Well, on one hand, chip stocks just had their worst week in about a year. The SMH, the major semiconductor ETF, fell 8.9%. The Philadelphia Semiconductor Index (SOX) dropped nearly 10%.
And on the other hand? Within just a few days, companies announced hundreds of billions of dollars in new investments. New systems, new factories, new chips.
So the market is selling while the companies themselves are buying.
Let's break it all down.
The biggest story of the week is Helios.
It's AMD's first rack scale AI system, and the first genuine challenger to Nvidia's Grace Blackwell and Vera Rubin systems in years.
The name isn't accidental either. Helios, the Greek god who pulls the sun across the sky with four horses. Those four horses represent the four core technologies AMD now builds itself: GPUs, CPUs, networking, and software.
Who's buying it? Microsoft.
On Monday, Satya Nadella announced that Helios will be deployed across Azure data centers. It joins a customer list that already includes Meta, OpenAI, Oracle, and India's Tata Consultancy Services. AMD shares climbed more than 1.5%.
"So... is it actually better?" you might be wondering.
AMD's head of data center solutions, Forrest Norrod, says the company's goal is the "lowest cost per token." CEO Lisa Su highlights significant advantages in inference performance and memory capacity.
Analysts at Futurum estimate Helios costs between $5 billion and $5.5 billion per deployment, compared with roughly $3.5 billion to $4 billion for Nvidia's Vera Rubin platform.
There is one major catch, though.
Nvidia still controls more than 95% of the data center GPU market, while AMD holds just 4.5%.
And as Counterpoint analyst Neil Shah points out, the hardware may be competitive, but Nvidia's secret weapon is CUDA.
Still, Futurum's Daniel Newman believes AMD could realistically grow its market share to 20% or even 25%.
We're talking about hundreds of billions of dollars in potential revenue.
Now let's move to the second story, which is just as remarkable.
TSMC, the company that manufactures chips for nearly everyone, has announced another $100 billion investment in Arizona.
The total commitment now stands at $265 billion.
Yes, you read that correctly.
Chief Financial Officer Wendell Huang summed it up perfectly in an interview with CNBC:
"We see a multi year, structural demand trend, and we don't intend to leave food on the table for anyone else."
At the same time, the company raised its annual capital spending to between $60 billion and $64 billion.
TSMC is converting production lines from 5 nanometers to 3 nanometers, while its first Arizona fab is already producing 4 nanometer chips.
Meanwhile, 2 nanometer technology is becoming the company's next major growth engine.
What's particularly interesting is that building semiconductor fabs in the United States costs four to five times more than building them in Taiwan.
They know that.
They're doing it anyway.
So what happened to the stock?
It fell 7.29% on Friday.
Even so, it's still up roughly 26% since the beginning of the year.
Huang's response was almost philosophical.
"We don't control the markets. What we do control are the fundamentals of our business."
This is where things get even more interesting.
According to The Information, Google is developing a new AI chip internally known as "Frozen v2," and Alphabet shares gained 3% following the report.
So what exactly is it?
In simple terms, it's a chip that permanently hardwires parts of Gemini's architecture directly into the silicon.
That means when you ask it a question, it requires fewer computations and much less data movement to generate an answer.
The result?
Google engineers estimate six to ten times more tokens per unit of energy compared with today's TPUs.
The target launch is 2028.
"But why is Google doing this?"
Because it has a serious problem.
Its internal shortage of computing capacity has become so severe that Google Cloud has reportedly been forced to turn customers away.
Last month, Google also agreed to pay SpaceX nearly $1 billion per month simply to help bridge that capacity gap.
Of course, nothing comes without trade offs.
The chip will only deliver those benefits if Google continues using the same underlying architecture.
At the same time, the next version of Gemini Pro has reportedly been delayed, researchers are leaving for competitors, and Chinese AI models now account for roughly 45% of token usage among U.S. companies.
If everyone is spending aggressively, why are the stocks falling?
And more importantly, is this a buying opportunity?
Wall Street analysts are split.
Mizuho says there's still "plenty of fuel left in the tank," arguing that AI infrastructure and power investments extend well beyond 2028 and 2029.
Morgan Stanley calls the current pullback an "attractive entry point," with Nvidia and Broadcom offering particularly compelling valuations.
J.P. Morgan agrees, noting that meaningful new chip supply is unlikely to arrive before 2028.
But there's another side to the story.
Evercore believes the SOX's 20% decline over the past four weeks is simply a correction within a longer term cycle.
However, they also expect another two to three weeks of weakness, with an additional 10% to 15% downside before the correction is complete.
Historically, once that happens, the average recovery has been about 36% over the following 20 weeks.
So what is everyone really saying?
Almost everyone agrees the sector is likely to move higher.
They just can't agree on when.
