Extropic Reveals Thermodynamic Compute Thesis for Physics Based AI Chips

Extropic has come out of stealth and introduced a big splash in processing AI. Introducing a hardware platform to accelerate AI processing power. You can find their Litepaper here: https://www.extropic.ai/future

What Extropic has proposed is using superconducting processors to run probabilistic algorithms on their new physics-based chips. This allows for energy-efficient broad market processing power in us of generative AI. This can significantly lower and distribute the cost of GPT’s.

At the moment processing an LLM is extremely expensive and inefficient even with dedicated devices.

While quantum computing is the biggest force multiplier, it’s time to developmentin it’s current state and costs put this in the when/then category. Extropic’s solution may be a much better short term solution to our growing computing needs.

A bottleneck for processors today is shuffling numbers around in memory. This eats up about 25% of the processors time and limits optimization around specific task compute. For general tasks, it’s okay. However, having dedicated AI processing chips on traditional computers and mobile devices will speed up the AI revolution and bring in a new era of architecture design.

-jm