Applications, cost, and future

FHE is a theoretical dream come true. But between "it is possible" and "it is usable every day," a major obstacle remains: speed.

Applications that inspire dreams

FHE unlocks uses impossible otherwise, all founded on the same principle — process without seeing:

• Confidential cloud: outsource storage and computation to powerful servers without ever revealing your data to them.
• Private machine learning: train or query an artificial intelligence model on encrypted medical data, without the model provider seeing the patients' records.
• Private search: query a database without the server knowing what you are looking for.

The major obstacle: cost

If FHE is not yet everywhere, it is because of its cost. It remains very slow — often thousands to millions of times slower than an equivalent computation in the clear — and the ciphertexts are bulky (a single encrypted bit can take up kilobytes).

Bootstrapping, in particular, is the most expensive operation: it is the price to pay for resetting the noise.

        base operation          ciphertext size
clear : 1 (reference)           a few bytes
FHE   : ×1000 to ×1000000       kilobytes per value

Possible use	Current limitation
Confidential cloud	Slowness, bulky ciphertexts
AI on medical data	Very high computation cost
Private search	Significant latency
Voting, secret auctions	Implementation complexity

Rapid progress

Research has been advancing fast since 2009. Several next-generation schemes have reduced costs:

• CKKS: designed for approximate computation on real numbers, ideal for machine learning.
• TFHE: optimizes bootstrapping to make it very fast on boolean operations.

On top of this come mature libraries and the development of dedicated hardware (specialized accelerators) that promises to gain several more orders of magnitude.

A promising future

FHE illustrates a deep trend: no longer having to choose between using a piece of data and protecting it. Still reserved for cases where confidentiality justifies the extra cost, it becomes more practical every year. What was science fiction in 2009 is today deployed in the first real services.

In summary

FHE promises the confidential cloud, AI on encrypted medical data, and private search, but its major obstacle remains its slowness (thousands to millions of times the cost of clear computation) and the size of the ciphertexts. Progress is rapid — CKKS, TFHE, libraries, and dedicated hardware — making the technology more practical every year.