Is the AI honeymoon over? Why 2026 is the year of the reality check

The most compelling argument for a bubble lies in the sheer scale of investment coupled with the elusive nature of profits.

Sky-high valuations and market masking

AI-related enterprises accounted for roughly 80 per cent of gains in the American stock market in 2025. But by January 2026, a new concern has emerged: AI infrastructure spending is masking slowdowns in consumer spending and the services sector. The market is pricing in a perfection that may take years to deliver.

The debt-for-data centre boom

The foundation of the boom is massive capital expenditure (CapEx). JPMorgan forecasts that investment-grade tech companies could accumulate US$1.5 trillion (around £1.2 trillion) in debt by 2030 to support these ventures. This debt load makes the system vulnerable to any reduction in revenue expectations, especially following the record-duration US government shutdown in late 2025.

The OpenAI Profitability Paradox

OpenAI remains the primary litmus test. Despite its leadership, the company is projected to burn through US$115 billion through 2029 before generating positive cash flow in 2030. In 2026, the “cash burn” question has become the defining query for investors: can even the most successful AI firm justify its valuation without immediate profit?

Read more: What to know about Scale AI, the data company powering OpenAI, Meta and Google

Kenneth Albolote: The article makes the case that today’s AI market exhibits bubble-like characteristics—not because investment levels are inherently excessive, but because capital intensity has become increasingly detached from near-term, verifiable unit economics. By definition, I believe the AI boom is not a classic speculative bubble. For the most part, if we are talking about the listed hyper-scalers, the investments are supported by profitable companies, strong balance sheets and real technological progress. However, equity markets are aggressively discounting a future of nearly universal AI adoption, rapid productivity gains and sustained margin expansion. 

History does offer many compelling, unmistakable warnings. Railroads, telecom networks and the early internet all demanded massive upfront investment long before monetisation. While each ultimately transformed the economy, early equity returns in the early years were often poor. Investors financed infrastructure that delivered extraordinary societal value but subpar near-term shareholder outcomes. For some AI players, they may be on track to repeat this pattern, with expectations resetting even as the technology itself continues to advance.

However, I believe that today’s AI boom is unfolding against a materially stronger foundation that is characterised by healthier balance sheets, robust cash flow generation and broadly profitable companies. Valuations may be considered elevated but can be justified by robust profit margins and capital discipline in leading tech names, with systemic risk signs only emerging. 

AI’s current impact is indirect, driving CapEx and wealth effects rather than immediate productivity gains; eventual benefits will materialise at a more moderate pace than what markets expect. Many early-stage growth investors, including Foxmont Capital, are fully aware of this “lag” and are betting on outsized returns for their winners to compensate for moderate scaling up in profits and cash flow generation for some players.

The debate reached a fever pitch at the World Economic Forum (WEF) 2026 in Davos. Speaking on Tuesday, Microsoft CEO Satya Nadella issued a rare and pointed warning: for this not to be a bubble “by definition”, the benefits of AI must be much more evenly spread across the global economy.

Nadella highlighted that a “tell-tale sign" of a bubble would be if AI benefits remain concentrated solely among tech companies rather than delivering real value to industries like healthcare, education and manufacturing. He argued that 2026 is a “pivotal year” where the tech must move beyond supply-side hype and “bend the productivity curve” for the wider world. Without this rapid diffusion, the current boom risks being remembered as a purely speculative investment rather than sustainable growth.

Kenneth Albolote: Nadella’s warning makes an important point. He is right to say that the real test is whether AI creates value beyond big tech and actually results in a significant improvement in productivity across broader industries. 

At the same time, technologies typically take significant time to spread. Early gains often stay with platform and infrastructure companies before wider adoption catches up. I believe that an AI bubble would be better defined not by limited impact in a single year, but by a mid-term or multi-year perspective where benefits remain concentrated, and there is no clear path for AI to spread across the broader economy or if productivity benefits only end up being marginal to companies.

Read more: The rooms that matter: the world’s most influential leadership gatherings to attend in 2026

A key structural flaw cited by critics is the practice of circular investment or “vendor finance”. Chipmakers and large cloud providers are not just selling to AI model builders; they are often investing in them. For example, Nvidia, the world’s most valuable chip company, holds a stake in and sells chips to cloud-computing provider CoreWeave, which in turn purchases billions of dollars in capacity from Nvidia. 

This creates an ecosystem where sales figures are reliant on capital flowing within a closed loop. If external funding dries up—or if high interest rates driven by “AI-driven inflation” persist—the whole circuit could short out.

Kenneth Albolote: The analysis of “circular cash” dynamics among AI players suggests that future downside risks in the event that “external financing dries out”. Vendor financing, cross-equity holdings and capital recycling among chipmakers, cloud providers and AI developers can artificially inflate demand while masking the true end-market consumption. This is similar to the late-1990s telecom boom, when vendor-financed growth did not make transparent weaker customer economics and often resulted in write-downs once assumptions unravelled.

While today’s AI leaders are stronger and less leveraged than their telecom-era predecessors, investors need to understand the key vulnerabilities that remain. Revenue concentration and the assumption of limitless compute demand echo earlier claims of infinite bandwidth—narratives that collapsed once capacity expansion outpaced monetisation. Ultimately, the quality or repeatability of revenues over the magnitude of revenues will determine the winners from the losers for investors.

Michael Blakey: Valuation bubbles are not solely the fault of overenthusiastic founders. Investors play a significant role. VCs are incentivised to look for home runs, and AI companies often require enormous amounts of upfront capital to build out infrastructure, train models, acquire GPUs and hire rare talent.

