Copy article

OpenAI's Enterprise Growth Victory Lap Masks Deeper Sustainability Questions

ended 09. December 2025

OpenAI's 2025 enterprise report shows dramatic adoption numbers: 8x message growth, 320x API token consumption, 19x custom GPT deployment. But buried in their own data is a more troubling story about the gap between capability and actual integration. 19% of monthly active enterprise users have never touched data analysis tools, 14% have never used reasoning features, and 12% have never tried search despite paying for them.

The report frames a "growing divide" between "frontier" and "laggard" workers, with frontier users sending 6x more messages and 17x more coding messages than median workers. This suggests  the creation of AI-dependent power users while the majority of paid seats generate minimal value. OpenAI's own chief economist admits the “primary constraints are no longer model performance or tooling, but rather organisational readiness”, which translates to “we've built capability most companies can't actually deploy.”

Most revealing: survey respondents report saving "40-60 minutes per day" but the methodology explicitly excludes time spent learning systems, crafting prompts, or correcting AI output. 

When only 25% of enterprises have enabled system integrations to give AI context-aware access to company data, and the report admits AI adoption requires treating it "almost like an operating system... basically a re-platforming of a lot of the company's operations," we're not discussing productivity tools, we're discussing infrastructure transformation with unclear ROI and substantial vendor dependency risk.

We want your views:

  • Should "time saved" metrics exclude the hours spent learning, prompting, and correcting AI systems—or does that misrepresent actual productivity impact?
  • When 19% of paying monthly users never touch core features like data analysis, what does "adoption" actually measure, capability purchased or capability deployed?
  • What happens when 75% of enterprises haven't enabled system integrations, yet AI requires "re-platforming company operations" to deliver promised value?
  • Should the 36% increase in non-technical coding be celebrated as capability expansion or flagged as security risk when OpenAI's own data shows most users avoid advanced evaluation features?
  • Where's the independent research validating that 320x token consumption growth correlates with business outcomes rather than experimentation burn-rate?
  • How do organisations measure true business impact when suppliers define "frontier firms" as those generating most messages, not necessarily most value?

3 responses from the Newspage community

Copy all

Star Quote
Copy

Colette MasonVerified
AI Ethics Consultant at Clever Clogs AI
Here's what OpenAI's report accidentally reveals: embedding AI isn't software deployment. It's open-heart surgery on your entire operation, performed on a platform that hallucinates, hijacks your processes, and changes fundamentally every few weeks.

The "frontier vs laggard" framing is patronising. Those workers aren't resistant to change, they're trying to get last quarter's tools stable enough to rely on before the next breaking update lands. You don't have capacity to learn the shiny new reasoning feature when you're spending an hour daily correcting output the survey conveniently doesn't measure.

AI requires complete workflow redesign on platforms you don't control, that change without warning, and can vanish when third-party infrastructure fails. Recent outages prove when your "system" goes dark it means total paralysis.

People are avoiding advanced features because they're still trying to stop the basic ones lying to staff or customers.
Copy

Rohit Parmar-MistryVerified
Founder at Pattrn Data
Claiming employees save '40-60 minutes a day' while conveniently excluding the hours spent wrestling with prompts and fact-checking hallucinations is pure fantasy. In the real world, that 'setup' time is the job. You can't celebrate the destination if you ignore the traffic jam to get there.

And let's be honest about this '320x token growth.' That isn't necessarily a productivity metric; it’s a burn rate. If 75% of enterprises haven't integrated these tools with their actual systems, they aren't transforming operations, they’re just paying for the world's most expensive autocomplete.

The divide isn't between 'frontier' and 'laggard' workers; it's between practical application and corporate theatre. Sending 6x more messages doesn't prove you're a power user; it might just mean the AI didn't understand you the first five times. We need to stop measuring success by server load and start measuring it by profit margins. Anything else is just vanity metrics for the boardroom.
Copy

Patricia McGirrVerified
Founder at Repossession Rescue Network
OpenAI’s report reads like an evangelical proclamation of success, but it exposes something far more serious. Businesses are embedding AI without the governance or stability needed to trust it. A handful of power users drive the numbers while the majority try to stop the basics hallucinating or veering off track. That’s not transformation. It’s vulnerability at the core of every business trumpeting AI as the fix to their problems.

We’re being pushed to rebuild operations on platforms that shift, break or vanish without warning. When the system falters, workers absorb the fallout, correcting outputs and carrying risks the metrics conveniently ignore. The fever pitch of AI adoption is miles ahead of robust oversight. Many organisations know they need stronger governance but don’t yet know how to build it. Or worse, they hope AI will fix that too. Until governance becomes the foundation, not the afterthought, this behemoth of tech innovation exposes businesses to real risk.