Dhaval Joshi AI Market Bubble: Why There Isn’t Just One AI Bubble

The debate over Dhaval Joshi AI market bubbles starts with a distinction that many market headlines miss. Joshi is not simply arguing that artificial intelligence is one giant speculative bubble waiting to burst. His thesis is that AI is producing a rolling sequence of bubbles as investors repeatedly change their view of which companies, sectors or physical assets will capture the technology’s economic value.

That interpretation matters because money does not have to leave the AI trade when one narrative fails. It can rotate. Software can fall out of favor while silver attracts an AI-infrastructure premium. Silver can correct while semiconductor shares become the preferred expression of demand for compute. Fortune’s August 16, 2026 report on Joshi’s thesis describes this as a rapid sequence of inflation, reassessment and deflation across different AI-linked markets.

The framework does not prove that every sharp move is a bubble, nor does it identify the next winner in advance. It offers a way to understand why a genuine technological transformation can coexist with repeated investment excesses, and why the collapse of one AI-related trade need not end the wider boom.

Who Is Dhaval Joshi?

Dhaval Joshi is a London-based global macro strategist whose current LinkedIn profile describes him as a “Global Macro Strategist”. Until recently, he was chief strategist for Counterpoint at BCA Research, where he analyzed cross-asset themes, market structure and global macroeconomic risks. BCA Research’s May 2026 publication archive still identifies him as chief strategist at the time of that report, while Fortune’s August reporting describes BCA as his former firm.

That timing is important. Joshi’s rolling-bubble argument is current commentary from his independent public profile and his August 2026 Fortune interview. His earlier BCA work provides intellectual context, particularly his focus on market complexity, speculative crowd behavior and the difference between technological value and shareholder profit.

What Does Dhaval Joshi Mean by a “Rolling Sequence of AI Bubbles”?

The traditional bubble model is linear: enthusiasm builds around one asset or sector, prices detach from defensible fundamentals, and the bubble eventually bursts. The correction then removes speculative capital from that market.

Joshi’s rolling-bubble model is different:

One AI narrative attracts capital, excess develops, the market reassesses the thesis, that trade corrects, and capital moves toward a new AI beneficiary.

In Joshi’s original rolling-bubbles commentary, he says investors are repeatedly assessing and then rapidly reassessing who will be the winners and losers from AI as a general-purpose technology. He identifies software, silver and semiconductors as successive expressions of that process.

This does not require all AI-linked assets to peak together. Each market has a different story, supply structure, valuation and investor base. SaaS depends on subscription economics and customer retention. Silver is a physical commodity with industrial and monetary demand. Semiconductors depend on capacity, technology leadership, pricing power and the durability of extraordinary margins. The shared element is the AI narrative, not identical fundamentals.

The amplitude and speed of the moves are central to Joshi’s definition. He told Fortune that a market move begins to resemble a bubble when large fortunes can be made in weeks or months and then lost just as quickly. In his view, ordinary price discovery becomes speculative excess when the magnitude and rapidity of the move overwhelm a reasonable reassessment of fundamentals.

Bubble #1: Why AI Turned Software and SaaS Into a Bubble

Why Software Initially Looked Like an AI Winner

Software initially appeared to be an obvious beneficiary of generative AI. Established SaaS companies already had enterprise customers, workflow data, distribution and subscription relationships. Adding AI assistants and automation promised better products, higher worker productivity and potentially more valuable customer relationships.

That was the first-stage narrative in Joshi’s sequence. Investors treated AI as a feature that incumbent software vendors could integrate and monetize. If customers could do more with the same platform, software companies appeared positioned to sell premium AI functions, deepen adoption and protect recurring revenue.

How That Narrative Reversed

The same technology later came to be viewed as a threat. More capable AI agents can perform tasks across applications, generate code, analyze data and create customized workflows. If an agent becomes the main interface through which work is completed, some traditional applications may lose user attention, seat-based pricing power or even their role as a distinct product.

