The overhead is much lower for software to begin with. We can consider the ratio of hardware manufacturers to even the hobbyist developer who makes a few bucks

The overhead is much lower for software to begin with. We can consider the ratio of hardware manufacturers to even the hobbyist developer who makes a few bucks on YouTube. Either way, multinationals are running out of ink signing their seats on spaceship AI and they pay in stacks. The critical test is achieving escape velocity for each individual GenAI "solution" (call it what you want...). Software giants with compute access are windmilling free models every day, hardware is racing to capture that compute, and those GenAI companies are entrenching their contracts while the competition is still hot. Once they have the contracts it's just build build build until another AI winter or possibly a nuclear winter.