Recommended Free Tools
Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Anthropic CEO Dario Amodei did not say that artificial intelligence is fake or that the entire sector is a bubble. His warning was narrower and more important for investors: AI can be genuinely useful while companies still lose enormous sums by committing to data centers, chips and financing arrangements faster than durable revenue develops.
At the New York Times DealBook Summit on December 3, 2025, Amodei said some unnamed AI companies were “YOLO-ing” their infrastructure spending—taking unusually large risks on the assumption that future demand and revenue will justify today’s commitments.
What Amodei’s “YOLO” warning actually means
“YOLO” means “you only live once.” In this context, Amodei used it to describe companies that may be pulling the risk dial too far by locking in massive amounts of compute capacity before they know exactly when, or whether, the associated revenue will arrive.
His argument separated two questions that are often treated as one:
#1 Best Overall
- Is the technology valuable? Amodei said he remained confident in AI’s technological potential.
- Will every company building around it become a good business? He warned that the answer is no, particularly when infrastructure spending, financing costs and depreciation outrun monetization.
Amodei did not identify the companies he meant. Some contemporary coverage interpreted the remarks as an indirect criticism of OpenAI because of the discussion around large infrastructure plans and industry financing. That remains an interpretation, not an explicit accusation by Amodei.
The central point is simple: real technology and real customers do not guarantee sensible valuations or sustainable corporate finances.
Why AI infrastructure creates unusual financial risk
AI companies need far more than software engineers. Frontier models require specialized chips, data centers, networking equipment, electricity, cooling systems and cloud capacity. Those resources create a timing problem:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Infrastructure takes time to secure. Data centers, power connections and chip orders may require commitments well before a company knows the exact level of customer demand.
- Demand is difficult to forecast. Early usage can reflect experimentation rather than long-term production workloads.
- Technology changes quickly. A newer generation of chips can make older equipment less productive or less economical even when it still works.
- Revenue may not cover incremental costs. More tokens processed can mean more sales, but also more inference, energy and support costs.
- Financing may need to continue. Companies pursuing large capacity plans may depend on repeated equity investment, debt, strategic funding, supplier credit or cloud commitments.
This creates two opposing risks. A company can overbuild and end up paying for capacity that customers do not use. It can also underbuild and lose customers because it cannot provide enough capacity. Competitive pressure encourages companies to spend defensively, even when a cautious financial plan would suggest waiting.
What are circular AI financing deals?
A circular deal can involve a chipmaker, cloud provider or infrastructure company investing in an AI company, after which the AI company uses some of the money to buy chips, cloud services or related capacity from that investor or its partners.
Rank #2
For example, imagine a supplier invests $10 billion in an AI company, and the AI company spends a substantial portion of that capital on the supplier’s hardware or services. That arrangement could be economically rational: the supplier helps a customer obtain scarce capacity, while the AI company receives funding and access to infrastructure.
But the transaction does not automatically prove that the AI company has generated $10 billion of independent end-user demand. The key question is who ultimately bears the risk if customers do not pay enough to support the commitments.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsAmodei defended circular arrangements in principle. His concern was that several such deals could be stacked together, creating commitments that depend on extremely optimistic forecasts of future revenue. Circular financing is therefore not automatically fraudulent or unsound. The risk is that headline funding, demand and valuation figures may look stronger than the underlying cash flow generated by unrelated customers.
Is Anthropic really different from its rivals?
Amodei portrayed Anthropic as more conservative in its planning. He said the company uses cautious assumptions about future revenue and chip economics rather than building its plans around the most optimistic possible forecast.
That is Anthropic’s claimed distinction, not an independently audited conclusion. Anthropic still needs substantial amounts of compute, cloud access, capital and infrastructure. Contemporary coverage also reported a planned $50 billion data-center investment, although the precise scope and accounting treatment of that figure matter.
The useful comparison is not “Anthropic spends little” versus “competitors spend a lot.” It is:
- Conservative planning: commit capacity against a range of possible demand and retain room for weaker outcomes.
- Aggressive planning: make very large commitments based on the upper end of uncertain revenue forecasts.
Because Anthropic was private at the time of Amodei’s warning, outside investors did not have the same level of audited disclosure available from a listed company. That makes claims about its profitability, cash burn and capital intensity difficult to verify fully.
What Anthropic’s revenue growth proves—and does not prove
According to figures reported from Amodei’s remarks by TechCrunch, Anthropic’s revenue grew from approximately $100 million in 2023 to approximately $1 billion in 2024. Amodei also cited an $8 billion to $10 billion year-end 2025 run rate.