This concern was not confined to Joshi’s chart. Reuters reported in February 2026 that software shares were being repriced as AI moved from a perceived tailwind to a possible source of disruption at the application layer. The same report also noted the counterargument: established vendors may retain advantages from specialized data, trusted workflows and enterprise integration, so disruption is not guaranteed.

Joshi’s claim is therefore about a reversal in market perception, not proof that SaaS will disappear. The original belief that AI would automatically enhance incumbent software value gave way to fear that agents could weaken parts of the subscription model.

What the SaaS Bubble Reveals

The software episode demonstrates the difference between operational benefit and shareholder value.

A company can use AI to make employees faster, reduce service costs or improve a product. Those benefits do not automatically make the company’s business model more defensible. If competitors gain access to similar capabilities, customers can switch more easily, or agents reduce the need for paid seats, the productivity benefit may be passed to customers rather than retained as higher margins.

Joshi had made this distinction before the current rolling-bubble thesis. In his earlier analysis of AI profit capture, he argued that an innovative technology does not guarantee exceptional profit for its creator unless the company has a moat that prevents those gains from being competed away. The SaaS reversal is a current example of that broader principle.

Bubble #2: Why Silver Became Part of the AI Bubble Story

How AI Infrastructure Created a Silver Narrative

AI is a digital technology built on physical infrastructure. Data centers require servers, circuitry, power distribution equipment and extensive electrical connections. Silver has unusually high electrical and thermal conductivity and is used in electronics and electrical components. A Silver Institute report prepared with Oxford Economics argues that expanding data centers and AI-related infrastructure should support silver demand over time.

That industrial link gave investors a simple narrative: more AI compute means more data centers, more electrification and therefore more demand for conductive metals. The narrative has a genuine fundamental basis. The investment question is whether the scale of the expected demand justified the scale and speed of the price move.

Why Joshi Says the Silver Move Went Too Far

Joshi argues that it did not. In his rolling-bubbles post, he acknowledged silver’s conductivity and its role in an electricity-intensive AI buildout, but said that connection could not justify what he described as a near tripling of the silver price, particularly because alternative conductors exist.

That is Joshi’s interpretation, not a settled valuation fact. Silver’s rally had several possible drivers, including its monetary role, investment demand, supply constraints and industrial use. Reuters’ February 2026 silver-market report showed why a single-cause explanation would be incomplete: physical investment demand was expected to rise, while industrial fabrication was forecast to decline as high prices encouraged thrift and substitution.

The balanced conclusion is that AI helped create a powerful demand narrative, but it was not the only force affecting silver. Joshi’s bubble claim rests on his judgment that the price move greatly exceeded what the incremental AI demand story could support.

What Silver Reveals About the Rolling-Bubble Theory

Silver expands the thesis beyond technology stocks. An AI bubble can form in an input, commodity or infrastructure bottleneck if investors believe that asset is essential to the buildout.

It also shows how a valid technological connection can become an overextended investment conclusion. AI infrastructure may consume more silver, yet demand can respond to price through substitution, thrift, recycling or additional supply. A true premise does not remove valuation discipline.

Bubble #3: Are Semiconductor Stocks the Next AI Bubble?

Why Semiconductor Stocks Became AI Winners

Advanced chips are the most direct physical enablers of modern AI. Training and running large models require accelerators, memory, networking components and specialized manufacturing equipment. As hyperscalers raced to secure compute capacity, supply bottlenecks and intense demand strengthened pricing power across important parts of the semiconductor chain.

Joshi describes the semiconductor stage as the market’s belief that AI’s demand for compute would allow chipmakers to charge exceptionally high prices for an extended period. In his sequence, the trade followed the software and silver episodes as capital searched for a beneficiary with clearer revenue and stronger near-term earnings.

Joshi’s Concern About Semiconductor Profit Margins

Joshi’s concern is not that chip demand will suddenly vanish. It is that investors may be capitalizing peak profitability as though it were permanent.

