The frontier AI market is entering its next phase. The industry is moving beyond consumer chatbots, with enterprises now starting to adopt long-running coding and work agents more deeply – all of which drive token-intensive workflows. That shift is dramatically increasing revenue and compute consumption, forcing leading AI labs like Anthropic and OpenAI to secure tens of gigawatts of future capacity.
Anthropic has captured the early lead in this agentic transition. Its enterprise-first strategy, usage-based pricing, and earlier push into agentic services have driven its exponential acceleration this year, allowing it to overtake OpenAI in quarterly revenue. OpenAI is now responding with its own enterprise platform, and has shifted its pricing structure deeper into metered APIs. Both now have rapidly expanding roadmaps to drastically increase AI capacity over the next several years, and are relying upon multiple clouds, neoclouds, chip providers, and buildout partners to get there.
This series provides a first look at Anthropic and OpenAI ahead of their coming (albeit delayed) IPOs. This post examines the rise of agentic services, the resulting revenue inflection, and the differing product and pricing strategies behind both companies. Follow-up posts then cover each AI lab's capacity buildouts and financing strategies.
This series will be of interest to hyperscale clouds, chip providers (including NVIDIA, AMD, Cerebras, and Broadcom), neocloud & site partners (including CoreWeave, IREN, Nscale, SB Energy, Hut 8, Core Scientific, TeraWulf, Cipher Mining, and Riot), and upstream suppliers.
- Agentic AI is now a top driver of compute demand. These coding agents, work agents, and managed long-running agents consume far more tokens than chatbot usage.
- Anthropic entered this market earlier, and built its operational model around usage-based enterprise revenue. That helped it reach a reported $65B revenue run rate in July, versus $40B for OpenAI. An estimated 75-85% of ARR is driven by consumption-based pricing, vs 15-25% from subscriptions.
- Anthropic has now spread its Claude models across the agentic AI platforms of all 3 top clouds, plus has embedded its full platform into AWS. These cloud partners now drive an estimated 15-20% of ARR, helping Anthropic capture enterprise customers that want to utilize cloud spend commitments.
- OpenAI has shifted its focus to enterprise over the past year, and has now assembled a set of competing agentic products (despite any potential headwinds from its major org restructuring, unwinding its exclusivity with Microsoft, and ongoing mgmt churn). Yet OpenAI has been roughly 3-5 months behind Anthropic in releasing competing agentic products.
- Critically, OpenAI has also been expanding more deeply into consumption-based pricing. This started in April with a new PAYG Enterprise plan, and its metered API is now backing several new agentic products, including a new managed agent service in AWS Bedrock.
- The gap between these competitors is beginning to close, as coding and agentic workloads have quickly risen in OpenAI's mix this year. Its consumer segment is expected to fall from 65% to 20% of the mix this year, and mgmt noted that its enterprise segment just surpassed consumer in July.
- Metered pricing should drastically improve the economics of both companies in this agentic demand wave. Revenue can expand with every additional workflow and token rather than adding seats. Anthropic is already hitting non-GAAP profitability, and OpenAI's shifts into consumption-based pricing should help reverse its worsening margins.
Compute requirements are rising alongside revenue. SemiAnalysis estimates that OpenAI and Anthropic currently have over 6GW of compute, with combined demand expected to exceed 100GW by the end of 2030. Next up are deep dives into Anthropic's and OpenAI's ongoing buildouts and financing strategies.