Inference Compute Jumps to Two-Thirds of AI Workloads, Reshaping Profit Pools
msharmas · x · 2026-08-07
A report by Activant Research highlights that AI inference is accelerating and overtaking training as the dominant workload. Key data points include:
- Compute Shift: Inference now accounts for roughly two-thirds of all AI compute, up from one-third in 2023. CoreWeave confirmed over half of its fleet now serves inference rather than training.
- Token & Revenue Surge: AWS served more tokens in Q1 2026 than in its entire prior history combined. Anthropic's revenue skyrocketed from $9 billion to $30 billion in a single quarter. Serverless inference providers like Together AI and Fireworks have crossed $1 billion in ARR.
The report argues that as open-source models close the capability gap with closed-source ones, every layer of the inference stack is commoditizing fast. The ultimate race will be to become the end-to-end provider governing the whole stack.
More from Venture
- AMD Acquires AI Inference Chip Startup Taalas to Boost Enterprise Compute — lurenjia_3x · 2026-08-07
- AI Face Rating App Dailyglowup Listed for Sale at $600k — marclou · 2026-08-07
- Keep Launches 'Super AI Membership': From Content Delivery to AI Agent Services — 量子位 · 2026-08-07
- Unitree Reportedly Set for IPO at $9B Valuation — zephyr_z9 · 2026-08-07
- SK Hynix 2Q26 OPe Forecast Misses Consensus Due to High HBM Mix and LTA Effects — zephyr_z9 · 2026-08-07
- Cathie Wood Cites Shopify Data: AI-Driven Orders Up 3x YoY — CathieDWood · 2026-08-07