In his June 2026 commentary on chipmaker margins, he argued that the bubble was in profit margins rather than headline valuation multiples. His mechanism is straightforward: extraordinary demand and constrained supply create exceptional pricing power; high returns attract competition and capacity; supply eventually catches up; and margins normalize.

This is a forecast and can be challenged. Technological leadership, software ecosystems, manufacturing complexity and scale can create durable moats. Joshi’s claim is not that every semiconductor company lacks competitive advantage. It is that current profit pools may be more exposed to supply normalization and competition than low price-to-earnings ratios suggest.

Why Strong AI Demand Does Not Guarantee Strong Semiconductor Returns

Demand, earnings and investment returns are related, but they are not interchangeable. AI demand could remain strong while new capacity reduces scarcity pricing. Competitors could capture a larger share of the market. Customers could develop alternatives or negotiate more aggressively. A stock’s valuation could fall even if revenue continues growing because expectations had been higher.

The Bank for International Settlements’ 2026 assessment of the AI investment boom supports the broader economic mechanism without endorsing Joshi’s specific market call. It warns that intense competition can push capital expenditure higher, reduce the net economic surplus available to the sector and create vulnerability if expected returns disappoint. Strong demand for compute can therefore coexist with weaker future returns on the capital committed to supplying it.

What Does Dhaval Joshi Mean by an “AI Profit-Margin Bubble”?

An earnings bubble and a profit-margin bubble focus on different parts of the valuation equation.

In an earnings bubble, reported or forecast earnings are too optimistic relative to the sustainable earning power of the business. A stock may appear reasonably valued on current estimates, but those estimates later prove excessive.

In Joshi’s profit-margin bubble, the immediate earnings may be real, and the price-to-earnings multiple may not look extreme. The vulnerability lies inside those earnings. Revenue is being converted into profit at an unusually high rate because scarcity, pricing power or temporary supply constraints have lifted margins. If those margins revert, earnings fall and the apparently reasonable valuation becomes less reassuring.

Joshi told Fortune that the market should ask not only whether the price-to-earnings ratio is high, but why the earnings denominator is so high. He contrasted his view with former BCA colleague Peter Berezin’s characterization of an earnings bubble, although the two ideas overlap. A margin reversal would ultimately reduce earnings as well.

The concept should be treated as Joshi’s analytical interpretation, not an established law. Some firms can sustain unusually high margins when network effects, intellectual property, switching costs or scale create durable barriers. The empirical question is whether today’s AI-linked chip profits reflect such structural advantages or a temporary imbalance between demand and supply.

Why AI Bubbles Can Keep Moving From One Market to Another

The rotation mechanism begins with uncertainty about value capture. Investors know AI is economically important, but they do not know in advance which business models, inputs or individuals will retain the gains.

Software offered distribution and recurring revenue, but agents raised questions about incumbent moats. Silver offered a physical bottleneck story, but substitution and the scale of the rally challenged that thesis. Semiconductors offered visible demand and profit, but the durability of scarcity margins became the next concern.

As each narrative weakens, the underlying belief in AI can survive. Capital then seeks a new expression of the same theme. That process can prevent sector-specific corrections from becoming a synchronized market-wide selloff.

The sequence is not automatic. Rotation requires liquidity, risk appetite and another plausible beneficiary. If those conditions disappear, capital may leave risky assets rather than move to the next AI trade.

What Could Be the Next AI Bubble?

Joshi does not claim to know. He told Fortune that identifying the next candidate is the central unanswered question. His examples should therefore be read as areas to watch, not forecasts of guaranteed price increases or collapses.

DDR3 Memory

DDR3 is an older generation of computer memory that has largely been superseded by DDR4 and DDR5. In Joshi’s July 2026 DDR3 commentary, he said its price had risen 600 percent in less than a year.

His explanation was a supply-chain ripple. AI data-center demand concentrated production and purchasing pressure on newer memory. That pressure spread into DDR4 and then into legacy DDR3 for users who could still use it. Joshi viewed the move as a scarcity squeeze likely to correct through new supply or demand destruction, and he suggested watching for the first break in the unusually long sequence of price increases.

