Macro & Thematic Indices
Leading indicators for the global AI economy. Serving institutional traders, hedge funds, and macro analysts tracking the compute cycle.
The Training Compute Index
The institutional benchmark tracking wholesale computational throughput costs ($/PFLOP-hr) for elite, scale-up interconnect GPU clusters (SXM and OAM fabrics) engineered for foundation model pre-training. For distributed systems engineers, it normalizes multi-node NVLink and InfiniBand scaling overhead, helping teams benchmark cloud quotes, optimize training budgets, and forecast cluster completion milestones. For technology investors, fund managers, and equity analysts, it serves as the definitive macroeconomic reference rate for AI capital expenditure—providing transparent, real-time insight into the operational cost structure of frontier AI development.
| B200 SXM6 | H200 SXM5 | H100 SXM5 | MI300X OAM | A100 SXM4 |
|---|---|---|---|---|
| -- | -- | -- | -- | -- |
The Inference Compute Index
Tracks the volume-weighted wholesale cost ($/PFLOP-hr) of modular, scale-out PCIe accelerator hardware purpose-built for production token serving, enterprise fine-tuning, and low-latency API workloads. For machine learning deployment engineers, it reveals the true market price of production serving capacity, empowering teams to slash recurring operating costs by migrating latency-tolerant workloads from costly SXM clusters to high-efficiency PCIe nodes. For credit funds and corporate finance leaders, it establishes the fundamental baseline for recurring AI operational expenses, enabling rigorous valuation and margin modeling for enterprise AI applications.
| H100 PCIe | L40S PCIe | L4 PCIe | A10G PCIe | RTX 4090 | T4 PCIe |
|---|---|---|---|---|---|
| -- | -- | -- | -- | -- | -- |
The NVIDIA CUDA Premium
Quantifies the exact financial premium commanded by NVIDIA's proprietary software ecosystem by tracking the normalized price-per-FLOP spread between NVIDIA hardware and rival silicon (AMD Instinct, Intel Gaudi). For engineering directors, it translates software friction and runtime porting complexity into an objective financial threshold—identifying the exact hardware discount needed to make non-CUDA alternative clusters economically attractive. For semiconductor analysts and portfolio managers, it acts as the primary empirical barometer of ecosystem lock-in, measuring in real time whether competing chip architectures are successfully expanding market share or eroding NVIDIA's pricing power.
| Frontier Training (H100 vs MI300X) | Enterprise Inference (H100 vs Gaudi2) |
|---|---|
| -- | -- |