loop you cannot see. Running where the liquidity flows fastest means knowing which liquidity is real.
The gap to watch is a single number: capital expenditure growth minus non-affiliated revenue growth. In AI, that spread is the bubble gauge. In crypto, swap capex for emissions and you have the same gauge. When the spread widens while price rises, the loop is doing the work, not the demand.
Contrarian
Everyone is reaching for 2008. I think that is lazy, and it is the part of the Bloomberg note I would push back on hardest.
2008 was a credit event. The mechanism was leverage, maturity mismatch, and collateral that was mispriced โ mortgages inside CDOs funded with short-term repo. The trigger was a funding run. What Bloomberg describes in AI is largely an equity-and-commitment phenomenon: strategic investment, purchase contracts, revenue recognition. Different plumbing. Different failure mode. An equity loop does not run. It re-rates. It bleeds through valuation cuts and write-downs, not a liquidity freeze.
So the 2008 analogy gets the fear right and the mechanism wrong. The real asymmetry is not leverage. It is disclosure. A loop with transparent related-party accounting is a strategic bet. The same loop in a footnote is a fraud waiting for a downturn. The variable that matters is not how big the loop is โ it is who is allowed to see it.
Here is my contrarian read. Crypto should not feel smug that this is an AI problem. Crypto is the loop's home field. Foundation treasuries, exchange ecosystem tokens, L2 incentive programs, restaking points, delegated governance โ every one is a circular funding structure, and most are undisclosed by design. I have watched delegation turn governance more centralized, not less, because users do not research; they hand their votes to the loudest KOL. That is a capital loop wearing a democracy costume. If the AI loop cracks first, the contagion path runs straight through the same funds and the same market makers into our order books. We are not spectators. We are the second domino.
Takeaway
What I am watching, in order. The Bloomberg original โ parties, amounts, structures. If it names hyperscalers and labs with numbers, this becomes a real signal. If it stays atmospheric, it stays a mood. Then quarterly filings: related-party revenue share, receivables days, capex versus external revenue. Then the regulatory track โ SEC related-party scrutiny, FTC and EU looks at interlocking AI investment. Then, on my own screens, the crypto tell: when a chain's headline revenue is its own treasury buying its own blockspace, the loop has already closed.
Seventy-two hours without sleep, zero doubts โ that is the job. But the discipline of a 7x24 desk is not speed. It is knowing which flash is a signal and which is a loop. Sensing the tremor before the earthquake hits is only half the trade. The other half is refusing to call a tremor an earthquake when the data has not landed.
The AI boom may be a bubble. The loop may be real. But the next 90 days of disclosures will decide whether this is 2008 or just another bull-market story we retell at the next cycle's bottom. Watch the filings. Watch the wallets. And watch what nobody is labeling.