Mitesh Agrawal

"the prices of memory does affect the entire ecosystem all the way from HBM to DRAM... the biggest point right now is everyone gets affected, first of all, on the memory side, especially actually the frontier ones, the ones that are using the CoWos, HBM gets even more."

"our silicon will become the first ever terabyte plus memory capacity chip in the entire world. For context, NVIDIA Rubin, that comes out later this year, 384 gigabytes or 0. 4 terabytes of memory. Positron chip that tapes out end of this year world's first 2. 3 terabyte memory or 2, 300 gigabyte chip"

"it's anywhere between 60 and 70 % of the total cost of owning the hardware, actually goes to the power cost. And now, we're seeing H100s could last eight to 10 years. This hardware is lasting a lot longer than people maybe initially thought. And so, it's the power cost that is very expensive over the cost of the hardware."

"I'm an energy maximalist, we need as much energy, but this is where putting Positron into the part of the story is we are a very energy- efficient chips. For every watt of power, we are basically trying to get three, four, five times more output for inference. And this is how we find a story into this overall narrative around, well, we need as much energy, but for the given amount of energy, if you can drive more output, the better it is for the production growth that is out there continuously."

"I think the world simulation models will be very interesting of not only just having a true physics engine at its core, but then, hey, do they open up new challenges for ecosystems for silicon, right? Do they work great on Positron or d- Matrix or NVIDIA or you need something else for those?"

"the biggest application growth is video generation and world simulation model, which will feed into robotics, training datasets and all those things."

"Only 30 % of that in 2024 was actually for inference. Now go to 2025 and it's closer to 60 to 80 % of their compute spend is now on inference. So we're going from a world of spending a lot of money on training to spending a lot of money on inference"

"you're seeing a whole class of AI inference chips that are coming into the space saying... Obviously, NVIDIA is the absolute behemoth in the space, but then all of this chip companies are really coming through and saying that, 'Look, depending on the workloads, this is going to get more and more bigger and bigger because every workload, the more dollars you save, the more electricity you save, the better it is for those workloads to grow.'"