Chip stocks suffered a sharp two-day selloff this week as fresh signals from Meta and Anthropic forced traders to reconsider whether the artificial intelligence infrastructure boom is entering a more disciplined phase, with the market shifting its focus from raw spending growth to returns, utilization and cost control.
The Philadelphia Semiconductor Index dropped more than 10% across Wednesday and Thursday, its steepest two-day decline in nearly a month. Memory chip shares were hit even harder. A basket of memory stocks tracked by Goldman Sachs plunged more than 18% over the same period, marking the sharpest two-day fall for that group in 12 years.
The trigger was not a collapse in demand for artificial intelligence. Instead, the selloff reflected growing concern that the next stage of AI development may not reward every part of the semiconductor supply chain equally. Traders moved quickly to reprice companies most exposed to expectations for endless data center expansion, especially semiconductor equipment makers and memory suppliers.
At the center of the reassessment were two separate reports involving Meta and Anthropic. Meta is reportedly considering ways to commercialize excess AI computing capacity, including leasing or selling access to parts of its infrastructure. The goal would be to generate better returns from the company’s multibillion-dollar data center investments.
A day later, reports said Anthropic was in talks with Samsung to develop custom AI chips that could use advanced 2-nanometer process technology. Such a move would be designed to reduce computing costs, improve efficiency and lessen reliance on outside suppliers for some workloads.
Taken together, the two developments suggested that major AI companies are no longer being judged only by how much they spend. The new question is how efficiently they can turn that spending into revenue, model performance and long-term competitive advantage.
The change in tone was enough to hit some of the market’s most popular AI hardware trades. Shares of Teradyne, Entegris, KLA, Applied Materials and Lam Research each fell more than 10% intraday on Thursday, reflecting their close connection to future chip-factory spending. ASML’s U.S.-listed shares dropped more than 5%, adding to the pressure across the semiconductor equipment sector.
SanDisk was among the hardest-hit names in the broader hardware and memory space, falling more than 23% in just two days and moving into bear-market territory. The contrast was notable: while chip and equipment stocks suffered severe declines, shares of large cloud service providers, which are among the biggest spenders on AI infrastructure, were relatively more stable.
That divergence showed how traders were drawing a sharper distinction between companies that buy and operate AI infrastructure and those whose revenue depends on continuous expansion in chip manufacturing, memory demand and equipment orders.
The selloff also exposed a broader shift in market psychology. For much of the past two years, the AI trade was built on a simple assumption: demand for graphics processing units, data centers, memory and networking equipment would continue to rise at an extraordinary pace as companies raced to train and deploy larger models. That assumption remains partly intact. But the market is now asking whether the spending cycle can continue at the same intensity without clearer evidence of returns.
Why the selloff accelerated
The semiconductor sector has been one of the clearest beneficiaries of the AI boom. Demand for advanced GPUs, high-bandwidth memory, networking chips and manufacturing equipment has supported large earnings upgrades and premium valuations across much of the industry.
That strength also made the sector vulnerable. When expectations are high, even a modest change in the narrative can trigger a rapid correction. This week’s decline reflected fear that the AI infrastructure cycle may be moving from an expansion-at-any-cost phase into a period of tighter capital discipline.
That does not mean AI demand is weakening. In fact, companies continue to spend heavily on training large models, supporting inference workloads and building data centers capable of handling rising enterprise and consumer usage. But the market is beginning to separate demand from profitability. Strong demand alone may no longer be enough if companies cannot show that each dollar of infrastructure spending produces measurable economic value.
This is why equipment makers were hit particularly hard. Their order books are highly sensitive to future capital expenditure plans. If chipmakers and AI companies become more selective about factory expansion, equipment suppliers can feel the impact earlier than other parts of the supply chain.
Semiconductor equipment stocks often act as a leveraged expression of confidence in long-term capacity growth. When traders believe chip demand will continue rising without interruption, these companies can outperform. When traders begin to question the timing or scale of new capacity, they can fall quickly.
The scale of Thursday’s intraday declines showed how aggressively traders moved to reduce exposure. The reaction was not limited to one company or one subsector. It spread across equipment, memory and AI hardware, suggesting a broad reassessment of the assumptions that have supported the trade.
The Meta signal
Meta’s reported interest in leasing or selling excess AI computing capacity raised an important question: has the industry already built more infrastructure than it can fully use in the near term?
The answer is not simple. AI data centers are designed for long-term needs, and utilization can fluctuate depending on model training schedules, product launches and internal research priorities. Periods of unused capacity do not necessarily mean the buildout was excessive. Large technology companies often build ahead of demand to avoid supply shortages later.
