- OpenAI is deprioritizing Sora and scaling back its AI hardware ambitions to refocus resources on revenue-driving products.
- The company’s “core” business is increasingly coding, because developers and enterprise customers generate the most consistent, high-value usage and spend.
- Sora is framed as technically impressive but commercially unviable due to high compute costs, inconsistent outputs, and a lack of enterprise buyers willing to pay enough.
- Enterprise API deals and premium tiers like ChatGPT Pro underpin the business model, while consumer ChatGPT remains large but converts poorly to paid.
- OpenAI’s pivot intensifies competition in AI coding tools against players like Anthropic, Cursor, and GitHub Copilot, with OpenAI betting on distribution and sticky integrations.
OpenAI quietly scrapped several of its most visible projects this month to concentrate on coding tools and enterprise customers. Sora, the video generation model that dominated AI discourse for most of 2024 and 2025, is being deprioritized. The hardware ambitions, including the much-discussed partnership with Jony Ive, are being scaled back. Internal teams are being redirected toward what the company calls its "core" business.
The core business, it turns out, is code.
The quiet pivot to code
OpenAI built its brand on the promise of general-purpose AI. ChatGPT was supposed to be the interface to everything: writing, research, analysis, creativity, coding, conversation. Sora was supposed to upend video production. The hardware project was supposed to create a new category of AI-native devices.
None of these side projects generated revenue proportional to their cost.
ChatGPT Pro at $200/month and the enterprise API are where the money comes from. And within those revenue streams, coding is the dominant use case. Developers write more prompts, use more tokens, and pay more consistently than any other user segment.
OpenAI looked at its revenue data and made the rational decision: stop spreading resources across speculative projects and double down on the customers who actually pay.
| Project | Status | Why |
|---|---|---|
| Sora (video) | Deprioritized | High compute cost, low revenue, no clear enterprise path |
| Hardware (Ive partnership) | Scaled back | Long development cycle, uncertain market, capital intensive |
| Coding tools (Codex, API) | Doubled down | Highest revenue per user, enterprise demand, sticky integrations |
| ChatGPT consumer | Maintained | Large user base but low conversion to paid |
Following the money to enterprise code
The enterprise developer market is where the economics make sense. A single enterprise contract for API access can be worth millions annually. A consumer ChatGPT subscription is $20/month. The math isn't subtle.
Codex 5.3, released in February, was optimized for autonomous code execution. GPT-5.4, released in March, shipped with native computer use. The product roadmap has been pointing toward developer tooling for months. The organizational restructuring just made the strategy explicit.
This puts OpenAI in direct competition with Anthropic's Claude Code, Cursor, GitHub Copilot (which ironically uses OpenAI's models through Microsoft), and a growing ecosystem of AI coding tools. The market is crowded, but OpenAI's advantage is distribution: millions of developers already use their API, and switching costs for embedded integrations are real.
Sora was a demo, not a business
Sora was technically impressive and commercially unviable. Generating video requires enormous compute per output. The results, while visually striking, weren't reliable enough for professional production workflows. And the market for AI video generation turned out to be smaller than the hype suggested: most video production still requires human direction, editing, and iteration that current AI can't handle autonomously.
The real problem was that Sora didn't have an enterprise buyer. Consumer creators wanted it for social media content, but they won't pay enterprise prices. Film and TV studios were interested but couldn't use it for production-quality work. Advertising agencies explored it but found the output too inconsistent for client-facing deliverables.
Without an enterprise buyer willing to pay proportional to the compute cost, Sora was a technology demonstration, not a business. OpenAI chose to stop funding the demonstration.
The hardware retreat
The Jony Ive hardware partnership generated massive press coverage and almost no product. The vision of an AI-native device that replaces the smartphone is appealing in the abstract and punishingly difficult in practice.
Hardware requires supply chains, manufacturing partnerships, retail distribution, inventory management, customer support, and warranty obligations. These are capabilities that OpenAI doesn't have and would take years to build. Meanwhile, Apple shipped the M5 MacBook Air with neural accelerators in every GPU core, and Samsung announced 800 million Gemini-equipped devices by year end. The hardware market for AI isn't empty. It's dominated by companies with decades of manufacturing expertise.
OpenAI stepping back from hardware is an acknowledgment that competing with Apple and Samsung on devices is a distraction from competing with Anthropic and Google on models and developer tools.
Enterprise code is the only AI business that works
OpenAI's pivot is a signal about where AI value accrues. Consumer AI products are expensive to run and hard to monetize. Enterprise developer tools are expensive to build but generate recurring revenue from customers with high switching costs.
Every major AI company is arriving at the same conclusion through different paths. Anthropic has always been enterprise-first. Google is pushing Gemini into Workspace and Cloud. Microsoft is building its own foundation models for enterprise products. And now OpenAI, the company that defined consumer AI, is redirecting toward enterprise code.
The consumer AI market isn't disappearing. ChatGPT will continue to exist. But the investment and talent are shifting to enterprise, because that's where the revenue justifies the compute cost.
For developers and engineering teams, this is good news. More competition in the AI coding tool market means better products and lower prices. For consumers who hoped that AI would transform creative work, video production, and personal computing, the message from OpenAI's pivot is less encouraging: those use cases will get attention when someone figures out how to make them profitable.
If you’re building with OpenAI, expect faster iteration and better support around coding agents, API workflows, and enterprise integrations, while video generation and experimental hardware may move slower or stay limited. Teams evaluating AI vendors should compare coding-tool ecosystems (model quality, tool execution, security, and switching costs) and avoid betting critical roadmaps on flashy demos without clear enterprise pricing and reliability. If you’re a creator excited about Sora-like tools, plan for alternatives and assume access, quality, or pricing could remain volatile until there’s a sustainable business model.