Chipmaker Stocks Sink as Memory Costs and Fed Rates Collide

The Philadelphia Semiconductor Index has fallen nearly 24% since late June, confirming a bear market for the world’s biggest chipmakers even as AI computing demand keeps breaking records. Nvidia, AMD and Broadcom extended the slide again on July 27, and Samsung’s stock fell the same week it posted a profit jump of more than 1,800%.

Traders have pinned the rout on AI bubble fears and profit-taking after a blistering rally. Underneath that story sit two more concrete forces: a memory chip shortage inflating hyperscaler budgets, and a newly hawkish Federal Reserve repricing every rate sensitive stock in the market at once.

A Bear Market Built on Record Profits

The SOX, Wall Street’s benchmark for chip stocks, spent the year through June on a historic run, up roughly 130% over twelve months on the back of AI infrastructure spending. That run reversed hard in the first week of July. The index dropped 11% in a single week and is now down close to 24% from its late June peak, the threshold traders use to call a bear market.

Micron Technology led the initial break, falling as much as 13% in one session and erasing roughly $138 billion in market value. Intel dropped 21% over seven trading days, Forbes reported on July 8. In South Korea, the KOSPI index plunged nearly 10% intraday and triggered circuit breakers as Samsung Electronics and SK Hynix each slid between 9% and 12%.

The strangest data point came from Samsung itself. The company reported preliminary second quarter operating profit of 89.4 trillion won (roughly $65 billion), an increase of more than 1,800% year over year. Its stock fell nearly 7% anyway.

Two separate mechanics explain why strong earnings and rising sales did not stop the bleeding, and both run deeper than sentiment.

The Memory Squeeze Behind the Selloff

SK Hynix, the world’s second largest memory chipmaker, is the clearest thread connecting the selloff to something other than mood. The company has pushed the ramp of its next generation HBM4 chips, the high bandwidth memory that feeds data to Nvidia’s and AMD’s most advanced AI accelerators, from the second quarter into the third.

The delay is not a supply failure. It is a business call. TrendForce reported the company is redirecting output toward higher margin commodity DRAM instead of ramping the AI specific chip as fast as originally planned. A semiconductor industry tracker posting as Jukan separately wrote that SK Hynix is throttling HBM4 output to chase DRAM profit instead.

Rising memory prices are not staying contained to memory makers either. Microsoft’s chief financial officer attributed $25 billion of the company’s higher 2026 capital spending forecast directly to rising memory and component costs, on top of the compute buildout itself. When the chips feeding AI servers get more expensive, the trade’s biggest buyers absorb that cost long before its sellers show any sign of a demand problem.

• DRAM prices climbing: average selling prices for commodity DRAM are on pace to rise a cumulative 62% by the end of 2026, squeezing every buyer from hyperscalers to PC makers.
• HBM4 pushed back: SK Hynix moved its next generation ramp from the second quarter to the third, keeping current generation HBM3E lines running longer than planned.
• Margins reset higher: Daishin Securities estimates commodity DRAM operating margins could reach a theoretical 90% this year, pulling supplier investment away from AI specific chips.
• Wafer shortages linger: SK Hynix chairman Chey Tae-won has said capacity constraints could persist until 2030.

None of this shows up as weaker AI orders. It shows up as higher bills for the companies placing them, which is a different problem wearing the same stock chart.

Warsh’s Fed Adds a Rate Shock on Top

The second force is monetary, not industrial. Kevin Warsh became Federal Reserve chairman after President Trump nominated him on January 30, replacing Jerome Powell when his term ended on May 15. At his first policy meeting on June 17, Warsh held the federal funds rate at 3.50% to 3.75% but signalled a hawkish shift that reset how traders price the rest of the year.

Nine of the eighteen officials on the Federal Open Market Committee now project at least one rate increase in 2026, a reversal from earlier forecasts that pointed toward cuts. Warsh has said prices are “too high” and has vowed to make elevated inflation, running near 4.2% year over year, “a thing of the past.”

The next FOMC gathering is already shaping up as contentious. Fortune previewed the meeting on July 25 as a brewing “family feud” among policymakers, with Warsh pushing for tighter policy against colleagues wary of slowing growth further. That fight matters more for chip stocks than for most sectors. Companies trading on earnings that arrive years from now are the most exposed when the risk free rate moves, since every basis point on long term yields shrinks the present value of the payoff all that AI infrastructure spending is supposed to deliver.

Is This 1999, or Just a Pricier Supply Chain?

The comparison holds only partly. Deutsche Bank’s economists have likened 2026 to 1999’s dot-com froth, yet hyperscalers keep raising their spending plans and describe their businesses as supply constrained rather than short on customers. That mix, real spending discipline colliding with real bottlenecks, explains more of the selloff’s timing than a demand collapse would.

Deutsche Bank’s global economics team framed the year bluntly in a June outlook.

2026 is looking a lot like 1999 meets 1990, but hopefully not 1973.

The bank’s analysts were pointing to AI driven stock optimism colliding with a Middle East energy shock, a pairing that echoes both a valuation bubble and an oil driven downturn at the same time.