Still, the idea that Meta could commercialize idle computing power shows that even the largest AI spenders are thinking more carefully about asset productivity. If a company has committed tens of billions of dollars to chips, servers, networking hardware, power systems and real estate, it has a clear incentive to maximize the revenue generated by that infrastructure.
Meta Chief Executive Mark Zuckerberg has previously indicated that selling excess capacity was “definitely on the table.” That comment now carries more weight as the market becomes more focused on whether AI spending can be converted into tangible financial results.
For Meta, commercializing unused capacity could serve several purposes. It could improve returns on existing infrastructure, reduce the effective cost of internal AI development and create a new revenue stream from assets that might otherwise sit underused during certain periods. It could also help the company compete more directly with cloud providers that already sell computing power to external customers.
But for semiconductor traders, the signal was more complicated. If major technology companies are finding ways to raise utilization of existing assets, they may not need to expand capacity as aggressively as previously assumed. Even if total demand keeps growing, better utilization can change the timing of future orders for chips and equipment.
That is why a strategy designed to improve Meta’s economics could be viewed as a negative short-term signal for parts of the semiconductor supply chain. The issue is not that Meta is spending less today. The issue is that the company appears more focused on extracting value from what it has already built.
The Anthropic signal
Anthropic’s reported talks with Samsung pointed to a different but related theme: AI companies want more control over their hardware costs.
The company is said to be exploring custom AI chips that could rely on Samsung’s advanced 2-nanometer process technology. If completed, such a partnership would align Anthropic with a broader industry trend toward in-house or customized silicon.
The logic is straightforward. General-purpose AI accelerators are powerful, but they can be expensive and may not be optimized for every model architecture or workload. By designing custom chips, AI companies can tailor performance, energy use, memory bandwidth and cost structures to their own needs.
This approach is not new. Google has developed its Tensor Processing Units for years. Amazon has built Trainium and Inferentia chips for AI workloads. Meta has been developing its own AI silicon as part of its effort to control costs and improve performance across its platforms. Microsoft and other major technology companies have also moved in similar directions.
Anthropic’s reported interest in custom chips suggests that model developers may increasingly seek a hybrid approach. They can continue using high-end GPUs for the most demanding training and research workloads while relying on custom chips for specific tasks where efficiency matters more.
For the broader market, that has major implications. Nvidia and other leading AI chip suppliers may still enjoy strong demand, especially for the most advanced systems. But as more AI companies explore custom hardware, the future supply chain may become more fragmented and more specialized.
This does not necessarily reduce total semiconductor demand. In some cases, custom chip development could increase demand for advanced manufacturing, packaging and design services. But it can change which companies benefit, when they benefit and how much pricing power they retain.
Samsung’s potential role is also significant. A deal involving advanced 2-nanometer technology would highlight the strategic importance of leading-edge manufacturing at a time when the race among Samsung, TSMC and Intel remains central to the future of high-performance computing. It would also show that AI companies are willing to look beyond existing supplier relationships if they believe custom designs can improve long-term economics.
From spending scale to capital efficiency
The market’s reaction suggests a broader transition in the AI trade. The first phase was dominated by spending scale. Companies announced large data center budgets, placed large GPU orders and emphasized the need to build computing capacity quickly. The market rewarded suppliers positioned to benefit from that rush.
The next phase may be defined by capital efficiency. Traders now want to know which companies can generate the most model output, product improvement and revenue per dollar spent. That shift changes the way the entire sector is valued.
In the early phase of a technology buildout, high spending is often treated as proof of commitment. Large capital budgets can signal ambition, confidence and competitive strength. But as the market matures, spending alone becomes less impressive. Companies must show that spending leads to durable advantages.
This is especially important in AI because infrastructure costs are enormous. Advanced GPUs, memory, servers, networking hardware, cooling systems and power contracts require massive upfront commitments. Training frontier models can cost hundreds of millions of dollars, while inference at scale can create ongoing expenses that rise with usage.
If revenue growth does not keep pace, even the most advanced AI platforms can face pressure. That is why companies are now looking for ways to reduce cost per query, improve chip utilization, automate data center operations and develop specialized hardware.
Meta and Anthropic are not retreating from AI. Their reported actions suggest the opposite. They are looking for ways to make AI economics more sustainable. The market, however, is adjusting to the idea that sustainable growth may not look like unlimited spending across every supplier category.
Why cloud providers held up better
One notable feature of the selloff was the relative stability of large cloud service providers. These companies are major buyers of AI infrastructure, but their shares did not fall as sharply as those of chip equipment and memory names.
That difference reflects how traders view their business models. Cloud providers can benefit from AI adoption in several ways. They can rent computing capacity to other companies, sell AI tools to enterprise customers, integrate AI into existing software products and use automation to improve internal efficiency.
In other words, they are not simply spending on infrastructure. They are also in a position to monetize it. If demand for AI services continues to rise, cloud companies may be able to convert infrastructure into recurring revenue.
