Confidential compute + GPU marketplaces
A deep‑dive on Bittensor/TargonCompute says the Targon Virtual Machine will stitch Intel TDX and NVIDIA confidential computing to run encrypted AI workloads — the pitch: permissionless GPU marketplaces for trust‑sensitive models (x.com). Meanwhile Microsoft Azure just became the first cloud to deploy NVIDIA’s Vera Rubin GPUs for next‑gen AI compute, signaling broader commercial availability for that class of accelerators (x.com) (x.com).
Manifold Labs released the Targon Virtual Machine whitepaper on March 23, 2026, and the document lists Intel engineers as co‑authors on the technical whitepaper hosted with Intel. (targon.com) The paper describes Confidential Virtual Machines (CVMs) that run on Intel® TDX to isolate CPU execution, produce hardware‑anchored cryptographic proofs of correct execution, and keep model code and data encrypted in memory. (targon.com) On the GPU side, Targon’s TVM integrates NVIDIA’s nvTrust/NVIDIA Attestation SDK to collect and issue verifiable GPU attestation evidence that checks firmware, GPU configuration and NVSwitch integrity before workloads run. (manifold.inc) Targon’s public repo and product pages show built‑in monetization primitives that let subnet operators register hardware, advertise secure GPU inventory, rent GPUs to buyers and return attestation tokens as proof of a secure runtime. (github.com) Independent reporting and on‑chain analytics cited by market writeups place Bittensor compute subnets’ combined revenue in the vicinity of $20 million annual recurring revenue, and public metrics have been cited showing hundreds of billions of tokens processed in peak months. (unsupervised.capital) At NVIDIA GTC on March 16, 2026 NVIDIA announced the Vera Rubin platform—describing seven new chips in production—and Microsoft said Azure/Foundry had begun powering on Vera Rubin NVL72 racks for Azure AI workloads. (nvidianews.nvidia.com) Commercial rollout signals include an nScale–Microsoft collaboration to deploy large Vera Rubin capacity and press reporting that NVL72 rack pricing can reach as high as $8.8 million apiece, illustrating the scale and capital cost of early Vera Rubin installations. (marketwatch.com)