This reminds me of the dot-com era, where I often get asked if things were as crazy as they seem now. Back then, start-ups were raising millions in seed rounds, but people forget that there was no cloud computing. Founders had to physically buy servers and hire people to manage them. What we are seeing now is a magnified version of the same story: capital-intensive infrastructure needs, but with AI.

Read more: From inflation to diversification: Why investors are investing beyond public markets and turning to private assets in 2026

If the American-led AI bubble pops, the financial fallout will be felt most acutely in Asia, the manufacturing backbone of the industry.

The semiconductor sensitivity

Markets in Taiwan and South Korea are intrinsically linked to the boom. Any drop in Western investor confidence immediately translates into reduced orders for Asian manufacturers. A November 2025 tech sell-off saw a 1.17 per cent drop in Taiwan’s Taiex index and a 1.84 per cent fall in South Korea’s Korea Composite Stock Price Index (KOSPI).

The precious metal signal

An unusual red flag in early 2026 is the behaviour of silver and gold. The price of silver doubled in 2025, reaching a peak at the end of December. Analysts suggest this is a massive “hedging” exercise by investors who suspect the AI bubble is nearing its limit.

Michael Blakey: The article makes a strong point that Asia could bear the brunt of a potential AI market crash due to its hardware exposure, but I would add that Southeast Asia is not a passive participant. In this region, we are seeing more practical AI adoption, with companies integrating AI into products to solve real-world problems. 

Where I see a challenge is that founders and investors focus too much on AI features rather than solving actual problems. The best companies will be those that treat AI as a tool, not a product.

Kenneth Albolote: The article correctly emphasises Asia’s downside exposure to the AI-fuelled investment boom. Semiconductor-driven economies, such as Taiwan and South Korea, are highly sensitive to Western capital flows. Any meaningful repricing of AI expectations would have global consequences, extending well beyond US equity markets.

The increased demand and resulting increasing prices of gold and silver point to a general cautionary approach and more of “correlated hedge behaviour” rather than a direct link to hedging a potential AI bubble, as the article points out. Movements in most precious metals reinforce this caution. Gold and silver rallies likely reflect broader hedging against inflation and geopolitical risk, but their timing underscores growing investor unease with highly valued growth narratives, which would include AI.

Read more: While the world watches humanoid robots dance, this founder thinks we’re missing the point

While the market shows clear signs of “froth”, predicting the exact moment the bubble bursts is, as Bridgewater Associates’ Ray Dalio suggests, impossible. Many strategists, however, anticipate a minimum 30 per cent sell-off in the S&P 500 in the second half of 2026.

The pop will likely be triggered when these core assumptions fail:

1. The ROI failure: A report by the Massachusetts Institute of Technology (MIT) found that despite US$30 to 40 billion in enterprise investment into generative AI, about 95 per cent of organisations are seeing zero return. If AI fails to translate into genuine productivity by late 2026, the bubble will deflate.
2. The competition arrives: The near-monopoly of top chipmakers will inevitably face erosion. The entrance of bespoke chips developed by hyperscalers like Google and Amazon, or a rival’s cheaper, more efficient product, will reduce scarcity, compress margins and deflate the valuations of market leaders.
3. AI-driven inflation: If the cost of AI labour and energy keeps inflation high, central banks will be unable to cut rates, placing immense pressure on debt-heavy tech firms.

The AI bubble is not about the failure of the technology; chatbots are already increasing margins in some sectors. Instead, it is a crisis of valuation. A pop would mark a painful but necessary course correction, shifting focus from speculative hype to measurable, real-world returns.

Read more: This scientist warns that civilisation could collapse—unless we change how we train AI

Kenneth Albolote: The article’s proposed triggers are plausible and interconnected, but it is worth noting that none require a technological failure, only an economic downturn or slowdown. MIT findings show that limited near-term enterprise ROI does directly challenge the most optimistic AI narratives. While investors may accept delayed monetisation, the central risk is not patience. It is the sustained mismatch between investment pace and realised productivity gains.

From an investor perspective, I believe the debate should be focused on whether current valuations adequately price execution risk, capital costs and time. A correction, if or when it occurs, in my view would likely strengthen the AI ecosystem by imposing greater discipline, improving transparency and refocusing capital on the leaders with demonstrated differentiation and wide moats for their competitive advantages.

With this in mind, investors such as Foxmont Capital will be focused on investing only in businesses with balance-sheet strength and cash-flow durability that can become long-term winners.

Michael Blakey: There is no doubt in my mind that the bubble will eventually deflate, but that doesn’t mean it was all a waste. Bubbles tend to accelerate infrastructure buildout. They push technology forward faster than would otherwise be possible under normal conditions. When the dot-com bubble burst, the internet did not disappear. What remained was the infrastructure of data centres, broadband capacity and payment systems that laid the foundation for the next two decades of digital innovation. The same will happen with AI.

Also, the concern that AI will wipe out jobs is valid but overblown. Every major technological shift has changed the nature of work, not eliminated it. It is not about replacing humans; it is about augmenting them and creating space for new types of value.

This article was produced as part of a Tatler experiment exploring the intersection of technology and journalism. The initial draft was generated by the AI model, Gemini 3, with factual data and structural guidance provided by human editors. While AI helped synthesise the vast data surrounding the discussion of the AI bubble, all insights and final editorial decisions were overseen by humans

NOW READ

This filmmaker found a new design partner in AI, but doesn’t want it to take over

Can an AI companion be ‘someone to talk to’ for lonely elderly people?

Tatler Most Influential 2025: Richard Teng took over Binance amid a scandal but is rebuilding its credibility, one rule at a time