The AI Gold rush: Boom, bubble or just the beginning?

Artificial Intelligence (AI) has moved from buzzword to backbone of markets. Innovation is accelerating, capital spending is surging, and investors worldwide are paying attention. The excitement has echoes of a modern-day gold rush – but beneath the hype lie real, long-term shifts in how companies create value and compete.

Last year, we were strongly optimistic about the AI opportunity. After another year of remarkable performance, we revisited our view. Our conclusion? We continue to see powerful reasons to stay constructive — but several risks are becoming increasingly important to monitor.

The case for the boom

AI adoption is only just getting started. It is already automating tasks, enhancing customer experience and enabling new business models, yet as of June 2025, only 41% of U.S. companies report adopting AI into their day-to-day business[1]. The potential for wider deployment across sectors remains significant.

Investment trends underline this. Hyperscalers such as Alphabet, Amazon and Meta continue to pour capital into chips, datacentres and cloud infrastructure, while governments step up spending to stay relevant. This looks less like a passing fad and more like a long-term strategic race.

For companies, the gap between early adopters and laggards is widening. Firms that embed AI into their processes are seeing productivity gains, better client engagement and an edge in attracting talent. Those without a clear AI roadmap risk ceding market share as the technology scales.

Crucially, AI is a recurring, not a one-off, revenue opportunity. Every query, model update and agent action consumes compute, supporting durable demand for infrastructure and cloud services. Yet, despite strong equity performance, valuations in the AI ecosystem are far more grounded - and balance sheets are significantly healthier – when compared to the dot-com bubble or other bubbles in financial history.

• Magnificent 7: Nvidia, Microsoft, Apple, Alphabet, Amazon, Meta Platforms, Tesla​
• Technology bubble leaders: Microsoft, Cisco Systems, Intel, Oracle, IBM, Lucent, Nortel Networks​
• Japan financial bubble: Nippon Telegraph and Telephone, Industrial Bank of Japan, Sumitomo Mitsui Banking, Bank of Tokyo-Mitsubishi, Fuji Bank, Dai-Ichi Kangyo Bank, Sakura Bank​
• Nifty 50: IBM, Eastman Kodak, Sears Roebuck, General Electric, Xerox, 3M, Proctor & Gamble

The case for the bubble

One growing concern is that power constraints could slow the expansion of AI infrastructure. Datacentres already consume vast amounts of electricity, and U.S. facilities alone could account for 12% of national power demand by 2030[2]. Yet investment in new generation capacity has lagged for years. If supply cannot scale quickly enough, some planned datacentre projects may be delayed or downsized — with knock-on effects across the entire AI value chain.

Interdependence within the AI ecosystem is also increasing. Companies are becoming tied together through shared investments, debt guarantees and deep strategic partnerships. While this accelerates innovation, it also creates new points of fragility: if a major player stumbles, the shock could ripple across suppliers, customers and investors.

The financing behind the AI boom also requires close attention. The scale of capital expenditure is unprecedented. OpenAI, for instance, reportedly has more than $1.4trn in commitments[3], compared with expected 2025 earnings of just $13bn[4]. Hyperscalers’ annual capex has surged from roughly $125bn in 2022 to an estimated $398bn in 2025 and could reach $400–600bn per year in the coming years[5]. Yet internal cash flows can cover only around half of the projected $2.9tr needed for datacentre buildouts through 2028[6], meaning the rest will likely depend on debt markets — introducing additional leverage into an already interconnected system.

The AI boom is not only an equity story — it is becoming a credit market story too

Financing the AI build-out: the growing role of credit markets

The acceleration of AI investment is transforming not just industries but global credit markets. Capital expenditure needs now exceed what many technology firms can fund through cash flow, pushing the AI ecosystem to rely more heavily on debt.

