group_split + furrr::map_dfr()

example_furrr

library(furrr)
library(dplyr)

future::plan(future::multisession()) # This is for windows, might want something else for Linux

data <- data.frame(letter = sample(letters,num_samples, replace = TRUE),
           value = runif(num_samples))

data %>%
  group_by(letter) %>%
  group_split() %>%
  furrr::future_map_dfr(function(.x){
    tibble(letter = first(.x$letter), value = sum(.x$value))
  })