Chip crash on AI fears: bubble or correction?

What happened?

On Monday and Tuesday of this week, shares of major chipmakers — including NVIDIA, AMD, TSMC, and Samsung Electronics — suffered double-digit declines in US and Asian stock markets. South Korea's Kospi index plunged 8%, triggering a temporary trading halt (circuit breaker), according to BBC Technology. Specifically, NVIDIA lost 9.5% over two sessions, AMD 8.2%, TSMC 7.8%, and Samsung 6.9%, wiping out more than $500 billion in combined market capitalization. The trigger was a combination of factors: quarterly results from some tech companies showing a slowdown in demand for AI chips, executive statements warning of overcapacity, and a Goldman Sachs report questioning whether massive spending on AI infrastructure was generating proportional returns. The report noted that hyperscalers (Google, Microsoft, Amazon) had increased capital spending by 45% year-over-year, but AI service revenues only grew 20%, suggesting diminishing efficiency.

Why is this important?

This move represents the biggest correction in the semiconductor sector since the dot-com bubble burst in 2000. The chip industry is the thermometer of the digital economy: it powers everything from smartphones to AI data centers. A sustained decline could freeze investments in new factories (fabs), disrupt the global supply chain, and slow corporate AI adoption. Moreover, the panic reflects a narrative shift: during 2023 and early 2024, investors bet unreservedly that generative AI (like ChatGPT) would skyrocket chip demand. Now doubts arise about whether tech giants (Google, Microsoft, Meta) are buying more chips than they actually need, creating an inventory bubble. Meta, for example, reported that its GPU inventory rose 30% in the last quarter, while its AI data center utilization fell to 65%. This imbalance between supply and real demand could lead to an inventory correction lasting several quarters.

Consequences for investors and companies

• Short-term volatility: More sharp swings are expected as the market digests big tech quarterly reports. NVIDIA futures already point to another 3% drop by the end of this week.
• Valuation reassessment: Companies like NVIDIA, with a P/E ratio above 70, could face further corrections if revenues miss expectations. Morgan Stanley analysts have cut their price target for NVIDIA from $950 to $800, citing oversupply risks.
• Impact on AI startups: Reduced capital availability and more expensive chips could stifle innovation at startups reliant on specialized hardware. According to PitchBook, AI startup funding in Q2 fell 18% from the previous quarter, and this correction could worsen it.
• Infrastructure investment cycle: Plans for new foundries (such as TSMC in Arizona or Intel in Ohio) could be delayed if demand fails to materialize. TSMC has already warned it may postpone production at its Arizona plant until 2026 if AI chip demand does not pick up.
• Ripple effect on other sectors: Software companies like Salesforce and Adobe, which integrate AI into their products, also saw 3-4% declines, reflecting fears of a broader slowdown.

What should readers know?

This is not a collapse of AI, but a necessary correction. The technology continues to advance, but the market was pricing in unrealistic growth. For end users, this could translate into better hardware prices in the medium term, as manufacturers look to clear inventories. For example, consumer GPU prices (RTX 40 series) are expected to drop up to 15% in the coming months, according to IDC estimates. For investors, it is time to diversify and not bet everything on the chip sector. Treasury bonds and defensive sectors like healthcare or utilities could offer shelter during volatility. Moreover, the correction could open buying opportunities for companies with solid fundamentals: AMD, with a P/E of 35 and a diversified product portfolio, might be a safer bet than NVIDIA.

"The chip stock decline is not the end of AI, but a sign that the speculative bubble is deflating. Real demand for AI semiconductors remains strong, but not at the pace investors had priced in." — Analyst at TheVortiq

Lessons from history

In 2000, the dot-com bubble burst when companies spent billions on fiber optics and data centers that later became underutilized. Today, AI data center operators are following a similar pattern: buying GPUs in bulk, but the profitability of AI models remains uncertain. However, unlike 2000, today's chip companies have solid balance sheets and AI has real applications (chatbots, assistants, automation), suggesting the correction will be shorter and shallower. Additionally, central banks are more willing to intervene: the Federal Reserve has already indicated it could cut rates if volatility spreads. Another key difference is that in 2000, overcapacity took years to absorb, while today demand for chips in automotive, IoT, and 5G continues to grow, which could cushion the blow. In summary, while the correction is painful, it is not an exact repeat of 2000, and long-term investors can find value amid the panic.