Technology hedge fund manager Dan Niles warns: By early 2027, AI-related stocks may be cut in half, down 30-50% from their highs.

Written by: Long Yue

Source: Wall Street Journal

The current market resembles 1997-1998—Wall Street tech veterans are already counting down to the end of the AI bull market.

On May 11, notable chip analyst from the internet bubble era and founder of Niles Investment Management, Dan Niles, was interviewed by The Master Investor Podcast, where he systematically elaborated on his judgment regarding the current AI market: The AI bull market has not yet ended, but he predicts a significant pullback could occur in early 2027, and investors should start preparing now.

At the same time, a survey by JPMorgan of 56 global investors found that 54% expect the U.S. stock market to experience a pullback of over 30% this year or next, with 45% believing it will happen in 2027—aligned closely with Dan Niles' assessment.

It is now 1997-1998, not 1999, and certainly not 2000

Niles likens the current market to 1997-1998, rather than the feared bubble peak of 1999-2000.

The logic is as follows: ChatGPT was launched at the end of 2022, and AI infrastructure development has now entered its fourth year. In the internet era, the Netscape browser was launched in 1994, and 1997-1998 marked the third and fourth years of its development.

In 1997, the Thai currency crisis erupted, and the S&P 500 dipped 11% at one point during the year, but ended the year up 31%. In 1998, Russia defaulted on its bonds, and Long-Term Capital Management (LTCM) collapsed, resulting in a maximum decline of 19% in the S&P 500 for the year, yet it still rose 27% for the year.

Niles said: "Back then, there was a backdrop of internet infrastructure development, so every macro shock became a buying opportunity. The same is true today."

He believes the oil price shock caused by the Iran war is a "man-made event," which is easier to resolve than previous currency crises or bond defaults, thus concluding that this will also be a temporary low point.

Agentic AI: The new fuel driving this bull market

Niles attributes the core driving force of this year's market to one term: Agentic AI.

In simple terms, previously when you asked ChatGPT a question, it would give you an answer; this is termed "conversational AI."

Agentic AI, however, is fundamentally different. Dan Niles illustrates: "You can tell it, 'This is Wilfred, go to the BBC website to get this data, go to Bloomberg to get that data, go to CNBC to retrieve other content, and then compile everything into an Excel spreadsheet.'" This series of operations requires a large number of concurrent calls, consuming 10 to 100 times the computing power tokens compared to conversational AI.

The data has already proven this: Two months before the release of Open Claw on January 30, 2026, the token growth rate was about 20%; in the two months following the release, the growth rate exceeded 120%.

This directly boosted the capital expenditure expectations of ultra-large cloud service providers: at the beginning of the year, the market expected a capex growth rate of about 30% for 2026, which rose to 60% after Q1 earnings and then further to 70% after the latest earnings round.

Niles concludes: This is not a small change, but a leap in magnitude, sufficient to support continued rises in AI-related stocks.

Hardware landscape changing: CPU rising, GPU under pressure

The architectural characteristics of Agentic AI are quietly rewriting the competitive landscape of AI hardware.

Training large models is about repeatedly performing the same task, which GPUs excel at; conversational AI reasoning is also sufficient; however, Agentic AI requires simultaneous scheduling of multiple applications and coordination of multi-step tasks, which is essentially "orchestration," a strength of CPUs.

Dan Niles said: "In the past, it was about 8 GPUs for every 1 CPU. After shifting to Agentic, this ratio will approach 1 to 1."

This means: Intel and AMD will benefit, while Nvidia will "marginally be impacted in its stock price performance."

But the semiconductor sector is severely overbought

Dan Niles shifts his tone: Short-term risks cannot be ignored. "In the short term, the current level of semiconductor overbuying is the most severe since the major crashes of 2000 or 1995. This is certain."

He specifically mentioned that semiconductor ETFs have risen about 70% this year, and the impact of the Iran war has not been enough to bring it down.

However, he also emphasizes that short-term overbought conditions do not equate to long-term logic being destroyed—Agentic AI’s demand for computing power is real. He is willing to endure the risk of Intel potentially falling 15-20% in the short-term because he believes this stock will be higher by the end of the year.

Where will the 30-50% pullback in 2027 come from?

Dan Niles is already forecasting the next cycle.

The explosion of Agentic AI began on January 30, 2026. Based on this, by early 2027, growth will start to compare against a high baseline, and the growth rate will naturally slow significantly. What will the market look like then?

"I believe these stocks, from the high point they will be at, could potentially fall 30% to 50%," he said.

The reference point is right in front of us: In 2022, the "Magnificent Seven" average dropped 46%—this was just the result of the tide receding from the technology construction wave during the pandemic, far smaller than the current AI frenzy.

Another potential ignition point is OpenAI. Dan Niles pointed out that OpenAI and Anthropic together account for about half of the backlog orders for ultra-large cloud providers. The two companies have grown from a combined annualized revenue of about $7 billion in early 2025 to nearly $70 billion now (Anthropic about $45 billion, OpenAI about $24 billion)—this growth is indeed remarkable, but this money must be squeezed out of budgets from other companies.

"When OpenAI’s revenue was still at $20 billion at the end of the year, it declared that it would commit $1.4 trillion in capital expenditures over the next eight years. Where is that money coming from? If OpenAI encounters issues, that would greatly accelerate this process."

He also pointed out a structural liquidity pressure: IPOs from companies like OpenAI, SpaceX, Anthropic, etc. are about to follow one after another, with each company’s valuation likely exceeding a trillion dollars. "This money needs to flow out from elsewhere, as fund managers do not sit on large amounts of idle cash; they need to sell other things."

