BlackRock’s iShares Future AI & Tech ETF has delivered a year-to-date return of 31.39%. The ETF holds exposure to 71 companies riding the AI boom: chip designers, foundries, memory suppliers, grid equipment makers like Eaton, and Big Tech hyperscalers through their cloud platforms.
In effect, ARTY captures the entire AI “hype stack”. As such, it mirrors the broader commitment to erect an algorithmic layer atop nearly all digital human interaction. We have already examined what this entails for the purpose of governance, with Palantir at the center.
However, are these unprecedented AI commitments sustainable, heading for an inevitable bubble burst in 2026, or are they just starting to ramp up?
The Scale of AI Investing Examined
December’s Dealogic report shows that the global tech sector issued $428.3 billion in bonds in 2025, meaning companies borrowed that much from investors in the form of debt securities. Expectedly, the US is the primary driver of this debt growth, accounting for $341.8 billion.
The bulk of that commitment is allocated for AI capital expenditures (capex). Case in point, the 5-year credit default swap (CDS) spreads for both and have nearly doubled since September.
According to the latest Crunchbase data, the AI sector received a total of $202.3 billion in capital during 2025, representing a 75% year-over-year increase from $114 billion in 2024. In Q3 this year, hyperscalers dedicated $106 billion to both AI and non-AI capex.
Yet, this may be only the modest beginning. The latest Goldman Sachs Research reports that capex estimates for 2026 have been revised higher, rising from $465 billion to $527 billion. The bank estimates that AI capex now accounts for 0.8% of GDP, still nearly half the 1.5% of GDP reached during comparable tech booms over the past 150 years.
Defining the AI Bubble Boundaries
In late January this year, we saw the first inkling of the AI hype contraction, as $600 billion was wiped from ’s market cap in a single day, setting a new loss record for publicly traded companies. Chinese DeepSeek triggered this sell-off, as the LLM matched the performance of top-tier US models but at a fraction of the training cost.
The implication is that massive investments in both power generation and data centers are not actually needed as previously thought. However, we maintained that DeepSeek’s efficiency tweaking doesn’t matter due to the extraordinary demand for compute resources beyond just text-to-text output into text-to-image, text-to-audio, and the most demanding of all – text-to-video output.
In other words, if there is to be an AI future, it is to be multi-modal, and we are still in its early growth. Indeed, the market contraction was short-lived, as Big Tech doubled down on AI capex.
It is also not surprising to see this development given the Jevons effect, named after the English economist William Stanley Jevons. Specifically, as AI models, chips, and data center infrastructure become more efficient, the cost per unit of computation falls. But rather than reducing total compute demand, such efficiency expands the addressable use cases.
Therefore, a more efficient AI would not restrain energy use and capital expenditures but multiply them. However, the Jevons effect applies only to AI that actually delivers on reasoning and productivity gains while eliminating confabulation.
This is the boundary of the AI bubble. If AI stalls with narrow incremental gains that fail to translate into measurable productivity, efficiency will stop compounding demand, breaking the Jevons effect in the process as overbuilt capacity chases diminishing marginal value.
Are Circular Financing Concerns Valid?
In addition to relying on academic breakthroughs and tweaks, the AI ecosystem is facing systemic risk in the form of circular financing. Case in point, when Larry Ellison of Oracle pushes for Stargate and a $300 billion investment in data centers for OpenAI, Nvidia is the primary supplier.
At the same time, Nvidia invests $100 billion in OpenAI, in addition to owning a stake in CoreWeave, which has a cloud deal with Oracle. CoreWeave is also Microsoft’s major customer – another voracious buyer of Nvidia chips – that has multi-billion dollar deals with OpenAI and .
This constant recycling of capital within a closed circle breeds tight interdependence and inflates headline growth. Put simply, the AI buildup is so unprecedented and capital-intensive that it cannot be funded solely by organic cash flow. That is to say, the entire system is leveraged on itself with no-exit optionality.
At the same time, AI remains an ahistorical phenomenon, driven by the extreme cost of innovation. And that cost is based on real monetization and demand, as we’ve recently explored with the Microsoft vs Alphabet roundup.
The Bottom Line
Calling AI a bubble has become so routine it now functions as a preemptive disclaimer. It operates less as analysis and more as a reflexive “don’t get fooled again” slogan. In late 2024, we leaned heavily against AI being a bubble due to the concerted nature of efforts involved, representing a fusion between corporate and political governance.
As the holy grail of governance, AI represents the ultimate instrument of control, one that scales the will of a select few to unprecedented, enormous levels. And AI can do this at a granular, intimate level of interaction. President Trump has demonstrated, in no uncertain terms, his dedication to serving the select few in this endeavor, just one year into his presidency.
Therefore, nearly all indicators point to 2026 being another AI boom year, benefiting both AI and AI-adjacent stocks.
***
Looking to start your trading day ahead of the curve?
Get up to speed before the bell with Bull Whisper—a sharp, daily premarket newsletter packed with key news, market-moving updates, and actionable insights for traders.
Start your day with an edge. Subscribe to Bull Whisper using this link.