Everyone ignored AMD. Everyone buried Intel. Then the AI buildout got too big for Nvidia to contain alone — and both stocks made investors look smart. That same setup may be playing out right now, in two stocks most people have never looked at.
Let's rewind to 2022.
Nvidia was the only AI chip story anyone wanted to tell. Jensen Huang was on magazine covers. NVDA was up 800% in two years.
AMD was the competitor everyone had written off — too slow, too behind, too late to the GPU party.
Intel? Forget it. The analysts called it a dying legacy company.
Then something simple happened.
The AI buildout kept growing. Data centers kept ordering. The hyperscalers — Google, Microsoft, Amazon, Meta — needed more chips than Nvidia alone could supply. Demand didn't wait for a single company to catch up. It overwhelmed the market and forced capital to find the next best option.
AMD surged 50% year-to-date. Intel posted its best single-day gain since 1987. Not because they suddenly became better than Nvidia.
The reason is simple.
When demand gets big enough, the entire food chain gets fed.
That is the most important lesson in semiconductor investing. And right now, it's happening again.
13 STRAIGHT DAYS. A 32-YEAR RECORD.
The Philadelphia Semiconductor Index just did something it has never done in its 32-year history. It closed higher for 18 consecutive trading days. The previous record was 15 days — set back in 2014 — and that entire run added just 7% to the index.
This one? The index rocketed 42% higher.
Credo Technology surged 115%. Astera Labs gained 93%. Marvell Technology — a name most retail investors have never heard of — ripped nearly 80% in the streak alone.
And Intel? The old, written-off, legacy chip company? It gained 62% during the streak and then added another 24% in a single session after its earnings beat.
But here's the question nobody is asking loudly enough yet:
If Nvidia, Intel, and AMD have already run — where does the demand go next?
ELON MUSK JUST CHANGED THE ENTIRE MAP.
On March 21, 2026, Elon Musk walked onto a stage in Austin and announced Terafab.
He wasn't subtle about it. "We either build the Terafab, or we don't have the chips," he said. "And we need the chips, so we build the Terafab." The project — a joint venture between Tesla, SpaceX, and xAI — is targeting one terawatt of annual computing output.
The entire United States currently produces about half that.
But here's the number that should make every semiconductor investor stop scrolling:
Terafab is targeting production capacity equal to roughly 70% to 90% of what TSMC currently produces for the entire world.
Think about what that means.
Taiwan Semiconductor is the most dominant chip foundry in history. It makes chips for Apple, Nvidia, AMD, Broadcom — virtually every advanced chip in every device you own runs through TSMC's fabs.
And Musk is building something in Austin that, at full scale, approaches that output.
That doesn't just affect Tesla.
It affects the entire semiconductor supply chain. Every company that builds fabs needs photomasks, purification chemicals, optical inspection equipment, fiber connectivity, and custom silicon design partners. Terafab needs all of it — and it needs it in Texas, not Taiwan.
Intel signed on as the manufacturing partner on April 7. Ground is already broken at Giga Texas. The research fab alone is a $3 billion commitment. Analysts at Bernstein put the full-scale cost somewhere between $5 trillion and $13 trillion.
When demand at that scale enters a market, it doesn't just feed the dominant players. It floods the entire ecosystem. The photomask suppliers. The fiber makers. The custom chip designers.
The companies that were second or third in line suddenly have more orders than they can fill.
Sound familiar? It should. It's exactly what happened to AMD and Intel when Nvidia's buildout got too big for one company to contain.
THE NEXT IN LINE
So where can the smart investors go now?
Smart ones would go into the companies that are positioned exactly where AMD was in 2022. Real businesses. Real revenue. Real demand. Just not yet fully priced.
Here are two.
Marvell Technology (MRVL)
Most investors still think of Marvell as a networking chip company. That's like calling Amazon a bookstore in 2005.
Marvell has quietly become the second-largest designer of custom AI accelerators on the planet — the chips that Google, Amazon, and Microsoft are building to replace Nvidia GPUs at scale.
