Supabase logs show every product page query running twice (generateMetadata + page)

The automatic dedupe applies to fetch() calls with the same URL and options inside one render. supabase-js does use fetch internally, but not in a way you can rely on for that. Wrap your function in React's cache, which dedupes by arguments for the duration of one request:

import { cache } from "react";

export const getProduct = cache(async (id: string) => {
  const supabase = await createClient();
  const { data } = await supabase.from("products").select("*").eq("id", id).single();
  return data;
});

Now generateMetadata and the page share one result. It's per request, not a cross-request cache, so no stale data risk.