Nvidia kicked off GTC 2026 last week with 30,000 attendees from 190 countries. Jensen Huang took the stage for a two-hour keynote and the big reveal was a new chip architecture codenamed Feynman. Built on 1.6 nanometre technology with something called silicon photonics, which means data inside the chip moves at the speed of light instead of through copper wires.
Feynman chips will not hit data centres until 2028. So why should you care about a chip that is two years away when you are running a 15-person business today? Because every AI tool you use runs on Nvidia hardware, and when the hardware gets cheaper and faster, the tools get cheaper and more capable. The cost of running Claude, ChatGPT, and Gemini is directly tied to the price of the chips underneath them.
How chip costs translate to your monthly bill
This is something I try to explain to clients whenever they ask about AI pricing. The reason these tools cost what they cost is largely down to inference, which is the compute power required every time the AI processes your request. When Nvidia pushes inference costs down, the AI providers either pass those savings on through lower prices or give you more capability for the same price. Usually both.
The jump from last year's Nvidia chips to the current generation already made Claude and ChatGPT noticeably faster and cheaper to run. Anthropic doubled everyone's Claude usage limits just last week, partly because the infrastructure costs are falling. When Feynman arrives in 2028, the drop will be even more significant. Things that are currently too expensive to run in real time, complex multi-step agents, real-time video analysis, continuous background processing, will become affordable for everyday business use.
Planning today for tools that arrive in 18 months
I realise "plan for what the tools will do in 18 months" sounds like vague futurism, but hear me out. If your team builds AI workflows now, on today's tools at today's prices, you are developing two things at once. You are getting immediate value from the workflows. And you are building the muscle, the habits, the processes, and the operational knowledge that will let you take full advantage of the next wave of tools the moment they arrive.
The businesses that waited until ChatGPT was "good enough" before they started are now scrambling to catch up with teams that started building a year earlier. The same thing is going to happen when AI agents become cheap enough to run continuously. The teams that have already learned how to integrate AI into their operations will adapt in days. The ones that are still dabbling will need months.
What this actually means for your budget
The practical takeaway is this. AI tools are going to get meaningfully cheaper and more powerful over the next two years. If you are currently holding back because the cost feels hard to justify, that calculation is going to shift in your favour. The question is whether you will be ready to take advantage of it when it does.
An AI subscription that costs you a few hundred pounds a month right now is buying you two things: the immediate productivity gain, and the learning curve. Your team is figuring out how to work with AI. They are developing instincts about where it helps and where it does not. That operational knowledge is worth more than the subscription fee, because when the tools get 10x cheaper and 10x more capable, the teams that already know how to use them will move at a completely different speed.
Nvidia's roadmap tells you where this is heading. The only question is whether your team will be ready when it arrives.
I help businesses build AI workflows that deliver value today and position the team for what comes next. If you want to start building that foundation now, here is how the fractional AI engagement works.
