Wafer's core innovation is using specialized AI agents to perform the complex, menial task of optimizing LLMs to run efficiently on GPUs. This automates the work of the ~500 highly-paid, scarce GPU kernel engineers in the world, dramatically accelerating performance optimization.
Nvidia's investment in MediaTek is a convertible note, a financially savvy structure providing downside protection. Since MediaTek builds Google's TPU for inference, Nvidia can now financially benefit from its main competitor's market traction, turning a competitive threat into a financial opportunity.
VC Finn Barnes argues that large funding rounds are driven by the internal mechanics of mega-funds. Partners with deployment quotas find it easier and more efficient to "mark up and follow quickly" on a hot deal than to find a competitor, contributing to valuation bubbles and market consolidation.
When heavily funded AI application companies fail to find product-market fit, they are increasingly pivoting to become "Neoclouds." Having already committed to massive capital and chip purchases, they default to the proven, less innovative business model of reselling compute to justify their valuation and spend.
The AI hardware ecosystem requires intricate partnerships, forcing rivals to invest in the same companies. Just as Google and Amazon both backed Anthropic, Nvidia is now a major financial partner to MediaTek, a key supplier for its competitor Google. This strategy is now essential for hedging bets and capturing value.
The current trend of usage-based pricing for AI is a response to the marginal cost of inference. VC Finn Barnes predicts that as smaller, efficient models become capable of running locally, the marginal cost to deliver AI features will approach zero, enabling a return to the predictable, high-margin SaaS subscription model.
AI startup Wafer has demonstrated that with proper software optimization, AMD chips can achieve 80-100% of Nvidia's performance (in tokens per second) for specific open-source models, at roughly half the cost. This challenges the notion of Nvidia's insurmountable hardware dominance by proving software is a key performance unlock.