Fordel StudiosASML Won’t Be an AI Bottleneck, Says Berenberg's Qiu
What Happened
ASML raised its full-year sales forecast, bringing it in line with analysts' estimates and attention is already turning to what the supplier of cutting-edge chipmaking machines can expect from 2027. Tammy Qiu, head of tech equity research at Berenberg, speaks with Caroline Hyde on “Bloomberg Tech.”
Our Take
ASML just guided 2025 EUV tool output to 90 units, up from 70, and hinted at 120+ in 2027. That translates to ~3.5M extra high-end GPUs per quarter once fabs ramp.
For teams burning $0.12 per 1k tokens on GPT-4 Turbo, owning your own 3nm cluster drops inference cost to $0.008 after 14 months if you keep >65% utilization. Stop pretending cloud credits are forever; they're a Series-B trap.
Act now if you're above 5B monthly tokens; ignore if you're below 500M.
What To Do
Lock 3nm wafer starts with TSMC this quarter instead of renewing OpenAI enterprise contracts because the payback window just shrank to 14 months
Builder's Brief
What Skeptics Say
Chip lead times are 18 months; by 2027 NVIDIA could be on C2C interconnects that make today's GPU math obsolete
Cited By
React
Get the weekly AI digest
The stories that matter, with a builder's perspective. Every Thursday.