How can I process marketplace images cheaply without exhausting serverless compute?

Sharp is not necessarily the problem; running it inside a Vercel request for every upload is. If you need reliable server-side variants, use an image service that integrates with your storage, such as Cloudflare Images or another image CDN. Those services can resize and transform images without consuming your application's serverless allowance. Another option is lazy processing: create each thumbnail or size only when first requested, cache it in R2 or a CDN, and reuse it afterward.

Client-side compression can work, but the biggest improvement is to stop routing the image bytes through your server at all. Have a lightweight API endpoint create a presigned R2 upload URL, then let the browser or app upload directly to storage. That endpoint only handles authorization and metadata, so it uses very little serverless CPU. The client can resize to a sensible maximum dimension, remove EXIF data, and use WebP when supported, with JPEG as a fallback. You should still validate the uploaded file's type, dimensions, and size outside the main request path because clients cannot be fully trusted.

For a very small project, a low-cost VPS could be simpler and more predictable than several free-tier services. You could run the processing worker there while keeping object storage separate. Even then, direct-to-storage uploads and client-side resizing are useful because they reduce bandwidth and processing work. I would avoid relying exclusively on client compression, though; keep validation and a fallback path for older phones and browsers.