Supreme intelligence so smart... The reason operators keep A100s racked at all, given that per-MW economics favor Blackwell roughly 2:1 even after the capital charge, is that they're not competing for the same megawatt. A 1kW liquid-cooled NVL72 rack physically can't go into the air-cooled legacy halls where Ampere lives. The A100 tail exists in the shadow of the power constraint — it monetizes stranded, low-density DC capacity that has no alternative use. Which means the 2029 A100 contract isn't evidence about GPU useful life at all; it's evidence about datacenter power scarcity.

Used our Gigawattonomics model, and the ROIC on investments in Grace Blackwell systems (assuming 10 yr life, which is a safe assumption at this point) assuming 75% utilization and $7 average (its much higher) is--260% That's cost to build, run, all power economics included.

Yesterday, @intel completed our planned capital raise of $20B in one of the largest follow-on equity offerings on record. With strong support from targeted long-term sovereign funds, fundamental growth and institutional investors, specialty hedge funds and retail investors, the round was oversubscribed >5X our initial goal. Many thanks to our underwriters JP Morgan, Goldman Sachs, Morgan Stanley and Citigroup for their excellent support. With this additional capital, Intel is now well positioned to meet the tremendous growth opportunity ahead of us in advanced node wafer manufacturing, advanced packaging and the massive CPU demand. We remain laser focused on execution across all our businesses. Along with the full support of our Board, my leadership team and I are fully committed to delivering strong returns to our shareholders in the coming years and to building a new and vibrant Intel.

New feature on @DiligenceStack. We teased this but will do it quarterly. Full review of our thesis, what changed, where we hold, raise, or shift a thesis, and more. The Changelog Q2.26

Lots of interesting conversations with investors the last few days talking about what it means, and the implications of selling intelligence when in no prior point in technology history has software been positioned or valued in such a way.

Via our Giagawattonomics model, we can confirm compute returns attractive economics. Whether any current generation (perhaps its the generic just compute is Jensen's point) is an investible asset will be fun to debate.

On the potential of an NVIDIA HBM downspec (option) Our model suggests that a broad HBM downspec across NVIDIA’s Rubin portfolio could support roughly 30% to 50% more GPU shipments from the same available pool of HBM bits, assuming CoWoS and compute-die supply keep pace. If those GPUs are deployed, power becomes the next question. A 30% to 50% uplift on roughly 7 million Rubin units across 2027–2028 would add 2.1–3.5 million GPUs. At 2kW per accelerator, that implies roughly 4–7GW of incremental ex-China chip-level power demand across the two years, or about 2–3.5GW annually if deployment is evenly split. Current assumption right now in accelerator forecast, is we need to energize (global ex-China) 26 GW in 2027 and 32-35GW in 2028.

This comment from Sumit Sadana, exactly why I made this chart. Since most commentary on memory has not covered this industry for very long. These deals are unlike any in the past in contract strength, durability, length, and pre payments. "we call them strategic customer agreements first and foremost because they are very, very different than the historical LTAs or long-term agreements. First of all, those LTAs were somewhat of a misnomer because there was nothing long-term about them; they were just 12-month agreements for the next calendar year, another important difference is that you know those LTAs had no binding terms in them, right? It was more of a handshake kind of an understanding with customers about ensuring that there is good level of supply chain planning that we do and documenting it so that you know they are putting more thought into what kind of supply they are intending to purchase from us. That was the LTA time. So when we saw some of this AI demand for a multi-year time frame become so urgent for our customers, there was an increased level of anxiety at our customers to secure the supply, and we came up with this proposal and idea of strategic customer agreements. And we have pioneered this concept in our industry, and we have been the first ones to work on it with our customers. And we have also, we believe signed the most number of SCAs across our industry since our earnings. We have signed up more SCA agreements with our customers, and you know we really see these agreements as being transformative of the business model that we are used to. For starters, these SCAs cover a long time horizon. Some customers that are smaller, like automotive customers, have mostly three-year SCAs. But the SEAs that cover the overwhelming amount of the revenue under SCA is going to be you know five-year type of terms through the end of calendar 2030, and so number one, these provide quite a long-term visibility to us. Second, these SCAs are binding commitments on purchases of these volumes by year, by customer, and these are take-or-pay agreements, and there are no contractual outs for our customers from these agreements. These are very much, very very different terms, very stringent and binding terms on the purchases. These are backed up by tremendous amounts of upfront cash and cash like commitments, like letters of credit. But the overwhelming amount of the commitment is upfront cash that we are going to have on our balance sheet."

$INTC Intel’s $15B Share Sale Multiple Times Oversubscribed. Intel’s equity raise, which is its first public stock sale since 1971, is drawing huge demand, according to Bloomberg. Dilution hit the shares today, but the oversubscription shows the market is still hungry for Intel.

This quote is telling from the $MU discussion at KeyBanc with Sumit Sadana: "So we have visibility now and deep engagements with our customers on the engineering side that stretches out to roadmaps at our customers that are beyond 2030 time frame. I mean, when we started, you know, in the memory industry many years ago, it was rare to even understand, you know, what next year's products are going to be at our customers. And now we have, I would argue, even longer term visibility than most in the semiconductor industry have across the different semiconductor markets, and we have these deep engagements"

This section is addressing the circular point, something many people miss. The customer is every company on the planet and any employee who uses a compute device for work. We are early in the diffusion of enterprise adoption. We don't have nearly enough compute.

Exactly the angle, point I made in this security report. Frontier labs, are already being used in cyber security, and will be well positioned for this extremely valuable enterprise software use case.