Three signals light up simultaneously: Stock market, oil prices, bonds, one must be wrong

Dan Niles’ first task every morning is to simultaneously look at oil prices, bond yields, and the stock market.

The current combination makes him uneasy: the stock market is hitting record highs, oil prices have risen about 60% this year, and yields on both 10-year and 30-year U.S. Treasury bonds have reached annual highs.

Historically, in the past 12 recessions, there were 10 instances where sustained rises in oil prices preceded them. If oil prices maintain around $90 for one to two quarters, inflation will rise, and consumers' purchasing power will erode, leading to an inevitable significant pullback in the stock market. McDonald's recently mentioned in its earnings report that lower-end consumers are feeling pressure, and same-store sales did not meet expectations—such signals are starting to emerge, and you cannot help but worry.

He also mentioned that incoming Federal Reserve Chairman Kevin Walsh tends to favor interest rate cuts and sees AI as a deflationary force, "this is a positive factor for pushing markets upward in the short term." However, he warned that if the 10-year or 30-year yields continue to climb, the market valuations will face real pressure. His conclusion is:

Among the stock market, oil prices, and bonds, one must be wrong. When one is re-priced, it could trigger a lot of market chaos.

His recommendation is succinct: "You should hold a lot of cash now. Back at the end of March, I thought it was a good time to enter the market actively. But now, I believe you should hold a lot of cash and remain highly alert to the eventual resolutions."

JPMorgan: 54% of institutional investors expect a significant pullback in 2027

Dan Niles' warning is not an isolated instance.

JPMorgan's global market strategy team released a roadshow feedback report on May 12, 2026, showing that analyst Eduardo Lecubarri led a team that visited five cities in Latin America and engaged with 56 institutional investors.

The core data from the report is as follows:

• 92% of surveyed investors believe the stock market returns for the entire year of 2026 will be positive, but none expect gains to exceed 20%;
• 54% of surveyed investors expect the stock market to experience a pullback of over 30% within 2026 to 2027 (of which 9% expect it in 2026, and 45% expect it in 2027);
• 75% of surveyed investors believe there is still over 20% upside before reaching the tech bubble peak;
• On regional allocations, surveyed investors have a highly uniform view on Europe—100% underweight Europe, 100% overweight the U.S.;
• The preferred sectors for investors in order are technology, utilities, and industrials.

This aligns closely with Dan Niles' "1998 logic": The bull market is far from over, but the timetable for a significant pullback has quietly formed in the market consensus.

Quantum computing: Huge potential, but don't rush

At the end of the interview, Dan Niles also talked about quantum computing. He is a strong long-term believer, "I firmly believe in quantum, and I think we will eventually get there"—but he quotes Bill Gates’ famous saying: "Technology is often overestimated in the short run and underestimated in the long run."

"The earliest AI papers were written over 50 years ago; when did ChatGPT appear? At the end of 2022. Quantum computing likely has a similar path. The arrival of quantum computing IPOs will bring market attention back, but truly disruptive applications will take longer to arrive than most people think."

Big tech company differentiation: Google stands out

The earnings reports from tech giants over the past few weeks have made Dan Niles' judgment clearer.

Google Cloud: Q4 year-on-year growth rate of 48%, accelerated to 63% in the latest quarter, a 15 percentage point increase.

AWS: accelerated from 24% to 28%, an increase of 4 percentage points—this is already quite good for the largest cloud provider.

Microsoft Cloud: accelerated from 38% to 39%, almost flatlining.

"These numbers tell you who is truly executing and who is gaining market share," Dan Niles said.

He directly concludes: "In the next 3 to 5 years, who is the best positioned among the big tech companies? Clearly, it is Google. They have a complete technology stack, and bets should be placed on them. Unless something drastically changes, they will continue to be winners because they have it all and have enormous cash flow to support them."

Google's advantages include: self-developed large language model Gemini, self-developed AI chips (over ten years of history, the longest among the three cloud providers), strong cash flow supported by advertising business, and the Android ecosystem covering over 75% of smartphones worldwide. Microsoft relies on OpenAI and does not have its own large model; Amazon’s AI product recognition is limited.

Meta's situation, however, is relatively concerning. Dan Niles points out that Meta has no public cloud and cannot sell excess computing power to external companies; its progress in developing self-developed ASIC chips is also late. More importantly, Meta saw a quarter-on-quarter decline in Family of Apps users for the first time this quarter—two growth engines of its advertising monetization model (user count and user price) have already reached an inflection point, "this is something to be worried about."

Regarding Apple, Dan Niles believes that the launch of an AI version of Siri and a foldable iPhone will drive a wave of upgrades—he cites the example of the iPhone 6: the screen size increased from 4 inches to 5.5 inches, pushing Apple's revenue growth from 7% to 28%.

Stay flexible, don’t be greedy

Dan Niles summarizes his market philosophy in one sentence: "Be nimble. Hold strong beliefs, but hold them loosely."

His judgment framework is: short-term momentum is upward, Agentic AI and loose monetary expectations are still the two main engines; however, by early 2027, these growth rates will start to compare against a high baseline, and the explosive growth brought by Agentic AI will enter a comparatively weak period, combined with the potential risks of OpenAI and liquidity shocks from super IPOs, "stock prices may fall 30% to 50% from the high point at that time."

What should be done now? Maintain cash reserves, closely monitor oil prices, Treasury bond yields, and the stock market every morning, and be ready to adjust.