This week Anthropic, the company behind Claude, filed its S-1 to list on the Nasdaq. For the first time anyone can read what a frontier AI lab earns, what it spends and who controls it. Revenue in the first half of 2026 was 16.3 billion dollars, against 1.2 billion a year earlier, and the second quarter brought a first operating profit of about one billion. The headline net loss of more than 80 billion dollars is mostly accounting on convertible notes. The two details that matter most to a business buyer are the channel and the control: 48 percent of revenue came through cloud marketplaces, and a long-term benefit trust elects a majority of the board.
Two days later Microsoft reported Azure growth of 48 percent. Its finance chief said customer demand continues to exceed available supply, with some partner capacity slipping into the first half of 2027. Google Cloud revenue grew 103 percent to 30.7 billion dollars in the quarter. Capacity, not capability, is the constraint, and the customers with capacity reserved in writing are served first.
That is the theme of this episode of Future Bytes AI News: treat AI providers as strategic suppliers. Know whether your contract is with the model maker or with the cloud that resells it. Read who controls the supplier. And secure capacity in writing before the season you need it.
The second story is about who gets the best models. Google's Gemini 4 Carbon now ranks first on the Artificial Analysis index, yet it is available only to a group of cyber defenders, with no date for businesses. In the same three days Anthropic's red team showed an unreleased model building working attack chains in Chrome, Android and Linux, with 62 Chrome vulnerabilities patched. OpenAI shipped GPT-6.1 Terra, which it says meets its Critical cybersecurity threshold, to paying customers with its top model's safeguards and at half the price of its predecessor. Three labs gave three answers: gate it, keep it in-house, or ship it with safeguards.
Magnus closes with four questions for your next planning session. What would break first if your main AI supplier changed its price, capacity or owner? Is your contract with the model maker or the cloud, and who do you call when it fails? Which plans assume the very best model, and what is plan B if it stays gated? And is your security team inside the early-access programmes, or will you meet these models first through attackers?
