Keep pulling the thread on Joseph Costello.
CognitChip's AI technology aims to reduce chip design time by 50% and the required resource commitment by up to 75%.
The process to design and build a new chip can cost $300 million and take four years to go into service.
GPU architecture is no longer optimal for AI workloads, as its fixed ratio of compute, memory, and memory bandwidth is out of step with the needs of advanced models.
The next generation of large language models is rumored to have 5 trillion parameters.
Some European countries have implemented a moratorium on the construction of new AI data centers due to concerns about stressing their power grids.
The world needs new AI hardware architectures that are memory-centric, allowing compute to be scaled independently to improve power efficiency and cost.
The GPU is the wrong architecture for AI workloads and would never have been designed for this purpose from first principles.
InPho is the first company to integrate Indium Phosphide transistors and photonics on a single die.
No silicon photonics device has ever integrated a silicon transistor with the photonic components on the same die because the manufacturing processes are incompatible.
The next major innovation in computing will come from system-level architectural rethinking, not from incremental improvements in silicon.
Lux Capital has over $6.5 billion in assets under management.
Lux Capital's investment strategy includes writing checks ranging from $100,000 to $100 million, with a focus on seed and series A rounds.