My median estimate sits at roughly 1.5% of 2026 frontier-lab training compute. Crucially, this metric targets training compute, not inference. While Dwarkesh Patel’s essay argues that 30–50% of a lab’s compute is currently wasted on non-learning inference, the fraction of training compute driven primarily by deployment data is an entirely different quantity 8 Predictions for the Era of Continual Learning. The denominator here—total fronti
Aligning this estimate with projections of the dominant personalization paradigms and data acquisition trends slightly tightened the upper bound of the forecast, reflecting that massive synthetic and web pretraining runs continue to overwhelmingly dominate deployment data compute.
Current Baseline and Operational Realities As of late 2026, the share of alignment and safety research dedicated to models undergoing weight updates from deployment data is small—likely in the 2–5% range. There is currently little operational pressure forcing labs to pivot: no publicly known frontier chat or reasoning model updates its weights from live sessions. Recent features like ChatGPT Memory, Claude's cross-provider memory import, and Agent Skills all rely on retrieval or context inje
Baseline and Historical Volatility The likely resolution source is Menlo Ventures' late-2028 enterprise survey. Their September 2026 edition placed Anthropic at a record 55% share of enterprise LLM API spend, with OpenAI at 31% and Google at 9% 2026: The State of Enterprise AI | Menlo Ventures 2026: The State of Enterprise AI | Menlo Ventures [2026: The Stat
Set against related questions, this was slightly lowered to reflect greater expected market fragmentation, while remaining structurally bounded above individual provider shares .
Which lab we are pricing Anthropic is the clear leader for the largest AI lab by revenue. Its Q3 2026 GAAP revenue was $16.96B, well ahead of OpenAI’s roughly $10B, and its annualized run-rate reached approximately $81B by the end of October Anthropic eyes $1.4 trillion IPO valuation in test of Wall .... While this comparison is imperfect—Anthropic books its cloud-marketplace sa
Set against related questions, this was slightly raised to balance the headline gross margins with the necessary deductions for partner fees .
The estimated distribution for the FY2028 blended gross margin on inference revenue is 56.0% (p10), 64.0% (p25), 71.0% (p50), 78.0% (p75), and 84.0% (p90).
Anthropic is heavily favored to be the largest AI lab by revenue in FY2028 given its Q3 2026 GAAP revenue of $16.96 billion against OpenAI's approximately $10 billion. Anthropic’s current S-1 filing implies a highly favorable gross margin on customer-serving compute—approximately 85.5% for the first nine months of 2026 and 85.7% in Q3 [Anthr
Set against related questions, this was shifted downward to reflect higher confidence that competitive price wars will compress future margins .
By late 2028, parameter-efficient adapters (principally LoRA) are expected to remain the dominant method for per-customer model customization by deployment count, driving a median estimate of roughly 18% for full-weight updates. The economics of multi-adapter serving heavily favor this outcome. Platforms like S-LoRA and Punica demonstrate that a single base model can host thousands of active adapters with roughly a 5% throughput penalty. Because of this, cloud and open-weight providers—including
Revised modestly downward to account for the ongoing deprecation of major self-serve fine-tuning platforms, which concentrates the full-weight market into a smaller set of high-end enterprise deployments.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited