Fascinating analysis, Dr. Cutress — especially how you frame compute as a financial asset class. I’m curious about the human and geopolitical layer of this deal: with such massive U.S.-based power requirements, TSMC dependencies, and ongoing export controls shaping AI hardware flow, how do you see those forces influencing AMD’s ability to execute this six-gigawatt vision? Does the “financial alignment” model hold if geopolitics or grid infrastructure slow real deployment?
That concern about running out of energy might not be as much of an issue in China, which is on a tear adding enormous power generating capacity from solar and wind every month, but that isn't the case in the US. The preference for using nuclear power here also means that it'll take years to have those reactors online. In the meantime, China adds more wind turbines, solar energy farms and megawatts of battery storage almost weekly. And all at lower costs, too.
Great analysis. (And I'm pained that it took me a month to find it)
It looks like the deals are tying the contracts to the economic interests of all the parties. The market's reaction confirms this.
What's your take on how these deals impact financial reporting? My first intuition is there will need to be a lot of adjustments to do industry comps between AI and semiconductor companies.
This is an incredibly thorough breakdown of a transformative deal structure. The warrant mechanism is particularly clever - by setting strike prices at $146, $157, and $168, AMD essentially gets paid to build capacity while OpenAI gets price protection against future cost escalation. What really strikes me is the inverted risk profile compared to Nvidia's model. Nvidia took equity risk by investing directly in customers, but AMD is getting guarenteed revenue while OpenAI assumes the utilization risk. The 6 gigawatt target is staggering - that's roughly 1.5% of total US electricity generation just for AI inference. Your point about power infrastructure being the real bottleneck is spot on. The chip supply might materialize, but can the grid capacity?
Fascinating analysis, Dr. Cutress — especially how you frame compute as a financial asset class. I’m curious about the human and geopolitical layer of this deal: with such massive U.S.-based power requirements, TSMC dependencies, and ongoing export controls shaping AI hardware flow, how do you see those forces influencing AMD’s ability to execute this six-gigawatt vision? Does the “financial alignment” model hold if geopolitics or grid infrastructure slow real deployment?
that was so quick, great article. were you aware of the structure of the real beforehand?
That concern about running out of energy might not be as much of an issue in China, which is on a tear adding enormous power generating capacity from solar and wind every month, but that isn't the case in the US. The preference for using nuclear power here also means that it'll take years to have those reactors online. In the meantime, China adds more wind turbines, solar energy farms and megawatts of battery storage almost weekly. And all at lower costs, too.
Who do you think did the better deal? AMD or Nvidia?
AMD warrant structure vs Nvidia equity investment.
Great analysis. (And I'm pained that it took me a month to find it)
It looks like the deals are tying the contracts to the economic interests of all the parties. The market's reaction confirms this.
What's your take on how these deals impact financial reporting? My first intuition is there will need to be a lot of adjustments to do industry comps between AI and semiconductor companies.
This is an incredibly thorough breakdown of a transformative deal structure. The warrant mechanism is particularly clever - by setting strike prices at $146, $157, and $168, AMD essentially gets paid to build capacity while OpenAI gets price protection against future cost escalation. What really strikes me is the inverted risk profile compared to Nvidia's model. Nvidia took equity risk by investing directly in customers, but AMD is getting guarenteed revenue while OpenAI assumes the utilization risk. The 6 gigawatt target is staggering - that's roughly 1.5% of total US electricity generation just for AI inference. Your point about power infrastructure being the real bottleneck is spot on. The chip supply might materialize, but can the grid capacity?
https://open.substack.com/pub/pramodhmallipatna/p/ais-grand-entanglement-the-subprime
Excellent write up!
Thanks for giving us the scoop.
https://open.substack.com/pub/faiezkirsten/p/silent-weapons-deployed-worldwide?r=1qpmbr&utm_campaign=post&utm_medium=web&showWelcomeOnShare=false
thank you so much for your analysis!