As I do at the start of each week, I looked back at recent AI developments. This time, three stories led me to the same question:
Where will the real bottlenecks emerge in the next phase of AI?
Why might AI model development need to slow down?
According to OpenAI, preliminary evidence suggests that one of its Astra models may meet the Critical cybersecurity capability threshold under its Preparedness Framework.
Following the OpenAI-Hugging Face incident, the company paused some reinforcement learning work for two weeks. Its largest planned frontier RL run remains on hold.
What stands out to me is this: model development may need to wait for the security infrastructure around it to catch up.
Who will finance AI infrastructure?
NVIDIA announced a $1.5 billion investment in SB Energy.
The PORTS-Pike campus in Ohio is expected to reach 8 GW of AI computing capacity. According to Reuters, NVIDIA has also agreed to provide up to $105 billion in guarantees supporting OpenAI’s 20-year lease.
Energy, data centres, chips, land and financing are increasingly determining how quickly AI can scale.
Why is the model access layer becoming more valuable?
Dublin-connected Stripe has agreed to acquire OpenRouter. The transaction has not yet closed.
OpenRouter provides access to more than 400 AI models from over 80 providers through a single layer. According to the company, its community includes more than 10 million developers and businesses.
The AI access layer is infrastructure that helps businesses compare models, switch providers and use different models within the same workflow.
Value may emerge not only among companies developing powerful models, but also among those making model choice and portability possible.
What do these developments mean for Ireland?
Stripe’s Dublin connections link a major AI infrastructure deal to Ireland’s ecosystem. Meanwhile, the AI Office of Ireland and the EU AI Act are increasing the importance of safety, governance and responsible adoption.
Who will build the most powerful model? That still matters.
But other questions are emerging: Who can develop AI safely? Who can finance the infrastructure? Who will control access between different models?
Perhaps the greatest value in the AI era will come not only from intelligence itself, but from the systems that make it safe, accessible and scalable.
