A maturing approach combines confidential virtual machines, using Intel TDX or AMD SEV-SNP, with Nvidia GPUs running in confidential-computing mode. The goal is to keep data encrypted and protected not only when it is stored or sent, but while it is actually being processed, including inside the GPU where AI models run.
That last gap, protecting data in use, has been the hard one, and it is exactly where sensitive information is most exposed.
What it means: for any business handling regulated or private data, this narrows a long-standing security hole. It makes it more realistic to run AI on confidential information, such as medical, legal or financial records, without simply trusting that the processing environment is safe.