Three games being played in AI

From the outside the three frontier labs look like competitors in the same market. Same models, same benchmarks, same launch cadence, the same press cycle. Open their pricing pages side by side, though, and the strategies separate cleanly. They aren't playing the same game; they're playing three different ones, against three different definitions of what "winning" means. Pricing and packaging are the cleanest tell — a company can fudge a keynote but it can't fudge what it actually sells.

Anthropic — depth on the operator

Anthropic is playing one user, very hard: the operator. An engineer, analyst, or founder using the model through an agentic harness to do their own work. The whole product surface optimises for that person. Claude Code is the wedge — a serious agentic coding tool that runs against the same models the API serves — and the entire packaging architecture is shaped around getting it in front of operators without a metered bill.

You can read this off the plans. Claude Max ($100 / $200 a month, April 2025) bundled Claude Code into a flat fee — and the enterprise API, which feeds the same product into companies, now drives the majority of revenue. Two sales motions, one product, one user class. What Anthropic conspicuously isn't doing is fighting OpenAI for the consumer chat default. They've ceded that surface to focus on the customer with the highest willingness to pay per seat.

The numbers are the loudest endorsement of any AI strategy on the board. Anthropic ARR went from roughly one billion in December 2024 to thirty billion by April 2026 — Amodei put the year's growth at eighty-fold. Claude Code alone hit $2.5B annualised by February 2026 and accounts for more than half of enterprise spend. Valuation tracked the same curve: $61.5B in March 2025, $380B in February 2026, in talks at $900B as of this month — potentially overtaking OpenAI's $852B. Reuters ran the headline plainly: Anthropic ARR surges on Claude Code strength.

OpenAI — breadth on the brand

OpenAI is playing the opposite game. ChatGPT is one of the most recognised consumer brands of the decade, with seven hundred million weekly users, and the strategy is to make it the default front door to AI for everyone. Pricing reflects that: ChatGPT Plus at $20 for consumers, Pro at $200 for power users, plus Team, Enterprise, and the API. Pro was the first $200 plan in the market (December 2024) but it was, at launch, a heavy chat tier — Codex was folded in later, in response to the operator-shaped move Anthropic had made.

The breadth is everywhere else too. Sora for video. The Jony Ive hardware partnership. Operator and Atlas pushing into the browser and agentic surfaces. The bet isn't depth on one user class — it's that the ChatGPT brand can host every adjacent product category as it emerges. It's a strategy with a real moat (brand and distribution at consumer scale are not cheap to replicate) and a real risk (a spread bet has to win in many places to justify itself).

The packaging tells you where it is and isn't working. Pro reportedly contributes under six per cent of consumer revenue — the $200 tier is real but not the engine. The engine is breadth: Plus subscriptions at scale, enterprise seats, API usage from the long tail of products built on the brand. ARR is in the $12–20B range tracking through 2025, with valuation at $852B as of March 2026. Bigger than Anthropic in absolute terms, growing at a third of the rate.

Google — defence on distribution

Google is playing a third game entirely, and the easiest to misread. Gemini isn't really a standalone product line; it's a feature embedded into the surfaces Google already owns. Search (AI Overviews), Workspace (Gmail, Docs, Sheets), Android, Chrome, Cloud, Vertex AI. The model goes where the user already is.

Read the AI Ultra plan as evidence of this. At $250 a month, it includes Gemini Code Assist, the Gemini CLI, twenty-five thousand AI credits — and a hundred dollars of monthly Google Cloud credit usable against Vertex AI or the Gemini API. That's not an operator plan in the Anthropic sense. It's a Cloud cross-sell dressed up as a consumer plan. The job of the bundle is to keep you inside Google Cloud, not to win you on the merits of the model.

Strategically this is defence. Google's real exposure to AI isn't losing a model race — it's losing Search to a chat-first front door, or losing developers to a different cloud. Embedding Gemini everywhere they already have distribution removes the need to win on standalone Gemini at all. It's slower than Anthropic's depth and quieter than OpenAI's breadth, but it's the only one of the three strategies that doesn't need to be first to be safe.

The scoreboard, one year in

A year in, depth is winning on the metrics that move fast. Anthropic identified one customer with high willingness to pay, built the tool that fit them, and priced it so the meter didn't get in the way. The revenue curve and the valuation curve both bent. Breadth and defence aren't losing — they're playing for slower things that don't register in twelve months. Consumer brand monetisation compounds across decades. Distribution moats erode at the pace of habit change, which is also slow. Twelve more months will look different again.

The more useful question isn't which strategy is right — all three are coherent for the company playing them. The useful question is which game you're playing as a buyer, because the answer changes what you're actually purchasing. From Anthropic you're buying a working surface for a specific class of worker. From OpenAI you're buying a front door to everything. From Google you're buying AI inside the tools you already use. The right answer depends on who's using it and what for — which is, as ever, the question worth asking before you reach for the default.