Record Profit Can't Save Samsung as AI Trade Cracks

Samsung Electronics posted record second-quarter profit and still saw its shares fall nearly 7%, a clean signal that the AI trade is no longer getting paid simply for being right on growth.

That matters most for investors crowded into semiconductors, memory stocks, and data center infrastructure, where expectations have run ahead of what even strong earnings can satisfy. The AI trade is losing steam because the market is shifting from “how big can this get?” to “who earns enough cash from this buildout to justify the price?” CoinDesk reported that the pullback hit names including Micron Technology, Sandisk, Samsung Electronics, and SK Hynix.

“The AI trade, which incorporates semiconductors and memory stocks, is showing signs of fatigue as investors reassess whether the extraordinary spending boom on chips and data centers can be sustained.”

That sentence captures the real issue. This is not proof that AI demand has vanished. It is proof that investors are demanding a harder link between infrastructure spending and durable profits.

The AI stock trade is being repriced because perfection got too expensive

The AI trade worked while investors treated infrastructure as the scarce asset. Chips, high-bandwidth memory, servers, and data centers became the obvious bottlenecks. If demand was assumed to keep rising, suppliers looked like toll collectors.

Now the trade is more fragile. Samsung delivered record quarterly profit, yet missed revenue estimates. That was enough to knock the shares down nearly 7% and extend pressure across AI-linked chipmakers. What happens when record earnings are no longer enough?

XOOMAR analysis: that is a valuation reset, not a demand collapse. The market is asking whether the next dollar of AI infrastructure spending produces the same return as the last one. When a stock has already priced in years of strong orders, even a small crack in the narrative can hit hard.

The rally had been extreme. CoinDesk reported Sandisk was up more than 525%, Micron had gained over 120%, and SK Hynix had climbed roughly 225% in 2026 before the latest pressure. After moves like that, investors don’t need bad news to sell. They only need less-perfect news.

Record earnings can’t hide the AI infrastructure spending problem

The pressure point is capital intensity. AI demand may still be real, but the infrastructure required to serve it is expensive and concentrated in a few layers: chips, memory, data centers, and cloud capacity.

Builders want volume, buyers want discipline

Chipmakers want hyperscalers to keep ordering. Hyperscalers want enough capacity to support near-term workloads without overbuilding. Enterprises, the end buyers in many AI use cases, still need AI tools to justify spending through measurable productivity or revenue gains.

That creates a circular trade:

Could hyperscalers delay projects or negotiate harder once they secure enough near-term capacity? CoinDesk says concerns are growing that they could slow AI infrastructure spending. That single concern now threatens the whole chain.

For crypto investors, this matters because CoinDesk also noted that bitcoin and the broader crypto market have suffered over the past year from the AI trade. We have tracked a separate crypto sentiment reset in MSTR Panic Fades as Bitcoin Market Bottom Takes Shape, but this AI unwind is about capital allocation, not just crypto-specific catalysts.

The numbers behind the AI capex reality check

The supplied numbers point to a trade that became crowded before the business model debate was settled.

Performance: Sandisk rose more than 525% in 2026, Micron gained over 120%, and SK Hynix climbed roughly 225%.

Drawdown: SK Hynix is down 25% from its all-time high ahead of its U.S. listing this week.

Earnings reaction: Samsung reported record second-quarter earnings, missed revenue estimates, and still fell nearly 7%.

Those figures show why the AI trade is sensitive to disappointment. Investors were not buying modest improvement. They were buying a sustained infrastructure supercycle.

What would prove the bulls right from here? Not vague AI excitement. Hard evidence would matter more:

• Order quality: Backlogs that remain firm after hyperscalers reassess spending.
• Cloud monetization: Clearer disclosure showing AI workloads turning into revenue.
• Utilization: Strong usage of already built AI capacity.
• Margins: Suppliers defending profitability as competition rises.
• Enterprise budgets: Buyers expanding AI spending because returns are visible.

CoinDesk does not provide hyperscaler capex totals, GPU utilization figures, data center vacancy rates, or depreciation data. That absence matters. Those are the numbers investors will need before treating this as a simple dip-buying moment.