Investment Grade: jumbo transactions and a changing landscape

U.S. hyperscalers have issued a surge of record-sized bonds to support expanding infrastructure. While strong balance sheets have anchored demand, the environment is shifting; higher coupons, softer secondary-market performance and widening spreads reflect investor concerns over rising leverage and multi-year spending plans. With issuance already exceeding $120 billion[7].

This trend is unlikely to fade. With AI infrastructure requirements still accelerating, the sector is poised to become a much larger component of IG indices. Over the coming year alone, additional issuance could exceed $150 billion[8] (or +60% YoY), driven primarily by the Big Five hyperscalers. While long-term fundamentals remain solid, we believe a slightly underweight and selectively positioned stance is prudent - particularly at the long end of the curve.

High Yield: The Rise of “Neoclouds”

A transformation is also underway in the high-yield market. A new group of AI-focused infrastructure providers — often called “neoclouds,” including CoreWeave, TeraWulf and Cipher Mining — is scaling quickly to deliver GPU-dense compute capacity. Many are former crypto miners pivoting to AI infrastructure, leveraging the shared GPU-intensive foundations of mining and model training. Neoclouds not only deploy hardware at speed, but are also purpose-built for AI workloads, as opposed to general-purpose computing. This often leads to lower latency and more competitive economics for training and inference.  

However, this agility comes with elevated financial risk. Neoclouds rely heavily on debt and asset-backed structures to fund their expansion. In some cases, leverage is rising extremely quickly: CoreWeave’s debt load is projected to jump from around $8 billion[9] to nearly $40 billion by 2027. While medium-term customer contracts offer some visibility, the model depends on uninterrupted access to financing and the assumption that customers will renew agreements or maintain demand.

Earlier in the year, this segment attracted strong investor appetite, with deals from CoreWeave, TeraWulf, Cipher Mining and SoftBank drawing heavy demand. Although technology represents only about 7% of the HY index and has room to grow (but it’s unlikely to mirror Telecom’s surge from 5% to 21% in the late 1990s), issuance has been concentrated in the U.S., with Europe seeing limited activity. But sentiment has become more selective. Applied Digital priced meaningfully above expectations, SoftBank’s credit protection costs have risen to multi-year highs (ex. Liberation Day), and CoreWeave’s yields have widened as markets reassess the sustainability of its business model. By contrast, issuers with strong strategic backing — including those supported by large technology partners — have proven more resilient, highlighting growing differentiation across credit quality. In high yield, opportunities remain but they require disciplined analysis and selective positioning.

Together, these trends point to a maturing phase in the AI credit cycle, where abundant early liquidity is giving way to a more discriminating assessment of balance sheet strength, customer concentration and the durability of capital spending plans.

A transformative boom — with manageable risks

We continue to believe the AI revolution is real and structural. The technology is still in the early phase of adoption, investment remains robust, and the economic potential is immense. The whole AI ecosystem now extends far beyond Nvidia, and a broader range of tech companies are positioned to benefit from this transformation.

But the path forward will not be linear. Power constraints, ecosystem fragility, and the financing required for large-scale infrastructure buildouts all deserve close attention.

At this stage, however, we see AI not as a speculative bubble, but as a long-term opportunity — one that still has significant room to grow.

[1] [1] Corporate AI adoption may be leveling off, according to Ramp data | TechCrunch
[2] Charted: The Energy Demand of U.S. Data Centers (2023-2030P)
[3] Sam Altman says OpenAI has $20B ARR and about $1.4 trillion in data center commitments | TechCrunch
[4] OpenAI has five years to turn $13 billion into $1 trillion | TechCrunch
[5] Data center investments surged to $455B last year: report | CIO Dive
[6] How big is the AI funding gap? Morgan Stanley calculated the three-year bill: the world needs to raise $1.5 trillion to "bet on tomorrow".
[7] Hyperscaler bond spreads widen as tech giants issue $121bn in debt By Investing.com
[8] Source: Wells Fargo
[9] CoreWeave CEO explains why its $8 billion debt is not a red flag for investors — TradingView News