It Won’t Be AGI Until H100s Fall to $30 an Hour
| Source: HN | Original article
The path to artificial general intelligence hinges on Nvidia H100 GPUs dropping to $30 per hour.
A recent commentary has sparked fresh debate about the economics of artificial‑general‑intelligence research, arguing that the field won’t move beyond narrow breakthroughs until Nvidia’s flagship H100 GPU can be rented for roughly $30 an hour. The piece, circulated among AI developers and cloud‑service providers, points to the current hourly rates for H100 instances – which remain well above the $30 mark – as a key barrier to the massive compute budgets required for training truly generalist models.
Why it matters is twofold. First, the cost of high‑end GPUs directly determines how many researchers and startups can afford to experiment at the scale of today’s largest language models. When compute is prohibitively expensive, progress concentrates in a handful of well‑funded labs, slowing broader innovation. Second, the price signal influences the economics of AI‑focused data centres. As we reported earlier on the $30 billion price tag of a 1 GW AI data centre in Texas, hardware costs dominate the budget; a reduction in H100 pricing would ripple through the entire supply chain, potentially lowering entry costs for new AI infrastructure projects.
What to watch next are the levers that could bring the $30‑per‑hour target within reach. Nvidia’s upcoming GPU generations, competitive offerings from other chipmakers, and pricing strategies of major cloud platforms will all shape the market. Additionally, any shifts in semiconductor supply chains or government incentives for AI compute could accelerate a price drop. Stakeholders will be monitoring announcements from Nvidia and cloud providers over the coming months for signs that the “$30‑an‑hour” benchmark is becoming realistic – a development that could unlock a new wave of AGI‑level research.
Sources
Back to AIPULSEN