My impression is that there are many parallels computing at large where custom hardware is becoming more and more prevalent. You can run arbitrary C programs with FHE by building a CPU out of binary gates and running on that, but it will run at 1Hz[2] emulated on 8GPUs. So, computing fibonacci(5) takes like 16 minutes. Conversely, you could create an arithmetic circuit that does it in like 16us. However, working with circuits is hard, let alone the additional requirements FHE imposes.
Today, our compiler lets you write Rust code that turns into an arithmetic circuit in the BFV scheme. It also manages parameter selection, which is another annoying part of FHE: choose parameters too small and decryption will fail due to excessive noise, but larger parameters slow down the computation and make ciphertexts larger.
Overall, the FHE has a ton of promise, but is currently in the chicken and egg phase whereby there isn't much commercially available because there isn't a market because there isn't anything available. We're trying to be an egg and grow along with a market. FHE is a big area and there's a ton to explore, like multi-key encryption[3]. FHE is currently nascent, but I believe its future is bright where it can be appropriately used.
[1]: https://docs.sunscreen.tech/
[2]: https://www.usenix.org/conference/usenixsecurity21/presentat...
Are there any companies doing pioneering work on this now? What aspects of FHE does your employer do? How would you say the future is looking for FHE?