ANJNEY MIDHA, an early Anthropic investor and former Andreessen Horowitz general partner, has raised $1.3 billion for Amp, a fund that pairs AI compute access with venture capital for startups and researchers.
The model upends the traditional VC playbook. Rather than offering capital alone, Amp provides GPU allocation directly to portfolio companies, removing one of the steepest cost barriers for AI startups training models at scale. Midha led investment rounds at Anthropic, Mistral, Black Forest Labs, and Periodic Labs before launching Amp.
Midha has been direct about his critique of the venture establishment. Traditional firms, he argues, failed to grasp that infrastructure access—not just deployed capital—would become the binding constraint for frontier AI development. That miss is Amp's opening.
The unit economics are instructive. A startup that must rent H100s at market rates faces opaque, volatile pricing and no guaranteed allocation during peak demand. A startup that has reserved compute as part of its funding gets predictability and, critically, the ability to iterate faster than competitors on the same budget.
Midha describes the current moment as the "revenge of the scientists," a shift where technical depth and direct hardware access now outweigh brand-name advisors or downstream network effects. Stripe's acquisition of OpenRouter—which Midha flagged as a significant strategic signal—suggests the same thesis: control the infrastructure layer, control the agentic internet.
Amp's competitive edge rests on capital efficiency for its LPs. By buying compute in bulk and allocating it across a portfolio, the fund arbitrages the gap between retail GPU pricing and industrial-scale procurement. Portfolio companies get cost-effective access; Amp captures the spread and the equity upside. Midha continues to hold Anthropic and has indicated he expects frontier AI companies to command higher multiples in public markets as the technology matures and revenue scales.
