Keep pulling the thread on Sulaiman Ghori.
Sulaiman Ghori believes xAI's biggest competitive advantage is its hardware deployment, which he asserts is significantly ahead of competitors.
xAI is able to produce new pre-trained model iterations for its MacroHard project daily, and sometimes multiple times per day.
xAI is capable of starting training runs on a newly installed rack of GPUs within a day, and in some cases, within a few hours.
Sulaiman Ghori believes that to be competitive in AI, companies cannot rely on public cloud providers like Amazon or Google for their compute infrastructure and must build it themselves.
xAI completed the construction of its first Colossus data center in 122 days.
xAI is exploring using the in-vehicle computers from Tesla cars as a deployment platform for its "human emulator" models.
Sulaiman Ghori states that running xAI's models on Tesla in-vehicle computers is more capital-efficient than using VMs on AWS or Oracle, or buying hardware directly from NVIDIA.
xAI is considering a business model where it would pay Tesla owners to lease compute time on their idle vehicles to run "human emulator" agents, potentially covering the owner's lease payments.
xAI's MacroHard project aims to create a "human emulator" that can perform any digital task a human does via keyboard and mouse inputs, analogous to how Tesla's Optimus robot targets physical tasks.
xAI's data centers must seamlessly switch from the public power grid to a fleet of over 80 mobile generators during periods of high municipal demand to avoid disrupting training runs.
A single engineer at xAI is rebuilding the company's core production APIs with the assistance of 20 AI agents.
xAI made a strategic decision that its MacroHard "human emulator" must be at least 1.5 times faster than a human, and current performance is trending towards 8x or more.