DanielRomero

Just so you know:
The data center business model remains one of the most misunderstood
> Wall Street has mostly figured it out, but remains skeptical
> Retail remains largely clueless
This type of ignorance is what keeps prices attractive for longer

Is Anthropic the largest $AMD deal?
The answer is no
> OpenAI: 6GW
> Meta: 6GW
> Anthropic: Up to 2GW
OpenAI and Meta have the largest overall commitments by far, with 1GW each committed to MI450 and the rest covering future generations
Anthropic has the largest explicit MI450-only commitment and receives an AMD investment instead of a warrant, which is much more favorable for AMD
It is not necessarily a smaller deal, just one covering a shorter time frame, while the others represent longer-term relationships

JUST IN
$AMD and Anthropic announced a strategic partnership to deploy up to 2GW of AMD Instinct MI450-series GPUs through AMD’s Helios rack-scale platform
The first gigawatt is expected to begin deployment in the first half of 2027, building on Anthropic’s existing use of MI355X GPUs
The companies will also collaborate on software, using Claude to optimize workloads for AMD hardware and accelerate ROCm development. AMD will adopt Claude across its engineering and product teams
AMD also committed to make a strategic equity investment of up to $5 billion in Anthropic

Remember when Google released Gemini 3 Pro last year and everyone thought they were actually going to win the LLM race?

> Leopold Aschenbrenner’s fund owns 5.6% of $NBIS
> $NVDA owns 9.3%
Smart money is clearly seeing something special here
On top of that, CEO Arkady Volozh owns 11.3% of the shares, giving him huge skin in the game as well

$AMD MI455X is now confirmed to be selected by:
> OpenAI
> Meta
> Microsoft
> Oracle
Plus possibly Anthropic, which seems to be an open secret
Clearly, it brings something to the table, because these companies could use ASICs or $NVDA, yet they still want to bet on $AMD

$NVDA is an investor in Revolut
Interesting that they’d go as far as investing in neobanks rather than simply buying back stock
That being said, Revolut is the best bank in Europe

Samsung P5 Phase 1 Nears $6.8 Billion in Equipment Orders
Phase 1 could begin operating in 2027
P5 will contain six cleanrooms across three floors and is designed to deliver more than 50% greater capacity than Samsung’s existing fabs
It will produce advanced DRAM and HBM, with foundry capacity also under consideration
Samsung holds approximately $99 billion in cash and short-term investments, $80 billion in net cash, and generated around $27 billion in operating cash flow in the first quarter alone
The company is in a prime position to deploy massive capex and take advantage of the current pri

$KOSPI -4% to start the week
The KOSPI selling off after all the news we’ve had makes absolutely 0 sense
Much less such an aggressive sell-off

You don’t get 30%+ drawdowns in AI companies during a bubble
It just doesn’t happen
> Either you think the bubble has already peaked and we’re experiencing the pop,
> Or you simply have to accept that we aren’t in one

Investors always wish they had bought more when the market is at all-time highs
→ Yet when a drawdown comes, there’s panic instead of an urge to accumulate
__________________________
The reason is that people see investing as gambling rather than business ownership
Irrationally, they get the same feeling as when they bet on a football team and it’s down 2–0 at halftime
You get the shivers
Luckily, investing in stocks has nothing to do with that
__________________________
You’re acquiring an equity stake in what hopefully is, or will become, a cash-generating machine
→ There’s no fixed deadline

The only reason memory is trading lower is that the market is still in denial about the pricing power
“Hyperscalers won’t spend 40% of their capex on memory.”
“Hyperscalers won’t accept memory suppliers earning 85% gross margins.”
It’s cognitive dissonance at this point
There isn’t even an argument or any supply-and-demand analysis
It’s just denial of reality

Open-source models improving gives tremendous power to neoclouds
This is what many people don’t understand
> $NBIS, for example, only offers open-source models through its managed inference platform
> It allows enterprises to fine-tune and deploy open models for internal workloads much more cheaply than using the OpenAI or Anthropic APIs
As open models improve, demand for this infrastructure increases
The GPU supply remains the same. The number of valuable workloads competing for that supply rises
That increases utilization, strengthens pricing power and gives neoclouds, especially leading one