Week of November 1, 2026

Pricing leads from Gong, Miro and Databricks share hard-won lessons on running AI credits. Gong ties one credit to roughly one analyzed call, and all three favor a single fungible pool with adjustable credit costs that only move in the customer's favor. They also urge tracking margins by workflow rather than by token and shipping admin usage caps at launch, which matters for both credit-based subscriptions and AI add-ons to existing SaaS.

Anthropic's S-1 shows H1 2026 revenue of $16.3B, an 85% gross margin and a first operating profit of $974M in Q2. About 48% of revenue, roughly $7.8B, came through Amazon, Google and Microsoft cloud marketplaces, and those channel fees are booked as sales costs, so they never touch gross margin. Analysts flag a gap between pay-as-you-go API revenue and decade-long compute commitments. They also note that the falling token price index mostly reflects model mix, not price cuts.

Anthropic's S-1 filing shows about 80% of its first-half 2026 revenue came from usage-based billing and only 20% from subscriptions, a large-scale test of per-token API pricing. Compute costs fell from 201% to 55% of revenue, and the second quarter brought a $974 million operating profit. The risks are price pressure, dependence on models released in the last six months, and 48% of sales flowing through cloud partners who also compete with it.

Google said fast growth in AI Mode queries weighed on Search revenue because those answers monetize worse, and its stock fell about 15%. That shows the strain on ad-supported AI. Anthropic also renamed its plans to Max at $100 and Ultra at $200 a month, and dropped the 5-hour limit for verified Ultra users. An unverified SemiAnalysis estimate says consumer subscriptions bring about 18% of Anthropic's revenue but use over half its compute, while its API margin is near 92%.

Anthropic's IPO filing shows that cloud marketplaces run by Amazon, Google and Microsoft carried 48% of its revenue in the first half of 2026, up from 11% in 2023. Reuters estimated Anthropic paid about $351 million in fees on $2.16 billion of 2025 marketplace sales, or roughly 16 cents per dollar. For sellers of model APIs, this shows what it costs to reach enterprises through the clouds, and why booking gross revenue makes comparisons with OpenAI tricky.

Google is taking ads in AI Mode to ten countries beyond the US, such as India, Canada and Nigeria, with Brazil, Mexico and the UK due in Q4 2026. Current Search, Shopping and Performance Max campaigns get enrolled automatically, billed through the usual cost-per-click auction at no extra charge, and advertisers cannot opt out. The push for ad-supported AI comes as Search revenue growth slowed to 12% and OpenAI widened ChatGPT Ads to 89 countries.

OpenAI is testing a visual ad format that appears beside ChatGPT image generation. It starts in the U.S. with a small advertiser group. It is also adding brand-suitability checks with DoubleVerify and IAS, conversion data pipelines, and many attribution and incrementality partners, so brands can prove return on spend. The push shows ad-supported AI maturing: since ads launched in February, weak measurement held back monetization of ChatGPT's 1.2 billion weekly users.

OpenAI doubled usage allowances on every ChatGPT Pro tier, priced at $100, $200 and $500 a month, and on Business Premium, while keeping prices the same. It also dropped a plan to shrink the Pro 200 allowance after users pushed back. The move shows how consumer AI subscriptions compete on usage caps rather than price, but OpenAI publishes no exact limits, so buyers cannot check the doubling.

Anthropic's S-1 shows $16.3B of revenue in the first half of 2026, up from $1.16B, with a first operating profit of $974M in Q2 and roughly 900% net revenue retention. Rory O'Driscoll notes about 80% of revenue comes from models under six months old, so per-token API sellers must keep funding costly new models to keep earning. The hosts also argue a US retreat from Chinese open-weight models could hurt inference resellers and give closed labs more pricing power.

Salesforce will begin charging for AI agent calls by volume from November, moving its AI revenue beyond seat-based pricing toward metered consumption. Agent Fabric upgrades add automated testing and cost controls so buyers can track agent spend. The piece notes that per-call fees and mandatory agent registration could make enterprises more cost-sensitive, testing how far an AI add-on to existing SaaS can push usage charges before retention suffers.