The AI boom may be built on shakier ground than most investors realize. Arthur Hayes, co-founder of BitMEX, laid out a pointed case in a June 26, 2026 interview with Bonnie Blockchain: the AI bubble risks a sharp correction, driven by three forces that are already quietly gathering momentum — rising energy costs, government policy unpredictability, and the accelerating rise of open-source alternatives.
Summary
Key takeaways
- Arthur Hayes warns the AI market faces a potential correction from three interconnected threats.
- Geopolitical tensions could push oil prices significantly higher within four to six months, raising AI computation costs.
- US government restrictions on Anthropic’s AI models show how policy can abruptly cut off user access.
- Chinese open-source AI models offer similar performance at roughly one-tenth the cost of US competitors.
- The shift toward open-source AI directly threatens the pricing power and margins of major US AI firms.
Three Threats the AI Market Hasn’t Fully Priced In
Most AI valuations are built on a clean set of assumptions: energy stays manageable, access remains open, and no cheap rival undermines premium pricing. Hayes argues all three of those assumptions are under pressure simultaneously — and that markets haven’t caught up yet.
What makes his analysis worth paying attention to is the interconnected nature of the risks. None of these factors operates in isolation. A geopolitical shock that drives up oil prices compounds operational costs at the same moment that policy restrictions are rattling enterprise buyers, while a wave of cheaper open-source models quietly erodes the premium that US AI firms have been charging. Together, they form a stress scenario that the current froth in AI valuations may not be equipped to absorb.
Energy Costs Could Pressure AI Computation and Profitability
AI computation is, at its core, an energy business. The data centers powering large-scale models consume enormous amounts of electricity, and that electricity is still heavily tied to fossil fuel pricing.
Impact of Rising Oil Prices
Hayes pointed to geopolitical tensions — including a possible US-Iran conflict — as a credible catalyst for a significant oil price spike. His estimate: prices could move markedly higher within four to six months. That kind of move would ripple directly into the operating costs of every company running large AI infrastructure.
Higher energy costs wouldn’t just sting margins. They would expose a structural weakness in how AI investments are underwritten. Many growth projections for AI firms assume relatively stable input costs. A sustained rise in energy expenses would force a reassessment of those profitability models — and that reassessment could be abrupt.
Why This Matters for Investors
According to Hayes, many portfolios built around AI growth assumptions may not have accounted for this variable at all. That’s not a minor oversight. If energy costs climb sharply enough, the entire financial architecture of certain AI businesses starts to look different.
Government Policies Create Uncertainty for AI Users
Regulatory risk in AI isn’t abstract — it has already materialized in concrete ways. Hayes cited US government restrictions on Anthropic’s Mythos and Fable models as a real-world example of how policy decisions can disrupt access without warning, affecting both external users and, at times, people inside the company itself.
US Restrictions on Anthropic’s AI Models
The restrictions on Anthropic’s models illustrate a broader dynamic: in a politically charged environment around AI, access to critical tools can be switched off at the government’s discretion. For domestic users, that’s a compliance headache. For international businesses and individuals who rely on US AI services, it’s a more serious operational risk.
Separately, the broader industry debate around open-weight models has intensified. According to Business Insider, over two dozen companies — including Nvidia, Meta, Microsoft, OpenAI, and Google — signed an open letter called “Open Weights and American AI Leadership,” endorsing open-weight AI development. Anthropic was notably absent from that letter, drawing criticism from figures including venture capitalist David Sacks and Benchmark general partner Bill Gurley, who suggested the company was protecting its own economic model by staying silent.
Risks for Non-US Companies and Users
For businesses outside the United States that have built operations around continuous access to US-based AI services, this is more than a political sideshow. A sudden policy shift could cut off service entirely, regardless of payment status or contractual agreements. Hayes argued this vulnerability is not fully reflected in current AI valuations — and that it could reshape how companies choose their AI providers going forward.
Open-Source AI Models Challenge US AI Firms’ Pricing and Control
The third threat Hayes outlined may be the most structurally disruptive. Open-source AI models — many developed by Chinese teams — are delivering performance that rivals closed US systems, at a fraction of the price.
Chinese Open-Source Models Offer Cost and Performance Advantages
Hayes said these models often cost roughly one-tenth of comparable US offerings while delivering similar performance. That price gap alone is enough to attract cost-sensitive enterprise buyers. Business Insider also reported on the recent launch of Kimi K3, an open-weight model built by Chinese AI lab Moonshot, which rivals top US closed models on several benchmarks — adding concrete weight to what Hayes described.
The competitive pressure is no longer theoretical. With models like Kimi K3 drawing serious attention across the industry, the question of whether US firms can maintain their pricing premium is becoming harder to dismiss.
Benefits of User Data Control and Reduced Dependency
Beyond cost, open-source models offer something that closed systems fundamentally cannot: user control over data. For companies wary of running sensitive operations through infrastructure tied to another government’s regulatory decisions, the appeal of self-hosted alternatives is growing fast.
This is exactly the kind of dynamic that erodes moats. US AI firms have justified high valuations partly on the basis of high margins. If a significant portion of the market migrates toward cheaper, self-hosted alternatives — driven by a combination of cost savings and data sovereignty concerns — those margins face sustained pressure. Hayes concluded that this shift directly threatens the pricing power of US AI companies, and the industry’s broader response to open-weight models suggests the threat is being taken seriously across Silicon Valley.
The deeper implication running through all three risks is that the AI market may be pricing in a best-case scenario on multiple fronts simultaneously. When energy stays cheap, governments stay permissive, and no affordable alternative emerges, premium AI valuations hold up. Strip away any one of those assumptions, and the math gets harder. Strip away all three at once, and the correction Hayes is warning about starts to look less like speculation and more like arithmetic.
FAQ
What are the main risks threatening the AI bubble according to Arthur Hayes?
Arthur Hayes highlights three main threats: rising energy costs driven by geopolitical tensions, restrictive government policies that can abruptly limit access to AI models, and growing competition from open-source AI alternatives — particularly from Chinese developers.
How could rising oil prices affect the AI industry?
Rising oil prices, potentially driven by geopolitical tensions such as a US-Iran conflict, could increase the energy costs of running AI data centers within four to six months. This would pressure the profitability of AI companies whose business models assume stable operating costs.
Why do government policies create uncertainty for AI users?
US government restrictions — such as those applied to Anthropic’s Mythos and Fable AI models — can abruptly cut off access for users, especially non-US businesses that depend on these services. This creates operational and service continuity risks that Hayes argues are not fully priced into current AI market valuations.
What impact do open-source AI models have on US AI firms?
Open-source AI models, particularly those from Chinese developers, offer similar performance to closed US systems at roughly one-tenth of the cost. They also give users greater control over their own data, reducing dependency on US platforms. Together, these factors threaten the pricing power and margins that underpin current AI company valuations.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.