Chipmakers, cloud giants and investors now want different things from the AI boom

The AI boom is entering a negotiation phase. The first phase rewarded scarcity. The next phase rewards control.

Chipmakers want demand to stay tight. Cloud giants want supply options. Enterprises want cheaper and more useful AI. Investors want proof that capex becomes repeatable cash flow. Can all four get what they want at the same time?

Zhipu AI complicates the answer. CoinDesk reported that the Chinese AI startup is exploring a custom AI chip as demand for its open-source GLM models grows. That points to a different path: lower-cost models, domestic hardware, and less dependence on cutting-edge U.S. chips.

XOOMAR analysis: the threat is not that custom chips instantly replace the dominant AI hardware suppliers. The threat is that customers start optimizing around them. Smaller models, open-source models, and custom silicon can weaken pricing power if they reduce the need for the most expensive infrastructure.

This is where the market signal becomes clearer. Investors are not just questioning AI demand. They are questioning who captures the economics.

The AI boom is starting to rhyme with past infrastructure manias

AI can be a real technology wave and still produce overextended stocks. Those two ideas can coexist.

Past infrastructure cycles often followed a similar pattern: first, demand looked unlimited. Then supply expanded. Then investors realized that the biggest early winners were not always the companies with the best long-term economics. Is AI different enough to escape that pattern?

XOOMAR analysis: the current evidence supports caution, not panic. Samsung’s record profit shows demand has not disappeared. The share-price reaction shows expectations were stretched. SK Hynix’s 25% decline from its all-time high shows investors are already reducing exposure before its U.S. listing draws more capital into the same sector.

The clean distinction is this: AI adoption can continue while AI infrastructure multiples compress. That is the scenario markets are starting to price.

What a cooler AI trade means for tech investors, enterprises and the data center industry

For tech investors, AI exposure now needs discrimination. Balance sheets, customer concentration, pricing power, and cash generation matter more when the market stops rewarding every AI-linked name equally.

For enterprises, a cooler infrastructure frenzy could shift bargaining power. If suppliers and cloud platforms face more pressure to prove demand, buyers may get more model choice and tougher vendor competition. Will that produce better AI economics for customers? It can, but only if vendors translate lower infrastructure costs into better pricing or measurable performance gains.

For data center and infrastructure investors, the signal is mixed. CoinDesk’s source material supports continued focus on chips and data centers, but also warns that hyperscaler spending could slow. That means projects tied to firm demand look stronger than speculative expansion built on endless AI capacity growth.

Crypto readers should also watch capital rotation. CoinDesk says if investor enthusiasm for AI fades, crypto bulls could see capital rotate back into digital assets. That does not guarantee a crypto rally. It does make AI fatigue relevant for bitcoin, alongside separate policy and market stories such as Trump Loses as US Housing Law Takes Effect Without Him.

The next phase of the AI market will reward proof over hype

The AI trade is not over. The easy version is.

The next phase will likely punish companies that need uninterrupted spending growth to defend their valuations. It will favor suppliers that can show durable orders, cloud platforms that can prove AI revenue, and software companies that can monetize AI without depending on endless infrastructure expansion.

Volatility should cluster around a few signals: hyperscaler spending guidance, chip shipment forecasts, gross margins at major suppliers, demand for memory, and enterprise AI budget data. If those indicators stay strong, the recent selloff may look like a reset inside a larger boom. If they weaken, the market will treat the AI infrastructure buildout as overextended.

The thesis is simple: AI demand can remain important while the AI trade loses its automatic premium. From here, proof beats narrative.

Disclaimer: This XOOMAR analysis is for informational and educational purposes only. It is not financial, investment, legal, tax, or professional advice. It does not provide buy, sell, hold, price-target, portfolio, or personalized recommendations. Verify information independently and consult qualified professionals before making decisions.

The Bottom Line

• AI-linked chip and memory stocks are being repriced as investors question stretched valuations.
• Samsung’s nearly 7% drop despite record profit shows expectations may have run ahead of fundamentals.
• The market is shifting focus from AI infrastructure growth to whether that spending produces lasting profits.