From an interview with an exec right in the middle of the data center buildout, selling power contracts to the hypersalers, neoclouds, and shell builders. The demand ratio for GPUs is 3 to 1, projected to go to 4 to 1... I thought there was no demand 🤣

Google $GOOG is acquiring data center and energy infrastructure startup Intersect for $4.75B, with the deal expected to close in 1H '26. Intersect's operations will remain separate from Google, and the acquisition will help bring data center generation capacity online faster.

The GPU as a service market is projected to rise more than 8X over the next decade, from ~$30 billion in 2025 to $251 billion by 2034. $NVDA $CRWV $NBIS $IREN

@Umbisam @MarcoRamireZuno @Agrippa_Inv @RHouseResearch @danroberts0101 @Umbisam hits it perfectly imo. And I’ll just build on it slightly, because this is exactly why I’m skeptical of using today’s standard depreciation cycles as a guide for residual value. The 5–6 year useful-life assumption and even the 4-year version is a product of the last 20

Unlevered IRR VS Levered IRR Let’s clear a few things up, because lately on X I keep seeing terms like ROE, ROIC, unlevered IRR, and levered IRR being thrown around interchangeably especially in the data center conversation. And most people immediately jump to levered IRR as the

@Agrippa_Inv @RHouseResearch @danroberts0101 I don’t think Agrippa’s method / approach is accurate. He should use NPV to better understand why, but happy to provide an example via IRR. To illustrate why ‘averaging’ subsequent periods is completely misleading, consider a simplified example with two phases: - Period 1

Capacity-Adjusted Market Share: Reframing the TPU vs Nvidia Narrative I wanted to share some important narrative nuance to the most recent TPU headlines. They suggest $GOOGL is preparing to 7× TPU volumes, which some people are quick to frame as big breakdown in Nvidia’s share

$AMD $GOOGL $METAq Okay no im gonna say something and it will be hard for some people to understand: what you are seeing today is the past and not the future. Sounds crazy right😂. Let me explain: I’m in $NVDA since 2018. And what happens today (see @Beth_Kindig post) is

Wow. Classic Jensen style, he ended the Nvidia vs. custom ASIC competition for good. 🫡 The level of confidence with which he explains. 🎯 He was answering to UBS research analyst question on how custom ASICs will affect NVIDIA or how they are going to compete with custom ASIC.

Apollo CEO Marc Rowan on the data center financing: “Hyperscalers are pretty safe; intermediaries that are taking up the debt and balance sheet risk not so much.” Neoclouds are leveraging themselves instead of hyperscalers, but they don’t have the cash flows from other

There will be a moment when companies are able to monetize AI to where capex costs are absorbed. In terms of the ability to monetize AI, this is being proven as we speak with OpenAI exhibiting the fastest run rate from $0 to $20 billion in all of tech’s history. I joined

Capacity-Adjusted Market Share: Reframing the TPU vs Nvidia Narrative I wanted to share some important narrative nuance to the most recent TPU headlines. They suggest $GOOGL is preparing to 7× TPU volumes, which some people are quick to frame as big breakdown in Nvidia’s share

According to Goldman Sachs, after 4 years of AI adoption, AMD, AVGO, and MRVL together are competing for only about 20% of the market. NVDA will still dominate with roughly an 80% market share.

For this week's post, I would like to go over Sofwave Medical Ltd $SOFW.TA, which is a newly profitable "razor-and-blade" compounder trading at a discount. They are the go-to for non-invasive skin tightening, capturing the 'Ozempic Face' tailwind while immune to the FDA warnings

When asked if customers would take on more of the XPU system design over time, $AVGO's CEO Hock Tan didn't give a great answer (see screenshots). Here's an angle he could have taken: "We are aware of the discussion around customer-owned tooling and insourcing. That has always