Those figures indicate strong adoption, but a run rate is not the same as recognized annual revenue. A run rate takes recent revenue and projects it across a full year. It does not prove that:
- customers will renew or maintain the same usage;
- gross margins are healthy;
- inference costs will decline quickly enough;
- research and sales expenses are under control;
- capital spending is covered by operating cash flow; or
- the valuation is justified by future free cash flow.
A company can grow revenue tenfold and still have poor unit economics if every additional dollar of sales requires nearly a dollar of compute and support costs. Investors should distinguish revenue growth from profitable growth and both from cash generation.
Why chip economics matter
The risk is not necessarily that AI chips stop functioning. It is that newer chips become faster, cheaper or more energy-efficient, reducing the economic value of older hardware before it has been fully depreciated.
An AI company that buys too much capacity too early may therefore face:
- lower productivity from its existing hardware;
- higher depreciation expense;
- more expensive power and cooling per unit of useful work;
- pressure to discount prices to keep capacity utilized; and
- additional spending to upgrade before the original investment has paid back.
This is why “AI demand is real” does not settle the bubble debate. Demand can be real while the industry still overpays for the infrastructure used to serve it.
Does real demand rule out an AI bubble?
No. “Bubble” can describe several different risks:
| Type of bubble | What it means |
|---|---|
| Technology | The underlying capabilities are overhyped or fail to deliver useful results. |
| Valuation | Investors price companies far above what their future cash flows can justify. |
| Infrastructure | The industry builds more data-center, chip, power or networking capacity than customers need. |
| Financing | Expansion depends on repeated fundraising, supplier credit or circular transactions. |
| Revenue quality | Usage and annualized revenue grow quickly but prove temporary or insufficiently profitable. |
A sector can have valuable products, genuine customers and rapidly increasing revenue while still containing companies whose spending plans or valuations are unsustainable. The question is not simply whether AI works. It is whether each company can convert demand into durable margins and cash flow.
Best Value
Later valuation and IPO context
Later developments should not be confused with information available at the December 2025 summit. A May 2026 report said Anthropic raised $65 billion at a reported $965 billion private valuation. That is a private-market valuation following a funding round, not a public-market capitalization or proof of profitability.
In June 2026, reports said Anthropic had confidentially filed for an IPO. As described in Fortune’s coverage, the offering’s share count, price, size and timing were not yet public in the cited reporting.
A public filing could provide a much clearer test of the “YOLO” warning by disclosing recognized revenue, gross margins, operating losses, capital expenditures, debt, purchase commitments, customer concentration, stock-based compensation and related-party transactions. Until those details are available, private valuation headlines should be treated as signals of investor expectations—not as audited evidence that the economics are sound.
How investors can test the warning
Investors evaluating Anthropic, OpenAI or AI infrastructure companies should look beyond funding rounds and headline revenue estimates.
Evidence that would support Amodei’s concern
- Revenue growth slows while fixed infrastructure commitments remain large.
- Gross margins fail to improve as usage expands.
- Customers reduce usage after experimentation or fail to renew enterprise contracts.
- Companies repeatedly require emergency fundraising or new strategic financing.
- Supplier or investor financing absorbs losses that customers cannot support.
- New hardware makes existing capacity uneconomic.
- Public filings reveal large debt, leases, purchase obligations or weak cash conversion.
- Data-center utilization falls or infrastructure providers report cancellations.
Evidence against the strongest bubble theory
- Enterprise customers renew and expand production deployments.
- Inference costs fall faster than prices.
- Utilization remains high across data centers.
- Companies increasingly fund expansion from operating cash flow.
- Hardware retains economic value for longer than skeptics expect.
- Customers can demonstrate measurable savings or new revenue from AI.
What this means for personal investors
For individual investors, the practical lesson is to avoid treating an AI company’s funding total, annualized revenue or private valuation as a substitute for financial statements.
When assessing an AI-related investment, ask:
- Is the reported figure recognized revenue, contracted revenue or merely a run rate?
- What are the company’s gross margins after inference and hosting costs?
- How much future capacity has it committed to buy or lease?
- Does it depend on a small number of strategic customers or investors?
- Are supplier investments creating genuine independent demand or mainly recycling capital within the ecosystem?
- How quickly could new chips make existing equipment less competitive?
- Is the valuation based on revenue multiples, expected profits or actual free cash flow?
- Are audited public disclosures available?
For enterprise customers, the same warning argues for portability: avoid making a critical application dependent on one model provider without fallback models, exportable data, clear service terms, cost controls and a plan for pricing or capacity changes.
Bottom line
Amodei’s “YOLO” warning is not a claim that AI has no value. It is a warning that companies can build valuable products and still become financially fragile if infrastructure commitments, chip depreciation and financing obligations grow faster than durable monetization.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →The decisive evidence will be recurring customer revenue, improving unit economics, resilient utilization, cash generation and transparent financial disclosures—not headline funding rounds or private valuation marks.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
