OpenAI or Anthropic?

My thesis: OpenAI is the category king in the race to become the default AI platform for work and everyday life. Anthropic is #2 in that platform thesis. OpenAI’s competitive advantage comes from distribution, product breadth, and the customer relationship, which extends positive compounding and capabilities over time.

• OpenAI has over 10% of the human population. OpenAI reports more than 1.2 billion people using ChatGPT each week. That gives it an enormous starting point for every new product. It can introduce agents, tools, and services to people who are already there. Distribution at that scale is hard to recreate. OpenAI, October 7
• ChatGPT is a known consumer brand. Its familiarity is an asset when people try a new agent, a business adopts AI, or OpenAI introduces a device. It can extend an existing customer relationship across more of daily life. OpenAI, October 7
• Other companies can build inside its product. OpenAI’s plugin extensions support sidebar apps, panels beside conversations, and file viewers and editors. That gives developers a reason to bring their products into ChatGPT. My bet is that useful third-party software makes ChatGPT more valuable, which attracts more users and more developers. Plugin extensions
• It has more ways to become useful throughout the day. Voice, image generation, and interactive answers extend the product beyond typing into a text box. Intelligent UI began rolling out to paid tiers on October 7, with Free and Go scheduled to follow on October 8. Availability varies by feature and account. Each is another reason to stay in ChatGPT. Voice, image generation, Intelligent UI
• It is moving up the stack into the work itself. Chat handles everyday questions. Work handles longer tasks and deliverables. Codex handles software development. These give OpenAI several ways to serve the same customer as their needs grow. The strategic prize is becoming the place where the work happens. ChatGPT Work and Codex
• Personal agents could deepen the relationship. OpenAI introduced dots as always-on agents for ongoing responsibilities, with access limited by plan, market, and workspace settings. If these agents become useful daily companions, OpenAI gets a stronger position in how people organize their lives and work. That is a bigger opportunity than answering the next question. DevDay recap
• The developer platform reaches beyond ChatGPT. The Agents API supports tasks in an OpenAI-hosted browser. Developers can build their own applications around that capability while OpenAI supplies the underlying service. OpenAI can benefit both when customers use its products and when other companies build products on its platform. Agents API computer use
• It has a route into large organizations. OpenAI’s enterprise strategy includes Frontier and deployment partners such as Accenture, BCG, Capgemini, and McKinsey. Those relationships matter because company-wide adoption requires integration, governance, and changes to how people work. A good model alone does not finish that job. OpenAI’s enterprise strategy
• Its ads platform turns attention into another revenue stream. ChatGPT Ads gives businesses a way to reach people as they explore products and make decisions. OpenAI is expanding attribution and measurement, with a visual ad test scheduled for later in October. Subscriptions, enterprise contracts, API usage, and ads give it several ways to fund the platform. ChatGPT Ads
• Forthcoming hardware gives it an internet access device, thereby disintermediating Apple. The io team merged with OpenAI, while Jony Ive and LoveFrom remain independent design partners. A dedicated AI device could put OpenAI closer to people throughout the day and reduce its dependence on other companies’ interfaces. That is future upside from a product still in development. Sam and Jony
• Compute access is a structural advantage. In April, OpenAI said it had surpassed its original goal of securing 10GW of U.S. AI infrastructure through Stargate. Secured capacity includes future deployment. As it comes online, that infrastructure can support more research, more customers, and more work per customer. Stargate infrastructure
• It is designing its own chips. OpenAI unveiled Jalapeño, its custom inference accelerator, with Broadcom and Celestica in June. Engineering samples were running workloads in the lab, with initial deployment targeted by the end of 2026. Designing hardware around its models could give OpenAI more control over cost, speed, and supply. Jalapeño
• The talent advantage reaches across the whole stack. OpenAI combines model research and product engineering with a hardware program led by Richard Ho and design work with Jony Ive’s team. My talent-density argument is about that combination: people who can improve the model, build the product, and design the hardware working toward the same customer experience. Hardware team, Design team
• Its access to capital gives it room to build. OpenAI closed a funding round with $122 billion in committed capital in March, backed by investors including Amazon, NVIDIA, SoftBank, and Microsoft. That scale of financing supports long-term bets on compute, research, products, and hardware. Committed funding gives it options; execution determines what those options are worth. Funding announcement
• I prefer the product story centered on useful work. OpenAI’s Work and Codex positioning gives buyers a practical question: what can this do for me? Anthropic explicitly researches model welfare and possible consciousness, while saying these questions remain uncertain. I prefer keeping that debate out of the buying decision and focusing on capability, reliability, control, and results. Work and Codex, Anthropic’s position
• Anthropic is a serious challenger with a strong position in valuable work. Claude Code and Cowork give it a credible path across software development and business functions. PwC’s announced rollout and Barclays’ expanded collaboration show why it cannot be dismissed. I put Anthropic second on the broader platform opportunity because OpenAI combines mass consumer reach with an expanding place for people, developers, and businesses to work. That does not make Claude the second-best tool for every job. PwC, Barclays

Your vendor selection shows your organization's capability at recognizing category kings and aligning for future success. If you made the wrong choice, perhaps that's recognition that you require outside help from deep expertise who understand mathematics, AI, and enterprise value creation at a foundational level.

At some point the entire org is going to have to align on a single provider, and get everyone using the same personal agent platform. Time to shift now if you got it wrong because Shared Memory and agent-to-agent communication is here, today.

— Sean