Fidji Simo announced her departure from OpenAI’s senior leadership the way many people learn difficult news about their own bodies: gradually, and then all at once. A three-month medical leave did not go as hoped; recovery will be longer and more complicated than expected, and the company will move forward without her. The news, dry and personal, flew under the radar, but it says a lot more than it lets on.

Simo’s exit is not an isolated event. Over the past two years, OpenAI has seen key figures leave: from Ilya Sutskever to several product executives, while Greg Brockman redefined his role. The governance, already shaken by the board coup in 2023, continues to show cracks. For anyone building applications on top of large language models – or buying access via API – the question is no longer whether OpenAI will survive, but how much it costs, in terms of risk, to depend on an unstable vendor.

This is where the reasoning goes beyond corporate gossip. Companies that handle sensitive data, operate in regulated industries, or simply don’t want to wake up one day to a change in pricing or model access policies are already shifting workloads toward on-premise deployment. The turbulence at OpenAI’s top accelerates this transition: if the commercial counterparty can change shape from one quarter to the next, direct control over infrastructure becomes a strategic asset.

You don’t need to be a tech giant to move. Frameworks like vLLM, Ollama or LM Studio now allow you to serve quantized LLMs on relatively affordable hardware. A GPU with enough VRAM, a bare metal server in colocation, and you get local inference without depending on an external entity. It’s not just a cost issue – TCO must be carefully assessed – but a sovereignty play: data stays within the corporate perimeter, latency drops, and compliance with GDPR or sector regulations becomes simpler.

The makers of chips and on-premise AI infrastructure – from GPU suppliers to specialized system integrators – find a powerful sales argument in this chronicle. OpenAI’s instability doesn’t just bruise its brand; it legitimizes the idea that critical artificial intelligence should be kept in-house. And while mature cloud alternatives exist (AWS Bedrock, Azure AI), on-prem offers a guarantee no service-level agreement can match: control over the runtime and the data.

We don’t know who will take Simo’s place, nor whether OpenAI will manage to stabilize. But the accumulation of departures paints a picture that technology decision-makers cannot ignore: AI as a commodity API has a breaking point, and the answer is coming from increasingly quiet and powerful server racks, inside the walls of those who use them.