They hold roughly 15% of the ASIC market. Only Broadcom is ahead of them.
Here's what changed everything this month.
Google entered talks with Marvell to co-develop two next-generation AI chips — a new Tensor Processing Unit and a memory processing unit designed specifically for efficient AI inference. The news added $5 billion in market cap in a single session. Then Nvidia — the company whose products Marvell's customers are trying to replace — turned around and invested $2 billion in Marvell directly.
Read that again.
Nvidia is investing in the company helping its customers build around it.
That's not competition.
That's confirmation that Marvell is now so embedded in the AI infrastructure ecosystem that even the dominant player can't afford to leave it out.
The numbers back it up.
Fiscal 2026 revenue hit a record $8.2 billion — up 42% year over year. Non-GAAP earnings per share of $2.84 were up 81%. Data center revenue is now 74% of the business. Management has guided to $11 billion next year and $15 billion by fiscal 2028. That's nearly doubling revenue in two years.
Custom AI chip sales are projected to grow 45% industry-wide in 2026.
GPU shipments? 16%.
The gap between those two numbers is where Marvell lives. As hyperscalers build custom silicon to reduce their dependence on Nvidia, every dollar they spend flows through Marvell's design shop.
Is it cheap? No. At 43x forward earnings, Marvell is priced for execution. But AMD wasn't cheap when it broke out either. The question isn't valuation in isolation. It's whether the earnings growth is real enough to grow into that multiple.
At 42% revenue growth with $15 billion in sight, the answer looks like yes.
Corning Incorporated (GLW)
Here is a company that has been making glass for 175 years. It made glass for Thomas Edison's lightbulb. It made the screen on your iPhone.
And right now, it is the single most important fiber optic supplier in the world — at the exact moment when AI data centers need more fiber than at any point in history.
Nobody saw Corning coming. That's what makes it interesting.
CEO Wendell Weeks started planning for AI-driven fiber demand more than five years ago, before most investors had heard of ChatGPT.
While the semiconductor world was chasing GPU stocks, Corning was quietly developing new products and building manufacturing capacity.
When the hyperscalers started pouring trillions into data centers, Corning was the only company ready to supply at scale.
Then the deals started arriving.
In January, Meta signed a multiyear agreement worth up to $6 billion for Corning optical fiber to power its AI data centers. The stock surged 18% in a single session — its first all-time high since the dot-com era.
Then, on this morning's earnings call, CEO Wendell Weeks dropped a line that should have stopped every investor cold:
"We now have concluded two more large long-term agreements with hyperscale customers. They are each similar in size and duration to the Meta agreement." — Wendell Weeks, Corning CEO — Q1 2026 Earnings Call, April 29, 2026.
Three deals.
Each is approximately $6 billion. Two customers not yet named. That is close to $18 billion in contracted fiber revenue that didn't exist eighteen months ago.
The financials are real. Q1 2026 sales grew 18% to $4.35 billion. EPS grew 30% to $0.70. Operating margins hit 20.2% — a milestone the company had targeted for end of 2027, reached a full year early.
The Optical Communications segment grew 36% year over year. Corning's stock is up 74% year-to-date.
And it just launched a product called Multicore Fiber — a cable that packs four times the data capacity into the same physical strand. That means a data center operator can get the same performance with 75% fewer cables.
In a world where every square foot of data center floor space costs a fortune, that is not a minor product update.
That is a pricing power event.
THE OPPORTUNITY IN FRONT OF INVESTORS
Every great trade has a simple thesis at its core.
The AMD thesis was simple: demand is too big for one company.
The thesis here is equally simple. Musk is building a chip complex that rivals TSMC. The semiconductor sector just posted its longest winning streak in 32 years. And the rotation out of the single dominant name — Nvidia — is already underway, exactly the way it was for AMD two years ago.
Marvell is the company building the custom chips that replace Nvidia GPUs at hyperscale. Corning is the company threading the fiber that holds the entire AI buildout together.
Both were ignored while Nvidia took all the oxygen. Both could now be getting their moment.