Confidential AI
Confidential AI can keep most of its speed when running models, in the right configuration. NVIDIA's September B200 test kept over 96% of baseline token throughput. Separate research shows much larger costs when software is not adapted or transfers between CPU and GPU are protected. Test your whole workload against the threats you face before relying on claims of near-zero overhead.
Research snapshot · updated Nov 12, 2026
Healthcare + financial + government provide reference verticals validating commercial procurement
BeeKeeperAI on Azure confidential computing (a Microsoft partnership, not an acquisition) anchors healthcare patient-data Confidential AI, with Mayo, UCSF and Sutter Health among named deployments. JPMorgan + Goldman + Citi cover financial services. Anthropic + Palantir reach federal and intelligence-community workloads. Reference customers reduce enterprise procurement friction; each $10M+ contract proves the procurement template.
Confidential AI in the cloud
Confidential AI runs models inside protected hardware so that the cloud operator cannot read the data. AWS Nitro Enclaves, for example, are isolated virtual machines that use processor and memory set apart from the parent instance, and they can prove what they are running through attestation. The cost in speed depends on the set-up. NVIDIA's September 2026 test on a B200 kept more than 96% of normal token throughput. Other research found much larger slowdowns when software is not adapted or when data moving between processor and GPU is protected. What each service protects varies, so check it against the workload you plan to run. We do not have a revenue figure for this market that we trust.
State of the art (2026)
Confidential AI has moved from research to production. NVIDIA Blackwell is the first TEE-I/O-capable GPU, encrypting HBM and NVLink traffic at near-zero throughput cost; in March 2026 Corvex verified a production confidential-computing deployment on HGX B200 with CPU and GPU remote attestation via Intel TDX and Trust Authority. Azure Confidential AI, Google Cloud Confidential Computing and AWS Nitro Enclaves are generally available, and healthcare and financial-services reference customers anchor early procurement. The expected EU regulatory tailwind has softened, though: the Digital Omnibus agreed in May 2026 defers high-risk AI Act obligations to December 2027, so near-term demand rests on enterprise data-protection economics rather than a hard compliance deadline.
EU AI Act is a slower regulatory tailwind after the 2026 deferral
The May 2026 Digital Omnibus defers high-risk Annex III obligations from August 2026 to 2 December 2027 (embedded Annex I products to August 2028). Conformity assessment, audit trails and cybersecurity duties still favour cryptographic attestation, but the hard compliance deadline that would have forced procurement has moved out by sixteen months. Near-term Confidential AI demand therefore rests on enterprise data-protection economics – running frontier models on regulated data without exporting it – rather than an imminent EU mandate. The regulatory pull is real but is now a 2027-2028 story.
Explore the deeper analysis in CanaryIQ
Explore what is covered below. These previews show the structure; log in to read the full analysis.
Signal stack
Evidence stacked leading → lagging
6 signals
Signal stack
Evidence stacked leading → lagging
Technology-native KPIs
Metrics that predict trajectory, tracked over time
5 tracked
Technology-native KPIs
Metrics that predict trajectory, tracked over time
Landscape map
Who builds what, and who depends on whom
140 players · 5 layers
Landscape map
Who builds what, and who depends on whom
Catalyst calendar
Dated events that will move the position
6 ahead
Catalyst calendar
Dated events that will move the position
Technology roadmap
Milestones on the path to maturity
7 milestones
Technology roadmap
Milestones on the path to maturity
Watchlists
Companies, people and papers, each with a condition for removing it
20 · 20
Watchlists
Companies, people and papers, each with a condition for removing it
Decision frameworks
What our view means for your job
Locked
Decision frameworks
What our view means for your job
Thesis changelog
When our view changed, and why
9 updates
Thesis changelog
When our view changed, and why
Change our mind
5 disconfirming conditions
You've read the verdict. The file is much deeper.
The full signal stack, technology-native KPIs tracked over time, the landscape of who depends on whom, the dated catalyst calendar, decision frameworks for every desk, live watchlists and a record of every time our view on Confidential AI has changed. All of it is in CanaryIQ.