← Back to feed

Anthropic has roughly 134 million monthly active users. OpenAI has roughly 900 million. Anthropic now generates more revenue. The Counterpoint Research analysis published in The Register makes the arithmetic straightforward: Anthropic earns $16.20 per user per month. OpenAI earns $2.20. That 7.4x gap, multiplied across the respective user bases, fully explains the revenue crossover that Bloomberg confirmed on April 8 when Anthropic’s ARR cleared $30B against OpenAI’s $24B.

The per-user figures for the rest of the market:

Company	Monthly Revenue Per Active User
Anthropic	$16.20
Microsoft	$5.00
OpenAI	$2.20
Google	$1.10
Meta	$0.10

What the ARPU Gap Is Actually Measuring

The difference is not pricing power in the abstract. It is the composition of who is paying.

Anthropic’s revenue is approximately 80% API, 20% consumer subscription. OpenAI’s is roughly the inverse: around 70% consumer (ChatGPT subscriptions and Plus plans) and 30% API. Enterprise API customers paying $1M or more per year generate fundamentally different per-user economics than consumer users on a $20 monthly plan or a free tier.

Anthropic’s enterprise segment has more than 1,000 customers at the $1M+ annual spend level — a number that doubled in under 60 days in early 2026. Eight of the Fortune 10 run on Claude. Those contracts are volume-based, multi-year, and grow automatically as the embedded workloads expand. A single enterprise agreement can represent 50,000 or more consumer-equivalent users in revenue terms.

Why Consumer Scale Stopped Compounding

OpenAI’s $2.20 ARPU is not a failure of pricing. ChatGPT charges $20/month for Plus and up to $200/month for Pro. The problem is the denominator: 900 million weekly active users includes a large base on the free tier that generates no direct revenue and consumes compute. The monetisation rate on consumer AI remains low because most casual users do not pay and most paid users are on flat-rate subscriptions that cap upside.

Meta’s $0.10 ARPU illustrates the extreme: 600M+ users across its products, essentially all on the free tier, with no enterprise API revenue to offset the compute cost. Google’s $1.10 reflects a hybrid position — consumer Gemini scale plus growing Workspace AI and Cloud enterprise, but with the free-tier drag of a consumer search company.

The Compounding Advantage

Enterprise ARPU compounds differently than consumer. A company that embeds Claude into its internal codebase, legal review workflow, or customer-facing product does not churn when a cheaper consumer option appears. The switching cost is integration depth, not preference. This is why Anthropic’s ARR tripled in roughly four months — from $9B at end-2025 to $30B by April 2026 — while OpenAI’s grew from $20B to $24B over the same window.

The Counterpoint data also has a product-mix implication for the AI field broadly. Microsoft’s $5.00 ARPU reflects Copilot embedded in enterprise M365 licensing — a distribution channel that charges per seat across large organisations. That is closer to Anthropic’s model than OpenAI’s, which explains why Microsoft’s per-user economics are 2.3x better than its primary AI partner.

What Changes If This Pattern Holds

The $16.20 vs $2.20 ARPU split matters for the next 12 months because it determines whose infrastructure spend is covered by actual cash flow. Anthropic’s $30B ARR at 134M users produces substantially more revenue per compute unit consumed than OpenAI’s $24B across 900M users. Enterprise API traffic is more predictable, more batch-processable, and less likely to generate the unpredictable spike load of consumer viral events.

The company with better ARPU can buy more compute per dollar of revenue, run it more efficiently, and reinvest faster. That is the compounding dynamic that OpenAI’s CFO Sarah Friar is now reportedly worried about: the compute commitments were sized for a revenue trajectory that assumed consumer scale would eventually monetise at enterprise rates. The Counterpoint data suggests that assumption has not held.