AI capex sustainability, financing fragility, and national security implications (with Jonathan Ferro, Lisa Abramowicz, Amarie Hordern)

South Korea is upping its AI play with massive investments in chips and data centers. Asks which argument Mamta Q puts more weight on: demand justifying investment or cost of serving AI going up.

Jonathan Ferro

Mamta Q

Overall AI sentiment is still positive, but the big question is that AI capex estimates keep moving up and we don't know when it will stop. Bottlenecks have shifted from GPUs to memory to power. On the demand side, enterprise adoption is starting but the timeline is stretched out.

If this is critical infrastructure that will pay off, it's worth it. Otherwise, capex keeps expanding and investors will be impatient.

Mamta Q

The question for AI has shifted from technology to financing. Public equity demand is not bursting at the seams for the $4 trillion of new market cap across SpaceX, Anthropic, OpenAI. Bondholders are cautious, demanding higher prices for technology risk and years of negative cash flows, which is uncommon in investment grade bonds.

The market is fragile: OpenAI delaying IPO news sent shares of the whole AI trade down, while SK Hynix and Samsung news brought demand back up.

Mamta Q

AI has moved from technology to national security and national competitiveness. China has a lead in open-source models and offering cheaper AI capabilities. Every country is trying to find its own edge.

Korea is taking hold on the memory side while the US has Nvidia and hyperscalers.

Mamta Q

AI is now a national issue, creating a tug-of-war between economic and national security sides. The Nvidia China supply question and Google limiting Meta's use of Gemini models show everyone is trying to get compute.

AI becoming durable won't be seen until costs come down significantly. There's a lot of focus on inference costs, and enterprise adoption is still waiting to be proven out.

SignalTube