Investors start asking what the AI buildout is actually worth

The valuation cut, argued as rational

The sharpest conversation in investing podcasts this week was about a number going down. On 20VC, the panel discussed OpenAI's valuation cut from $1.4 trillion to $1.25 trillion, and argued it was a rational market response to liquidity and growth comparisons rather than a sign of the AI wave cracking. The same episode noted that pricing pressure in the language model market is intensifying, with recent models offering equivalent performance at 30 percent lower costs, and that Anthropic's net new bookings growth decelerated in the third quarter, which the panel read as a sign of potential market nervousness. m:…9089

The panel also argued that the era of free subsidies for AI agents is coming to an end, because persistent virtual machines are expensive to run. And it raised a threat to the delivery business: AI agents pose an existential risk to delivery platforms like DoorDash by enabling users to find cheaper, direct-to-restaurant delivery options. That framing matters because DoorDash is buying Instacart for $17 billion, a deal the panel called a logical move to consolidate delivery networks and reach 80 percent market share — a bet on scale made just as the panel argued agents could undercut the model. m:…9089

The firm that won't buy in

A very different answer came from Mavik on the Capital Allocators podcast. Mavik is a $2 billion flexible special situations investment firm, and the firm avoids AI infrastructure and data center investments because of uncertainty about their long-term fundamental value. Instead, the firm argues that commercial real estate is currently a compelling environment for restructurings and recapitalizations, and noted that regional bank lending for new construction has not returned to pre-2023 levels. Mavik's discipline is part of the story: in 2023, the firm did not execute a single deal for almost a year, and its strategy is to let the opportunity shape the investment rather than fitting into rigid mandates. Every eligible employee at Mavik is invested in the firm's funds, and the firm shares incentive fees with every person across the organization. m:…8039

The bull case, with a caveat

On How Leaders Lead, Laffont offered the bull case — and then admitted its limits. Laffont said the semiconductor industry market cap grew from $2 trillion in 2020 to $24 trillion today, and argued that the current era, which he calls the Age, began in 2022 with new semiconductor architectures like the GPU. His investment strategy focuses on what he calls Big Ideas, large shifts that act as a root for multiple related business opportunities. But Laffont also admitted to missing the cloud computing wave, which he said resulted in poor performance for his firm for several years — a reminder that even the investors who name the era correctly can miss the wave inside it. m:…6731

Laffont also argued that AI will not destroy jobs but shift them, in the same way work shifted from agriculture to industrialization, and that data centers can improve grid reliability by agreeing to power down during peak demand periods. His most striking claim was about where the infrastructure goes next: he argued that data center infrastructure will eventually move to space, to avoid terrestrial constraints like permits and pollution. Taken together, his case is that the buildout is real and still early — and that the constraint is not demand for computing, but where to put it and what it costs. m:…6731

Where the three agree, and where they don't

The three podcasts do not agree with each other, which is what makes the week interesting. The 20VC panel treated the valuation cut as healthy repricing, Mavik treated AI infrastructure as unpriceable and stayed out, and Laffont treated the whole era as a once-in-a-generation shift that is still under way. This suggests the market is not arguing about whether AI matters — nobody on any of these shows argued that — but about what the infrastructure is worth and who should own the risk. The Canadian government's investment of up to $1.14 billion in Cohere, discussed on 20VC as a move to ensure sovereign AI capabilities, shows that governments are now part of that argument too. The open question for the next few months is whether the pricing pressure the 20VC panel described — equivalent performance at 30 percent lower costs — turns into a genuine correction, or just a cheaper on-ramp to the same buildout. m:…9089 m:…8039 m:…6731