McKinsey: Scaling AI inference and the 5 shifts that define it

By McKinsey & Company · Wed, 14 Oct 2026.

This email, sent by Bill Wiseman and Marc de Jong, global leaders of McKinsey's Semiconductors practice, introduces a new piece titled «Scaling AI inference». The introductory text notes that inference is reshaping the economics of AI, infrastructure architectures and investment priorities across the semiconductor ecosystem.

The main article linked in the email is titled «5 shifts that define AI inference». The body of the email does not explain or list these five shifts. It provides only the headline and the introductory sentence, so they cannot be detailed here without inventing content.

In the «Also Consider» section, the email recommends two more pieces: «Frontiers of compute: The technologies to reduce AI inference costs» and «The next era of semiconductor value creation».

The email goes to subscribers of McKinsey's «Advanced Industries» alert list. It invites readers to contact the team to discuss how these ideas could affect their organization.

For industry context, inference has become a growing focus of cost and investment in AI infrastructure. Inference means running already-trained models to handle requests from users and agents. Autonomous agents, which chain many model calls together, tend to increase that demand. Readers who want the details of the five shifts need to read the full article on McKinsey's website.