That does not mean cloud providers face no risk. Heavy spending can pressure margins, and competition for AI customers remains intense. Power constraints, supply shortages and depreciation costs are also important concerns. But compared with equipment suppliers, cloud companies have more direct control over pricing, packaging and customer relationships.
The market appeared to recognize that distinction. A more disciplined AI spending environment could even benefit the strongest cloud providers if it favors companies with scale, distribution and the ability to run infrastructure efficiently.
The pressure on memory and equipment
Memory chip stocks were among the biggest casualties because AI has been a major source of optimism for the sector. High-bandwidth memory is essential for advanced AI accelerators, and demand has helped improve pricing and earnings expectations after a difficult downcycle.
The Goldman Sachs-tracked memory basket’s 18% two-day drop showed how much optimism had been priced in. When traders began to question whether AI infrastructure spending could keep accelerating at the same pace, memory shares were vulnerable to a sharp pullback.
Equipment makers faced a similar problem. Companies such as Applied Materials, Lam Research, KLA and ASML are essential to semiconductor manufacturing. Their tools are used to produce advanced chips, inspect wafers, etch circuits and support the complex processes required for leading-edge nodes.
But their revenue depends heavily on chipmakers’ willingness to invest in new capacity. If customers slow factory expansion, push out orders or focus on improving utilization of existing facilities, equipment demand can soften.
This is why even a shift in expectations can matter. The market does not need to see actual order cancellations for these stocks to fall. A belief that future orders may be delayed can be enough to reduce valuations.
A correction, not a collapse
Despite the severity of the two-day move, many analysts view the decline as a correction rather than a sign that the AI cycle is ending. AI adoption remains in an early stage across many industries, and the need for computing power continues to grow as models become more widely used.
Training frontier models remains compute-intensive. Inference demand can scale rapidly as AI products attract more users. Enterprise adoption is still developing, and many companies have only begun to integrate AI into software, customer service, research, logistics and internal operations.
That creates a long runway for infrastructure demand. The question is whether that demand will be broad and indiscriminate, as the market once assumed, or more selective and efficiency-driven.
The answer may determine which parts of the semiconductor sector continue to outperform. Companies that offer essential bottleneck technologies, clear performance advantages or strong cost benefits may remain well positioned. Companies whose valuations depend mainly on ever-rising capital expenditure may face more volatility.
For traders, the key issue is not whether AI will grow. It is how that growth will be financed, monetized and distributed across the supply chain.
Broader signals for risk markets
The shift toward demonstrable returns in AI also carries implications beyond technology stocks. When the market’s most important growth theme begins to emphasize efficiency over narrative, other speculative areas can come under pressure.
Digital assets have already shown signs of stress. Total market capitalization for digital assets fell to about $2.11 trillion at the start of the month, down from a peak of roughly $3.5 trillion in late 2025. At the same time, institutional investment vehicles recorded around $7 billion in outflows during May and June 2026, suggesting that some large-scale buyers were cutting exposure.
That pattern fits a wider change in risk appetite. When capital is abundant and confidence is high, assets driven by long-term narratives can attract strong inflows. When traders begin demanding clearer evidence of utility, revenue and durability, weaker projects can struggle.
The AI correction may therefore serve as a warning for other high-growth markets. The market is not rejecting innovation. It is asking for proof that innovation can support sustainable economics.
For digital assets, that means greater attention may fall on networks with real usage, clear fee generation, resilient liquidity and practical applications. Projects that rely mainly on promotional momentum or distant promises may face a more difficult environment if capital becomes more selective.
A more mature phase for AI
Meta’s and Anthropic’s reported initiatives point to a more mature phase in artificial intelligence infrastructure. The industry is still expanding, but the nature of competition is changing. Scale remains important, yet scale without efficiency is no longer enough.
Meta’s reported plan to monetize excess computing capacity is a utilization story. Anthropic’s reported chip talks with Samsung are a cost-control story. Both reflect the same underlying reality: AI companies are under pressure to improve the economics of growth.
For the semiconductor market, that transition has produced a sharp short-term correction. For the AI industry, it may be a healthy adjustment. A focus on efficiency can reduce waste, improve business models and support longer-lasting growth.
The immediate market impact has been painful for chip stocks, especially equipment makers and memory suppliers. But the selloff does not necessarily mark the end of the AI infrastructure cycle. It marks a reassessment of who benefits most as the cycle evolves.
The central question for traders is now clearer than it was before the selloff. The AI boom is no longer only about who can spend the most. It is increasingly about who can turn massive infrastructure commitments into measurable, sustainable returns.
As AI infrastructure matures, explore diversified crypto opportunities in volatile tech cycles with our market opportunities hub today.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
