Anthropic generated $11.6 billion in second-quarter revenue and reached a modest operating profit, according to financial figures reported by the Wall Street Journal. The quarterly total more than doubled Anthropic's first-quarter sales. In contrast, OpenAI told investors that its second-quarter revenue reached $6.7 billion, up 18 percent from $5.7 billion in the first quarter, while its operating loss widened to $12.3 billion.
OpenAI's reported loss included stock-based compensation, expanding from a $9.3 billion deficit in the previous quarter. Earlier reporting cited by Bloomberg indicated that Anthropic's annualized revenue run rate climbed above $65 billion by late July, compared with an internal OpenAI run-rate estimate exceeding $40 billion. The disclosures reflect private company reporting based on people and internal documents rather than audited public filings.
Enterprise Workloads Versus Consumer Delivery
Revenue accounting differences can alter direct comparisons between the two developers. Sales routed across cloud partners, capitalized training expenses, and adjusted profitability definitions vary across private ledgers. Nevertheless, the reported disparity highlights diverging business models. Anthropic has focused heavily on coding assistants and corporate software integrations, which appear to generate cash more rapidly than OpenAI's large consumer funnel.
OpenAI continues to absorb massive infrastructure costs to support hundreds of millions of consumer users alongside commercial deployments. Serving, training, and securing frontier models across a broad, free-tier audience requires substantial ongoing capital. The second-quarter figures show that broad consumer engagement does not inherently produce immediate operating margins if delivery costs outstrip subscription and enterprise income.
CME Prepares October Launch for Compute Futures
As frontier training expenses mount, financial exchanges are moving to treat compute capacity as a measurable financial input. A TechCrunch Equity report highlighted Silicon Data's efforts to establish benchmark pricing for artificial intelligence hardware. The firm secured $30.5 million earlier this month to build index benchmarks and risk tools supporting CME Group's planned compute derivatives.
CME Group plans to introduce cash-settled futures contracts tied to Nvidia H100 and B200 graphics processors on October 5, pending regulatory review. The contracts are intended to provide enterprise buyers and cloud operators with financial instruments to hedge against fluctuations in processing costs, similar to established derivative markets in energy, agricultural commodities, and industrial raw materials.
Structural Limits of Hardware Hedging
Standardizing compute derivatives presents distinct technical hurdles that do not exist in conventional commodity trading. Unlike crude oil or wheat, data center capacity cannot be stored in tanks or shipped between physical hubs. Two server clusters running identical chips often yield substantially different functional output depending on network topology, software efficiency, cluster uptime, and active utilization rates.
Because physical hardware cannot be warehoused, derivative contracts depend heavily on the accuracy of underlying performance benchmarks. If pricing indices fail to capture actual computing throughput or prove susceptible to optimization games, financial hedging will detach from real operational costs. For commercial developers, compute exposure ultimately reflects cost per finished task rather than nominal hardware rental rates.
Z.ai Introduces GLM-5.3 Amid Safety Review
International competitors are also changing the cost equation for specialized developer tooling. Z.ai released GLM-5.3 through its API and commercial coding plan, targeting programming workflows and multi-step autonomous tasks. The developer reported performance improvements on vendor evaluations, with scores rising from 4.6 to 28.3 on Terminal-Bench 3.0 and from 46.2 to 66.9 on DeepSWE v1.1.
Z.ai disclosed that GLM-5.3 scored 84.5 percent on the CyberGym vulnerability-discovery evaluation following defensive training. Consequently, the company postponed releasing the model's open weights for approximately two weeks while implementing additional safety controls. While unreleased weights undergo safety reviews, the hosted API permits ongoing surveillance and rate limits that allow developers to monitor anomalous querying patterns.
Enterprise Focus Moves to Unit Economics
Shifting margins among model developers are beginning to alter enterprise vendor negotiations. Facing mounting operational expenses, providers may bundle specialized tools, adjust inference price points, or restrict usage ceilings to steer enterprise clients toward proprietary platforms. Corporate technology buyers must evaluate full workflow costs, including secondary human verification, retrieval systems, and security monitoring, rather than standalone token pricing.
Corporate procurement teams are increasingly relying on structured evaluations to determine automation value. Tracking completed business outcomes alongside system maintenance, data privacy controls, and failure recovery provides visibility into true operational returns. Regulatory mandates for audit logs, access verification, and human oversight functions similarly operate as diagnostic tools, identifying which automated processes maintain positive unit margins under production workloads.
The emergence of corporate revenue divergence, compute futures contracts, and competitive API alternatives signals an evolving commercial landscape for artificial intelligence. Sustained adoption now depends on unit economics rather than sheer parameter scale. Developers and corporate buyers alike must prove that automated task execution generates measurable economic return after accounting for infrastructure expenses, review overhead, and operational risks.
AI news questions, answered
What were the reported second-quarter financial results for OpenAI and Anthropic?
According to reports by the Wall Street Journal, OpenAI generated $6.7 billion in second-quarter sales with an operating loss of $12.3 billion, while Anthropic generated $11.6 billion in revenue and achieved a small operating profit.
When will CME Group launch compute futures contracts?
CME Group has scheduled cash-settled futures contracts for Nvidia H100 and B200 processors for October 5, pending regulatory review.
Why did Z.ai delay releasing open weights for GLM-5.3?
Z.ai delayed releasing the open weights for approximately two weeks to strengthen safety controls after the model scored 84.5 percent on the CyberGym vulnerability-discovery benchmark.
Get daily AI news by email
Short morning and evening AI-only updates from TweeLabs Digital. No general tech noise.