A concentrated 48 hours of announcements has clarified the three parallel contests that will define competitive AI in Q4 2026: the product and platform race, the capital and infrastructure race, and the emerging physical-world layer. Each moved decisively this week.
OpenAI DevDay: Dots, an Agent Runtime, and $70 Billion in Annual Revenue
At its annual developer conference in San Francisco on 29 September, OpenAI unveiled more than 20 updates across ChatGPT and its Codex coding assistant. The centrepiece is Dots — always-on agents, each running on GPT-6 Astra, provisioned with their own cloud computer and browser, working continuously on goals you assign. Dots connect to more than 4,000 applications through OpenAI’s plugin ecosystem and require owner approval before taking consequential actions. They are rolling out now to Pro and Business Premium subscribers; Enterprise and Education workspaces follow after an administrator opts in.
For builders, the headline announcement is the Agents API, now in public beta: a managed runtime where OpenAI handles session state, orchestration, context compaction, and fault recovery while the developer defines tools and environments. Agents can invoke an OpenAI-hosted browser for computer use, connect to external MCP servers, spawn parallel subagents, and maintain durable cross-session state. Alongside it, OpenAI launched Plugin Extensions — a distribution channel for developers to build full applications (dashboards, editors, complete workspaces) that surface natively inside ChatGPT and Codex. A new Ultrafast speed tier generates tokens up to eight times faster in Codex and up to six times faster through the API; it is bundled in the new Pro 500 plan alongside the highest usage allowances OpenAI currently offers.
Separate from DevDay, Axios reported that OpenAI’s annualised revenue run rate has reached nearly $70 billion, up more than 70 per cent since the start of Q3. B2B revenue more than doubled in the same period, and consumer revenue earned in Q3 alone exceeded what the company generated from consumers throughout all of 2025. For operators evaluating platform concentration risk, the gap between OpenAI’s commercial velocity and every other model provider is now measurable rather than merely asserted.
Anthropic’s Leaked IPO Filing: $518 Billion and an Existential Warning
Anthropic’s confidential IPO prospectus circulated publicly on 29 September, disclosing the full scale of its infrastructure commitments. The company plans to spend at least $518 billion over the next decade through supply agreements with six partners: at least $111.1 billion with Google, $110 billion with Amazon, and $31.4 billion with Microsoft, with three further partners accounting for the remainder. Approximately 80 per cent of those obligations are non-cancelable or payable regardless of usage — structural costs that will persist even if revenue growth slows.
Revenue grew 12-fold in 2025 to nearly $4.6 billion, even as net losses reached $42 billion, largely from liability writedowns tied to prior fundraising rounds. The 261-page prospectus devotes roughly 80 pages to risk disclosures — nearly double the 48 pages covering the business plan. One passage explicitly acknowledges that Anthropic’s own models could pose an “existential risk to humanity.” A potential IPO at a valuation above $2 trillion would represent more than a doubling from the $965 billion Anthropic reached in May. What the filing makes structural is the hyperscaler dependency: Anthropic competes with Google DeepMind and OpenAI while simultaneously being locked into decade-long compute contracts with the same cloud providers. The circular arrangement is not incidental — it is the only financing mechanism the current model of frontier AI permits.
AMD Acquires World Labs for $8.2 Billion
AMD announced a definitive agreement to acquire World Labs, the San Francisco spatial-intelligence lab founded by Fei-Fei Li, for $8.2 billion in an all-stock transaction. Closing is expected by year-end pending regulatory review. Li will join AMD as executive vice president and chief scientist. World Labs develops models that generate, reconstruct, and simulate interactive 3D environments from text, image, and video inputs, alongside technology for robotic learning and simulation. The company had an existing inference optimisation and training partnership with AMD before the deal. The $8.2 billion figure is AMD’s second-largest acquisition on record, after the roughly $50 billion paid for Xilinx in 2022.
The strategic logic is a wager on the next layer of AI value. As inference moves from data-centre servers into robots, autonomous vehicles, and physical automation environments, the ability to simulate and navigate the physical world becomes a differentiating capability. AMD has consistently trailed NVIDIA in raw training-scale silicon; World Labs gives it a research and product position in spatial AI that NVIDIA does not yet own cleanly. For operators running or considering robotics and computer-vision pipelines, this acquisition signals that the hardware and model layers are converging at AMD’s level, not just at the hyperscaler level.
Three separate competitive contests advanced sharply in 48 hours. OpenAI is moving to own the application and orchestration layer with Dots and the Agents API. Anthropic is locking in a decade of compute capacity and preparing to access public markets. AMD is buying the physical-world intelligence stack. An operator deciding where to place infrastructure bets today faces a landscape where the winning position may require a view across all three.