Can individual firms reach viable scale for vision automation?

Svanberg, Li, Fleming, Goehring and Thompson's MIT FutureTech cost model finds that getting bigger, by itself, barely moves a firm toward cost-effective computer-vision automation. "The median employee works in a firm where close to none of the vision tasks are cost-effective to automate," and "even a firm with 5,000 employees, i.e., larger than 99.9% of firms in the United States, could only cost-effectively automate less than one tenth of their existing vision labor at the current cost structure." Pushed to the extreme, "even a hypothetical firm as big as Walmart lacks the scale to make automating 15% of their vision tasks attractive."

The mechanism is the same minimum viable scale threshold from the companion note — the deployment size at which an AI system's amortized cost matches human labor cost — but applied to firm size specifically rather than to tasks in general. The authors test whether a more concentrated economy (fewer, bigger firms, or one occupation per firm) would change the picture and find it would not: "even a perfectly concentrated economy with exactly one occupation per firm would only have a tenfold multiplier on task value compared to our base assumption," not enough to close the gap. They note this squares with McElheran et al. (2023), who find fewer than 6% of firms use AI-related technologies, disproportionately large ones representing 18% of employment — consistent with scale mattering but still being scarce even among big firms. Their proposed way past the ceiling is aggregation across firms rather than within one: AI-as-a-service, where "one firm develops the AI system for a task and others that also need the system outsource" to it, pooling the wages paid across every firm in a NAICS industry code against the cost of one shared system. They cite a diamond-classification tool built by NavTech and an NVIDIA self-driving platform collaboration as existing examples of this model.

This sharpens Do firms substitute labor for AI at different rates?: that note shows exposed firms substitute labor for AI faster and more cheaply, which could read as returns to scale internal to the firm, but this paper's Walmart example suggests individual firm scale runs into a ceiling well before most vision tasks clear minimum viable scale — the returns to scale that matter more may be external, pooled across firms via AI-as-a-service. It also qualifies Does concentrated AI exposure enable workers to adapt and reallocate?: reallocation within a firm is one adaptation path, but this paper implies the more consequential lever for unlocking automation is market-level demand aggregation, not firm-level task concentration. See also Which vision tasks actually make economic sense to automate? for the headline figure this firm-size evidence supports.

The Walmart case is explicitly "hypothetical," not an observed firm, and the AI-as-a-service proliferation path is framed as a belief ("We believe that aggregating demand for AI solutions is more likely to happen through AI-as-a-service business models") rather than a measured market outcome. The paper does not estimate how fast such aggregated markets would actually form, only that they are the more promising route past the scale ceiling than firm growth alone.