$AMD
Lisa Su says they collaborated for 2 years with OpenAI before their industry-shaking deal

$AMD Lisa Su gets asked if she sees a bubble in AI:
"We don't see a bubble.
What we do see is very well-capitalized companies, companies that have significant resources using those resources at this point in time because it's such a special point in time in terms of AI learning and AI capabilities."

$AMD
Lisa Su at the UBS Conference:
"We are 2 years into a 10-year supercycle."

$AMD Lisa Su gives her opinion on $GOOG TPUs at the UBS Conference:
"Google has done a good job with the TPU architecture over the years. But it is a more purpose-built design. It lacks the programmability, model flexibility, and balanced training and inference capabilities that GPUs offer. GPUs combine a highly parallel architecture with high programmability, which enables fast innovation.
From our perspective, there is room for all types of accelerators. However, over the next five years, GPUs should remain the clear majority of the market because we are still early in the cycle, and softwar

Lisa Su responded to the question of whether $AMD is going to build ASICs:
"From our perspective, differentiation comes when we combine our intellectual property with our customers’ intellectual property and know-how to create outcomes where one plus one is greater than three. We excel at deep customer partnerships, and that has been true for more than a decade. In addition to our standard products across CPUs, GPUs, and FPGAs, we have built a strong semi-custom business. I don’t view this as an ASIC business. With ASICs, you design any chip a customer asks for. That is not where we add the mo

$AMD's Lisa Su gives her opiniojn on $GOOG's TPUs in the UBS Conference:
"Google has done a good job with the TPU architecture over the years. But it is a more purpose-built design. It lacks the programmability, model flexibility, and balanced training and inference capabilities that GPUs offer. GPUs combine a highly parallel architecture with high programmability, which enables fast innovation.
From our perspective, there is room for all types of accelerators. However, over the next five years, GPUs should remain the clear majority of the market because we are still early in the cycle, and so

Valuations aren’t anywhere near the levels of the dot-com bubble in the public markets
However, nowadays many companies, especially unprofitable tech names, choose to stay private
I wonder how this graph would look if the multiples of the 100 largest private companies were included as well

Although it seems otherwise, data center growth has declined from 2024 to 2025

$NBIS hasn't disclosed:
• The exact MWs or GPUs delivered to $MSFT
• How much they are paying to rent any of their data centers
They have not disclosed the terms of a single leasing contract, which every other company like $MSFT, $IREN, $AMZN, $CIFR, and $WULF does disclose
• Electricity rates in any of their data centers
• Which data center the $META deal is tied to, and the GPUs or MWs to deliver
• What GPUs they are buying, when they are buying them, and at what price
And there are probably more details missing
Why?
It makes the job of retail investors very hard. I am sure they share

The market is turning on $NVDA, the largest company on Earth leading the AI and robotics race, because of some rumors about Google selling hardware that was already accessible through their cloud
And it’s turning on OpenAI because Google delivered a good model, as if the model leaderboard didn’t change every month after each update
Two months from now OpenAI could launch a killer model, and $NVDA could say they’re booked for 80% of Rubin capacity already, and the market would love them again
It’s all a constant overreaction followed by another overreaction

One month OpenAI is going to destroy $GOOG
The next, $GOOG is going to destroy OpenAI
Everything has to go either to infinity or to zero
Investors can be so emotional and shortsighted

$GOOG will sell TPU capacity
$NVDA will sell GPUs
$AMD will sell GPUs
$INTC will sell CPUs and foundry capacity
$TSMC will sell foundry capacity
$ASML will sell machines to make chips
$MU will sell memory
Everyone will be happy
Stop thinking in zero-sum games

Has $NBIS disclosed the exact MWs of the $MSFT deal?
I find it surprising how little information was released in the 6-K and the following filings
I still can’t find whether the project is 250 MW IT load, 300 MW gross, or the 350 MW figure shown on DataOne’s site
Anyone have more details?

I asked Nano Banana for the cross-section of a data center
And this is what it gave me on the first try
Very impressive

In the next 10 years:
$NBIS will likely become the AWS of AI with a over 200 billion enterprise value
$AMD will become a multi-trillion market cap company by satisfying the demand for efficient compute
$MU will remain the only US manufacturer of a key component for superintelligence, memory, reaching a multi-trillion market cap
Data center operators will be the most important real estate owners. The future of humanity depends on their infrastructure. That includes $IREN, $CIFR, $WULF, etc.

The MIT sees AI already able to replace 11.7% of the US workforce
It is impossible not to be uber bullish on the AI Cycle
Each day that passes, I am more convinced any dip must be bought
How can you not want to be all in for a revolution this size?
And now robotics are coming

If the $CLSK CEO says they are getting hyperscaler interest in bitcoin mining sites, I can imagine it is only a matter of time before $IREN, $CIFR, $RIOT, $WULF and others strike more deals.
Here is the full commentary:
“I had a conversation, not with a or anybody like that, but with the senior director of site development for a global hyperscaler last night on my way out. He told us their 2026 forecasts are so constrained that they are already evaluating alternative types of builds just to meet demand for 2026. The takeaway was Sandersville and Sealy, because both sites can be energized. Sand

3 industries to be invested in over the next years:
> Memory: $MU and 000660(SK Hynix)
> Optics: $COHR $LITE $AVGO
> Data center operators: $IREN $NBIS $CIFR $GLXY $BITF $WULF
If you want to take advantage of the AI supercycle, these are a must, as gigawatts of data centers are being deployed and both scale up and scale out architectures are needed for these massive deployments.