Elon Musk's xAI has been criticized for overbuilding its compute infrastructure, leaving significant capacity unused. According to Harrison Rolfes, a Senior Research Analyst at PitchBook, Musk's strategy of constructing at maximum scale has resulted in excess resources that competitors have capitalized on. In 2024, xAI's $10 billion server lease with Oracle collapsed due to slow cluster construction, leading Musk to build the Memphis data center independently. The unused GPU capacity was later acquired by OpenAI. Further issues arose when Musk redirected 12,000 NVIDIA H100 chips from Tesla to xAI, delaying Tesla's projects. Despite these efforts, xAI's Colossus 1 data center was underutilized, with only 11% of its compute capacity being used. This underutilization led to leasing out over 220,000 NVIDIA GPUs to Anthropic, as xAI's Grok platform could not support the massive scale initially intended.