An Interview with Ben Thompson at the MoffettNathanson Media, Internet & Communications Conference

The more pointed claim buried in Thompson's argument is that compute constraints may actually protect incumbents with existing infrastructure relationships rather than opening the field to challengers, which cuts against the usual 'disruption favors the newcomer' framing.

Modelwire has no prior coverage directly connected to this piece, so it sits somewhat on its own. That said, it belongs to a broader conversation that has been building across the industry around whether AI economics eventually converge toward utility-style infrastructure or remain winner-take-all platform plays. Thompson's compute-scarcity framing is a specific structural argument that would sharpen considerably when read alongside any coverage of hyperscaler capacity commitments or inference pricing moves, neither of which we have in the archive yet.

Watch whether any of the major inference providers, particularly those without their own silicon, announce pricing changes or capacity rationing in the next two quarters. If inference costs stay flat or drop while Thompson's scarcity thesis holds, the vertical integration argument weakens materially.