The denominator for this estimate—total frontier-lab training compute in 2026—is dominated by massive pretraining runs and increasingly large post-training reinforcement learning (RL) phases. Pretraining historically accounts for the vast majority of compute and relies almost entirely on web, licensed, curated, or synthetic corpora, contributing roughly zero to the numerator. While reasoning and RL post-training compute are growing rapidly, the evidence strongly suggests these modern runs are dr
The forecast remained unchanged, as evaluating this alongside timelines for continuous weight updates confirmed that production data continues to represent only a minimal fraction of overall training compute.
The current baseline for alignment and safety research focused on weight-updating models is in the low single digits, with a 10th percentile of 2.5% and a 25th percentile of 4.5%. At the frontier, all deployment currently relies on fixed weights augmented by retrieval or context injection. Consequently, operational pressure to dedicate headcount to live-update safety remains near zero. Established lab governance documents—including Anthropic's RSP v3.4 anthropic.com, OpenAI's Preparedness Framework [
The estimate was adjusted slightly downward to reflect the expectation that widespread implementation of continuous weight updates on frontier models will likely not arrive until the 2030s, delaying the need for dedicated safety staffing.
The final estimated distribution is p10: 27.1, p25: 33.6, p50: 40.0, p75: 47.0, p90: 55.6.
Initial logic and baseline parameters regarding enterprise survey anchors are validated. The historical spend share baseline across major API providers remains established context, confirming that market leadership transitions and share volatility establish the foundation for exceptionally wide uncertainty bands menlovc.com.
Forces for Fragmentation Market fragmentation and multi-homing
Evaluating the slow timeline for continuous learning architectures and explicit data-for-access lock-in slightly firmed up the median near 40% while preserving a very wide distribution.
Final Estimates: 45.0% (p10), 56.0% (p25), 64.5% (p50), 72.0% (p75), 79.0% (p90).
Identifying the Target Lab Initial logic and parameters are validated. Established context confirms Anthropic as the target lab being priced 55 sources.
The Central Estimate Standard processing applied. The baseline inference margins and consumer subscription impacts are validated as established context 55 sources.
Downside and Upside Risks
Set against related forecasts, this distribution remained largely stable but the lower tail was slightly adjusted to maintain a consistent trajectory of near-term versus long-term price competition risks.
Determining the FY2028 Revenue Leader The forecast centers on Anthropic, which is currently the most likely candidate to be the largest AI lab by revenue in FY2028. As of mid-May 2026, Anthropic claimed a $47B annualized run rate, heavily concentrated in enterprise and API deployments anthropic.com.
Baselines and the Strict Margin Definition The metric is narrowly defined as the blended gross margin on inference revenue—serving revenue minus the direct cost of serving. Under this definition,
Set against a related near-term forecast , the lower tail of this distribution was widened to better reflect the compounding downside risk of multi-year price wars.
Multi-tenant serving economics overwhelmingly favor parameter-efficient adapters over full-weight updates. This structural reality underpins our expectation that commercial per-customer customization will heavily rely on LoRA. Technical headwinds against adapters under sequential learning regimes are counterbalanced by the sheer operational efficiency of multi-tenant adapters, anchoring the median at 18.0%. Set against related questions, the distribution of percentiles (including 5.0% at p10 and
Set against related questions, these percentiles were slightly adjusted to consistently reflect that multi-tenant serving economics heavily favor adapters over full-weight updates.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited