AI development is consuming resources at an unprecedented scale, repurposing electrical grids, GPUs, and talent in a race reminiscent of the philosophical paperclip maximizer. According to Anjney Midha, founder and CEO of AMP and former general partner at Andreessen Horowitz, this resource intensity has cemented the dominance of major tech companies. Their financial capacity to justify massive spending for Artificial General Intelligence (AGI) creates a high barrier to entry.
Midha argues that access to compute infrastructure must not remain a bottleneck for innovative new labs. While big players chase AGI, he identifies multiple frontiers in AI development beyond a single goal. The technical recipe for frontier models remains consistent: pretraining, mid-training, post-training, and a continuous feedback loop. In this landscape, Anthropic stands out as a role model for efficient development, contrasting with the heavier operational footprints of larger competitors like Google.
For enterprises, the key to value lies in deep integration. Midha notes that embedding AI into core functions such as HR, IT, and procurement can reduce costs by millions. However, effectiveness varies by domain. AI excels in fields with verifiable feedback, such as software engineering and material science, where objective validation prevents hallucinations. Conversely, subjective tasks like creative writing lack these structured loops, limiting current model reliability. The future of competitive AI will depend on equitable compute access and the ability to implement rigorous, verifiable feedback mechanisms.