--- title: "When the Magic Becomes Incremental: Why OpenAI’s Pivot Won’t Solve the Real Strategic Drift" url: https://sam-page.com/thinking/when-the-magic-becomes-incremental-why-openais-pivot-wont-solve-the-real-strateg date: 2025-12-04 author: Sam Page type: Post or article categories: [Emerging tech and AI] reading_time: 4 min summary: "OpenAI’s pivot back to the core won’t fix what’s really going on." first_published: https://www.linkedin.com/pulse/when-magic-becomes-incremental-why-openais-pivot-wont-sam-page-fwihe --- # When the Magic Becomes Incremental: Why OpenAI’s Pivot Won’t Solve the Real Strategic Drift OpenAI’s pivot back to the core won’t fix what’s really going on. OpenAI hitting pause on all the side quests… on the surface it sounds sensible. Back to the core. Back to what made them, them. But if I’m honest, I don’t think this reset touches the deeper problem. And it’s not because the teams aren’t brilliant. It’s because the industry is running head-first into its own limits. Over the past year they’ve stretched themselves into all sorts of horizontal territory. Checkout. Pulse as a personal assistant. Early ads. Even hints of a jobs product. All of it chasing new revenue, not deepening the thing people actually come for. And whenever I see that kind of expansion, it usually means one thing: **the core is starting to flatten.** If you talk to real users, most will tell you the same story. GPT 3 to 4 felt huge. 4 to 5… helpful, yes. But not life-changing. The magic is becoming incremental. And competitors are on the same trajectory. Then Google dropped Gemini 3 and, apparently, everything at OpenAI went code-red. Side projects shelved. Everyone rushed back to the core model. It makes sense. It also feels a bit like rearranging the furniture while the foundations are shifting. Because there’s a deeper issue here that no amount of “try harder on the core” will solve. ### The architecture ceiling ![](https://sam-page.com/img/thinking/when-the-magic-becomes-incremental-why-openais-pivot-wont-so-1.webp) We’re bumping up against the natural limits of this generation of LLMs. You can throw more data at it, more training runs, more of everything. You’ll get gains. But they’re marginal. And marginal doesn’t justify premium pricing forever. If your story becomes “slightly better autocomplete”, it’s very hard to tell a category-changing narrative. You lose the emotional pull. You lose the defensibility. You lose the permission to charge what you used to. ### Compute isn’t a detail, it’s a constraint ![](https://sam-page.com/img/thinking/when-the-magic-becomes-incremental-why-openais-pivot-wont-so-2.webp) OpenAI is still largely dependent on the same chips, the same supply chain that everyone else is battling over. That leaves them fundamentally exposed. Costs wobble. Capacity bottlenecks hit at the worst possible moment. You’re reliant on a tiny handful of vendors who, frankly, have their own incentives. Meanwhile Google, and a growing group of Chinese players, have quietly built their own silicon advantage. And this is the bit that always gets overlooked: controlling your own stack compounds over time. Every efficiency, every margin improvement, every optimisation… they get to keep it. When you zoom out, you can see this anxiety bubbling in the market already. NVIDIA’s wobbling share price. Investors asking who really captures the long-term economics of AI. The whole thing feels slightly overextended. ### The missed Enterprise moment ![](https://sam-page.com/img/thinking/when-the-magic-becomes-incremental-why-openais-pivot-wont-so-3.webp) But the biggest issue, for me anyway, sits in Enterprise. Technology should be in service of people, and the people in large organisations need stability, trust, clarity… not a company sprinting in ten different directions at once. While OpenAI chased consumer scale, Anthropic, Google and even Meta did the slow, unglamorous work: security teams, legal reviews, procurement, risk frameworks. All the bits that create confidence. And unless every report is wrong, Anthropic is now ahead of OpenAI in Enterprise adoption, and Google’s closing fast. The percentages don’t matter. The pattern does. OpenAI is no longer the safe default for long-term, high-stakes AI programmes. And Enterprise is where the durable value lives. Longer contracts. Stickier integrations. Higher switching costs. The stuff that shapes how organisations actually work. A better model doesn’t fix a go-to-market problem. It doesn’t rebuild trust. **Why the pivot won’t be enough** So even if OpenAI ships a stronger model next quarter… even if they retake the leaderboard… they haven’t changed the underlying dynamics. The architecture is flattening. Compute is constrained. Enterprise confidence has drifted. And the business model still leans heavily on undifferentiated API usage and a consumer subscription product that everyone else can copy. **It’s not a bold reset. It’s a necessary stall.** The real question isn’t “how fast can they improve the core?” It’s “can they rethink where and how they compete?” Because the next phase of AI, at least the meaningful phase, won’t be won on marginal model improvements. It’ll be won on trust, on grounded value, on building technology that’s genuinely in service of people, and helping people deploy that technology effectively. That’s what I care about. That’s what I love. And it’s the piece OpenAI still hasn’t fully figured out.