“Did 2025 end badly for OpenAI?” is the wrong question. Here are the 2 questions that do matter.“Did 2025 end badly for OpenAI?” is the wrong question. Here are the 2 questions that do matter.
OpenAI once again weighed on a raft of its suppliers and partners after The Wall Street Journal reported**** that the company has missed a number of internal revenue and user targets, as its competition with Anthropic and others heats up.
The goals missed**** reportedly include a target to hit 1 billion weekly active users by the end of 2025, its annual revenue target for ChatGPT last year, and multiple monthly revenue targets this year, as Anthropic has surged ahead in the enterprise markets.
So, where to from here?
- We spent the final two months of 2025 punishing stocks for being close to OpenAI, and until April, those names remained in the penalty box, lagging the Nasdaq 100 and significantly trailing Google-linked stocks.
- What really matters? Does OpenAI understand why it lost market share among enterprises, and can OpenAI compete on quality? Solve those two questions, and this is just a rough patch. Fail to, and no amount of compute can solve its issues.
- Most of OpenAI’s internal and externalcommuniques in 2026 have taken care to spotlight the growth of Codex (its AI coding tool) and how enterprise revenues are gaining ground on consumer sales within the firm. This appears to be a company that’s better balancing the need for enterprise depth to go along with its consumer breadth.
One thing OpenAI does have going for it is that the best ability is availability. OpenAI has sought to make this a key differentiating and selling point relative to Anthropic. The Claude developer has been bedeviled by complaints about use limits and is in the midst of a mad scramble for compute that’s seen it strike or expand deals withCoreWeave,**** Amazon,**** Google,**** andBroadcom over the past month.
The Takeaway
Commoditization might sound like a bit of a dirty word, or like it’s devaluing the impact of a potentially revolutionary technology. But at its essence, all we’re describing here is the ability of AI tools to produce a (roughly) standardized and reliable output: you don’t think twice about whether the gas you’re putting in your car at Exxon Mobil will be any better or worse than Shell’s. Both get you where you want to be.
To tie these two points together: if Exxon Mobil is closed and Shell is open, well, then there’s really no choice for whose fuel you’ll be using
I still like GPT for research, but, it doesn't produce much different (and valid) findings that Claude won't find on things like code review or impact analysis - for me - anymore, it's mostly less complete. Maybe it's a templating issue (the template being too restrictive) or an instruction structure issue in my setup - can't ever be sure if it's on my end or theirs.
I'm not sure about commoditization though, I'm still thinking differentiation. This is of course not really useful when you've been claiming for years that your little magic trick fixes everything. But, maybe we don't need one solution to fix everything half baked, but instead many solutions that each fix different things really well.
đź‘Ź