DDR3 is not a definitive prediction of the next bubble. It is an illustration of how AI demand can reprice adjacent and even obsolete components when supply is inflexible.

Could Crypto Become an AI Bubble?

Joshi told Fortune that crypto’s limited participation in the sequence was unusual. He added that crypto could become a candidate if AI and blockchain technology developed meaningful synergies.

The conditional language is essential. He did not predict that crypto would be the next AI bubble, nor did he identify a specific token, use case or timetable. He identified it as a possible future area of interest, rather than making a definitive prediction. Without a defensible mechanism that links AI adoption to sustainable blockchain demand, the idea remains speculative.

When Does Dhaval Joshi Expect AI Capex to Peak?

Joshi told Fortune that AI capital expenditure would most likely peak in late 2026 or the first half of 2027. This is his forecast as of August 2026, not an observed turning point.

AI capex matters because it is the spending stream connecting the entire infrastructure chain. Cloud providers finance data centers. Data centers require chips, memory, networking, cooling and power equipment. Suppliers plan capacity around expected orders. If spending growth peaks, investors may reassess revenue growth, scarcity pricing and margins throughout that chain even if the absolute level of spending remains high.

The distinction between a peak in capex and a collapse is crucial. A peak can mean spending stops accelerating, reaches a high plateau or begins a gradual decline. It does not automatically imply canceled projects or a market crash.

Current evidence does show why the timing matters. Reuters reported in July 2026 that Microsoft, Alphabet, Amazon, Meta and Oracle were on course, based on consensus estimates, to spend more collectively on capital expenditure than they generated in free cash flow by 2027. Reuters also noted that company disclosures do not cleanly separate AI capex from broader cloud and infrastructure spending. Any forecast about an AI-specific peak therefore carries substantial measurement uncertainty.

What Could Break the Entire AI Bubble Sequence?

Joshi identified three conditions that could stop capital from rotating and instead push it out of risky assets more broadly.

A Sharp Rise in Real Interest Rates or Bond Yields

Real yields represent the return available on safer assets after accounting for inflation expectations. A sharp increase raises the discount rate applied to future corporate cash flows and makes speculative assets less attractive relative to bonds. It also increases financing costs for data centers, suppliers and leveraged investors.

Joshi told Fortune that accommodative monetary conditions are an important support for bubbles. If real rates or real bond yields rose sharply, capital might not move into the next AI beneficiary. It could leave risk assets altogether.

A Sharp Unwinding of AI Capex

A gradual peak in spending is different from a sharp unwind. A sudden reduction in hyperscaler orders would affect chipmakers, memory producers, data-center developers, power-equipment suppliers and creditors at the same time.

This risk is independently consistent with the BIS assessment. The institution warned that disappointment in AI returns could trigger a financing pullback, turn the capex boom into a prolonged investment bust and transmit stress through suppliers and credit markets. Joshi’s point is narrower: a sufficiently sharp capex reversal could synchronize the bubbles that had previously corrected one at a time.

A Significant Recession

Joshi’s third condition was what he called a “non-mild recession.” A serious downturn would reduce revenue expectations, tighten financing conditions, weaken corporate investment and raise risk aversion across sectors.

In that environment, investors may not have the confidence or liquidity to rotate from a falling software, commodity or semiconductor trade into a new AI narrative. The macro shock would dominate the sector-specific story.

Does Dhaval Joshi Think the AI Boom Is About to Crash?

No, not in the simple sense implied by that question. Joshi is not making a verified prediction that the entire AI market will crash in 2026.

His rolling-bubble thesis says individual AI-linked markets can inflate and deflate while the wider investment cycle continues. Software can be repriced without ending data-center construction. Silver can correct without eliminating semiconductor demand. Semiconductor margins can normalize without proving that AI has failed as a technology.