The spending data cuts against a pure bubble read. Google, Amazon, Microsoft and Meta plan to spend a combined $725 billion on capex in 2026, up 77% from last year’s $410 billion, and Goldman Sachs projects the four largest hyperscalers will spend $5.3 trillion combined between fiscal 2025 and fiscal 2030. Microsoft alone set 2026 capex at $190 billion, well above the $152 billion Wall Street had expected. Alphabet raised its own guidance by $5 billion to as much as $190 billion, and Meta added another $10 billion on top of its prior plan. Amazon projected $200 billion, mostly for data centers.

The skepticism is not confined to chipmakers, either. Morgan Stanley’s own valuation work has pushed SpaceX’s stock within $15 of a zero value AI verdict for one of its units, and the model layer has its own credibility fight, with ChatGPT, Gemini and Claude trading praise while billions ride on the outcome. Chip stocks are simply the most liquid way to bet against the whole trade at once.

Six Weeks of Chipmaker Pain

The decline did not arrive in a single session. It built over more than six weeks, with each stage adding a new pressure point.

1. June 10: Broadcom and AMD each sink 4% and Nvidia slides 3%, an early tremor that outlets were already calling a deepening chip selloff.
2. June 17: Kevin Warsh runs his first FOMC meeting as chairman, holding rates but signalling the hawkish shift that resets rate expectations for the year.
3. June 23: TrendForce reports SK Hynix is slowing its HBM4 ramp to chase commodity DRAM profit instead.
4. Early July: The SOX drops 11% in a week, Micron falls as much as 13% in a session, Intel slides 21% over seven trading days, and South Korea’s KOSPI triggers circuit breakers.
5. July 25: Fortune previews the next FOMC meeting as a brewing clash between Warsh and colleagues over how far to tighten.
6. July 27: Nvidia and AMD fall again, down 3.46% and 3.39% respectively, and Broadcom trades 21% below its June peak.

Each leg down added a new explanation rather than replacing the last one, which is why the sector’s losses kept compounding instead of stabilizing after the first shock.

What Could Break the Trade From Here

Meta’s own hedge is telling. Reports surfaced on July 1 that the company is preparing Meta Compute, a unit that would resell spare AI training and inference capacity, plus access to its Llama models, to enterprise customers. A company that expected to run permanently short of capacity would not normally build a business to resell the extra.

Three things will decide whether the selloff keeps deepening or stabilizes. SK Hynix’s delayed HBM4 output is due in the third quarter; if it ships close to schedule, the memory bottleneck argument weakens fast. The next FOMC decision will show whether Warsh’s committee actually delivers the hike nine of eighteen members have penciled in, or holds steady as some forecasters expect. And the next round of hyperscaler earnings this autumn will test whether supply constrained cloud businesses stay that way once the current capex bills come due.

For now, the chips still work, the data centers keep filling up, and the stock prices keep falling anyway. That gap closes only when SK Hynix ships on schedule, Warsh’s committee votes, or hyperscalers report what all that spending actually bought.

Frequently Asked Questions

What Caused the AI Chip Stock Selloff in July 2026?

Three forces combined: profit-taking after a roughly 130% twelve month rally in the Philadelphia Semiconductor Index, a memory chip shortage raising costs for chip buyers even as it lifts supplier margins, and a hawkish pivot at the Federal Reserve under new chairman Kevin Warsh that pushed up the rates used to value future earnings. This was the second leg down, not the first; chip stocks had already wobbled in mid-June before the sharper break in early July.

Why Did Samsung’s Stock Fall Despite Record Profit?

Samsung’s 89.4 trillion won operating profit, up more than 1,800% year over year, was driven heavily by the same DRAM price spike squeezing hyperscalers’ budgets elsewhere. Investors read that as a supply shortage temporarily inflating margins rather than durable demand growth, which is part of why the stock fell nearly 7% on the news instead of rallying.

What Is HBM4 and Why Does Its Delay Matter?

HBM4, short for fourth generation high bandwidth memory, stacks multiple memory dies to feed data to AI processors far faster than standard DRAM, and it is the component Nvidia’s and AMD’s top end accelerators depend on most. SK Hynix pushing its ramp from the second quarter to the third means less of that specific chip reaches the market on the original schedule, even though overall memory supply keeps growing.

Will the Federal Reserve Raise Interest Rates in 2026?

It remains unresolved. Nine of the eighteen Federal Open Market Committee members penciled in at least one 2026 rate hike after Warsh’s June meeting, with inflation running near 4.2% year over year. The Motley Fool’s investing desk has nevertheless predicted the committee holds steady through year end, arguing Warsh’s hawkish rhetoric targets inflation expectations more than an actual near term hike.

Are Nvidia and AMD Still Growing Despite the Selloff?

Stock prices and business performance have diverged. Hyperscalers, which are Nvidia’s and AMD’s biggest customers, raised their combined 2026 capex plans to $725 billion, up 77% from last year, and describe demand as outstripping supply. Nvidia shares remain up sharply over the past twelve months even after dropping 7.1% in a single week, which is why several analysts describe the move as profit-taking rather than a demand problem.

Disclaimer: This article is for informational purposes only and is not investment advice. Chip and technology stocks carry significant price volatility, and readers should consult a licensed financial adviser before making decisions; figures above are accurate as of publication on July 28, 2026.