Why AI and Semis Are Selling Off: The Real Reasons Explained
Nine drivers. Ranked by how much they matter.
The screens were red this week and the headlines said "AI selloff."
I think they were looking at the wrong market.
This wasn’t an AI selloff. It was a financing selloff.
The distinction matters.
Data centers are still being built. Memory is still in short supply. Power is still the biggest constraint.
What changed was the market's willingness to fund AI at any cost.
The AI trade quietly split into two very different businesses.
One depends on cheap capital. The other depends on physical bottlenecks.
The first sold off. The second got cheaper.
Everything that happened this week makes sense once you understand that distinction.
Here’s what actually happened.
Nine stories hit the market.
1. The Circular Financing Structure Got Exposed
Nvidia is reportedly in talks to back $250 billion in debt so OpenAI can build a 10 GW AI campus in Ohio, plus another $350 billion in financing for GPU purchases.
That matters because OpenAI cannot borrow that amount on its own.
Nvidia’s balance sheet effectively becomes the credit support behind the financing.
In simple terms:
Banks weren’t willing to lend to OpenAI on its own.
Nvidia may use its own financial strength to help OpenAI borrow the money.
OpenAI then uses that money to buy Nvidia GPUs.
Michael Burry increased his short position and posted, “around and around we go.” Nvidia fell nearly 5% on Monday.
AI Demand may still be real.
But when suppliers help finance their customers, investors begin asking a different question.
Is demand accelerating?
Or is financing making it possible?
The market didn’t answer that question.
It simply assigned a lower multiple to the uncertainty.
2. Credit Cracked Before Equities Did
The first warning didn’t come from the stock market.
It came from the bond market. Most retail investors are not looking at it.
Oracle's five-year credit default swap spread rose to its highest level in nearly 18 years, and S&P downgraded the company to BBB-, one notch above junk.
AI-related debt issuance is projected to reach roughly $570 billion this year, while demand for hyperscaler bonds has weakened significantly.
The message was clear.
Investors became less willing to finance AI infrastructure at the same pace and price as before.
Stocks reacted after credit markets did.
3. Alphabet Showed Everyone The New Scoreboard
Alphabet reported strong results.
Revenue grew 24%. Cloud revenue grew 82%. Cloud backlog reached $514 billion.
The stock still fell about 5%.
Investors focused on higher capital spending, negative quarterly free cash flow, and increased spending guidance.
Strong revenue was no longer enough.
The market wanted evidence that AI spending would generate acceptable returns.
4. Rates are Moving The Wrong Direction
The Fed is not cutting.
It is debating hiking.
Inflation remains at 3.7%, well above the Fed’s 2% target.
Markets have priced a meaningful chance of another rate hike this year.
Higher rates increase the cost of capital.
That matters because every AI valuation depends on future cash flows.
Brent crude fell 8.7% on easing geopolitical tensions.
Lower energy prices could reduce inflation pressure and limit further rate hikes.
5. China Added a New Memory Competitor
CXMT surged 466% in its Shanghai debut, raising about $8.6 billion to expand DRAM production.
The company is now the world’s fourth-largest DRAM manufacturer with roughly 7.7% global market share.
The market interpreted the IPO as a sign that future memory supply could increase.
That weighed on memory stocks despite an industry that remains supply constrained.
6. China Made Progress in Lithography
China has started producing domestic immersion DUV lithography machines, with deliveries expected to SMIC, Hua Hong, and CXMT.
But scale remains limited.
Roughly five machines are expected in 2026, compared with ASML’s capacity for about 130 immersion DUV systems.
The impact on earnings is small today.
The impact on future expectations is not.
7. Korea Became a Forced-Selling Event
The Kospi fell 10.84%.
Samsung dropped 13.4%.
SK Hynix fell more than 14.7%.
JPMorgan estimated Korea-linked leveraged ETF assets fell from $50 billion to $26 billion, while foreign investors sold $110 billion, with 90% concentrated in semiconductors.
Regulators also raised the minimum cash requirement for single-stock leveraged ETFs.
This wasn’t fundamentals alone.
It was forced selling.
8. Kimi K3 Release
Moonshot AI released Kimi K3, an open-weight model claiming frontier-level performance at a fraction of the cost.
Barclays also warned that enthusiasm around AI capital spending is beginning to cool.
The concern wasn’t AI adoption.
It was whether cheaper AI models reduce future infrastructure spending.
History suggests the opposite.
Jevons paradox says that when a technology becomes cheaper and more efficient, total demand often increases rather than decreases.
9. The Biggest Week for AI Is Still Ahead
Microsoft, Meta, Apple, and Amazon report earnings this week.
The Fed also announces its latest policy decision on Wednesday.
Markets will be watching one thing.
Will companies keep increasing AI spending?
Or will they begin demanding higher returns on every dollar invested?
The answer will shape the next leg of the AI trade.
What Changed. What Didn’t.
What Changed
Three things changed this week.
1. Capital became more expensive.
Credit spreads widened, debt financing became harder, and the market raised the hurdle for AI investment.
2. China became a more credible competitor.
CXMT’s IPO and progress in domestic lithography reminded investors that future supply will not come from the U.S., Taiwan, and South Korea alone.
3. The market changed how it values AI.
Revenue growth is no longer enough.
Investors now want evidence that AI spending will generate attractive returns.
What Didn’t
The physical constraints behind AI did not change.
93% of leading-edge DRAM capacity remains allocated to HBM, with no meaningful supply relief expected until 2028.
Alphabet’s cloud backlog still stands at $514 billion.
Power remains the largest infrastructure bottleneck.
Advanced packaging capacity remains constrained.
Optical bandwidth demand continues to grow with every new AI cluster.
The financing story changed.
The physics did not.
That distinction is the opportunity.
Final Thoughts
This wasn’t an AI selloff.
It was a repricing of AI financing.
Higher funding costs, tighter credit, and rising return expectations changed how the market values AI investments.
They did not change what AI requires.
Data centers still need power.
AI models still need memory.
Chips still need advanced packaging.
Clusters still need faster networking.
Those constraints remain.
Markets can reprice valuations overnight.
They cannot reprice physical bottlenecks.
That’s where I’m focused.
Disclosure: For informational and educational purposes only. Not investment advice. This reflects my personal opinions and may change without notice. I may hold positions in securities mentioned. Always do your own research before investing.
It’s interesting how quickly valuations can change while the underlying demand for power, memory, and data centers stays the same.
Excellent insights. Something’s gotta give. My guess, the price of tokens is going to go up not down.