The entire sequence would require a broader break, such as sharply higher real yields, a severe capex unwind or a significant recession. Even Joshi’s forecast that AI capex may peak around late 2026 or the first half of 2027 does not by itself amount to a crash call.

The accurate interpretation is more nuanced: Joshi expects repeated mispricing as markets search for the ultimate winners, and he believes some current profit margins are unsustainable. That is a warning about sector-level valuation and value capture, not a precise market-wide crash forecast.

What Is the Real Risk Behind the AI Bubble?

The deepest risk is capital misallocation even if AI proves technologically transformative.

Companies may spend on capacity that later earns inadequate returns. Investors may value temporary scarcity margins as permanent. Incumbents may improve their operations with AI while losing pricing power to new competitors. Suppliers may expand just as demand growth slows. Markets may identify the right technology but the wrong companies or inputs as long-term winners.

This is the central distinction between technology success and investment success. Railways, telecommunications and the internet created enormous social value while many individual investments failed to earn acceptable returns. The relevant lesson is not that AI must follow the same path, but that genuine usefulness does not protect capital from overpayment or excessive competition.

The BIS makes a similar distinction at the macro level. Its 2026 analysis says competitive investment can raise capex while reducing the sector’s net economic surplus. In practical terms, firms can collectively build valuable infrastructure and still earn less than investors expected.

Who Actually Captures the Economic Value Created by AI?

Joshi’s value-capture framework offers three possible destinations for the economic surplus. Fortune reported these as scenarios, not mutually exclusive predictions.

Companies With Durable Competitive Advantages

The first scenario resembles the Web 2.0 model. Companies with genuine moats, such as network effects, proprietary data, scale, distribution or high switching costs, retain a large share of the value. They can keep prices and margins high because competitors cannot easily reproduce the full offering.

This is the outcome many AI equity valuations implicitly require. It is also the assumption Joshi questions when he argues that chipmakers or software firms may not have permanent protection around their profits.

Highly Productive Individuals

The second scenario shifts value to “superstar” workers. Joshi used the example of a highly skilled lawyer or consultant who uses AI to reduce staffing costs while continuing to charge for premium output. The individual, rather than the AI vendor or the client, captures much of the productivity gain.

This outcome would make AI economically valuable without guaranteeing extraordinary profit for every company that supplies the technology.

Consumers Through Lower Prices

The third scenario is intense competition. If many firms can deploy similar AI capabilities, none may sustain exceptional margins. Competition passes productivity gains to customers through lower prices, better products or both.

Society can benefit substantially under this scenario while investors in individual AI companies earn disappointing returns. It is the clearest expression of Joshi’s central point: technological success does not determine in advance who captures the profit.

Dhaval Joshi AI Market Bubbles Explained

The following table summarizes Joshi’s framework. It does not mean that every stage is universally accepted as a confirmed bubble.

Conclusion: The AI Bubble May Be a Sequence, Not a Single Event

The Dhaval Joshi AI market bubbles thesis reframes the debate. AI is not one homogeneous investment, and its technological importance does not require every AI-linked asset to rise or fall together. Different markets can experience separate speculative cycles as investors search for the companies, workers and inputs most likely to capture the gains.

Software and SaaS show how a perceived beneficiary can become a disruption target. Silver shows how AI enthusiasm can migrate into physical inputs. Semiconductors show why exceptional demand does not guarantee permanently exceptional margins. Together, they illustrate Joshi’s rolling-bubble framework without proving that every price move is irrational.

The central question is value capture. Durable corporate moats, highly productive individuals or consumers through lower prices could receive the largest share of AI’s economic benefits. If competition erodes margins, the technology may succeed while many investments disappoint.

A correction in one sector does not necessarily end the AI cycle. A sharp rise in real yields, a severe capex reversal or a significant recession could potentially break the broader sequence by forcing capital out of risk assets altogether.

The more useful question may not be whether AI is a bubble, but which AI-linked market is being priced for perfection and what happens when investors move on to the next one.