In-house inference enables pricing flexibility
Shortwave
Owning the inference layer lets Shortwave sell email software like software, not like metered model access. Because most AI work runs on open source models on Shortwave controlled hardware, the company can package AI into seat based plans, absorb heavy usage from power users more efficiently, and keep premium tiers tied to workflow value such as deeper search history, stronger models, and onboarding rather than simply passing through third party API bills.
-
Shortwave already prices around the workload that actually costs it money. Its billing and plan structure gates AI search depth, advanced assistant features, and higher usage limits, while keeping email access itself broadly available. That is easier to sustain when compute is mostly internal, because gross margin does not rise and fall one for one with external API spend.
-
The same build it yourself pattern shows up elsewhere in the product. Shortwave connects directly to Gmail rather than through an email API middleman, and processes email heavily on its own servers to power real time delivery, normalized rendering, and team features. Inference running in house fits that broader choice to control the critical workflow layer instead of outsourcing it.
-
This also sharpens the contrast with premium email peers like Superhuman. Superhuman sells speed and AI organization at a much higher price point, while Shortwave has room to bundle more AI usage into lower team plans and reserve the most expensive tier for the heaviest users. That creates more freedom to undercut on price while still expanding margins as usage grows.
The next step is for email pricing to look more like a blended software and compute model. As Shortwave adds more automations, integrations, and agent style actions inside the inbox, controlling inference in house should let it keep pushing AI deeper into the product without making every new feature a margin hit, which strengthens both expansion revenue and competitive pricing power.