Chip stocks just sank — all it took was OpenAI getting efficient and Meta showing off Iris
The semiconductor index fell 6% after OpenAI reportedly halved inference costs and Meta confirmed its in-house Iris AI chip enters production in September. AMD, Intel and Nvidia all slid.
The fear that has hovered over the semiconductor market for months now has a concrete question attached: if AI labs can squeeze far more out of the chips they already own, how many new chips do they really need to buy? That doubt swept through Wall Street on Friday and shoved the semiconductor index down around 6%.
Why did chip stocks fall?
Two blows on the same day. First, a report in The Information described OpenAI engineers celebrating optimisations that cut inference costs by more than half — and Broadcom CEO Hock Tan himself pointed to roughly 50% lower cost per token versus current-generation GPUs in early tests of the Jalapeño chip OpenAI designed with Broadcom. Then Meta confirmed that Iris, its home-designed AI chip, enters production in September, feeding the buildout meant to take the company to 14 gigawatts of compute by 2027. The market did the maths on its own: AMD closed down 6.9%, Intel lost 9% and Nvidia slipped 1.3%.
Is this the end of Nvidia’s party?
Steady on. Demand for compute keeps climbing — Meta itself talks about doubling its infrastructure, and OpenAI needs entire fabs to produce the chips it has designed. What changes is how the pie gets sliced: every giant that designs its own silicon is a customer buying fewer off-the-shelf GPUs, and every efficiency gain stretches the hardware already installed. For investors used to treating demand as infinite, Friday was a reminder that “more AI” no longer automatically means “more chips for everyone”.
None of this gets settled in one trading session — but the warning is on the record now: in the AI race, efficiency has become a weapon as sharp as scale.
By Oliver Grant
Image: Radiotrefoil / Wikimedia Commons (CC BY-SA 4.0)
