Broadcom's CEO has confirmed the company plans to ship AI chips equivalent to 10 gigawatts of power capacity by 2027. The figure, disclosed in a recent statement, marks an aggressive push into the booming market for artificial-intelligence hardware.
Why gigawatts are the new metric
Measuring chip shipments by power consumption, not unit count, is unusual. But it reflects a key reality: large-scale AI training clusters now draw electricity on par with small cities. As models grow, so does the energy needed to run them. By stating a target in gigawatts, Broadcom is betting that power, not just transistor count, will define the next generation of AI infrastructure.
What the 10 GW target means
Ten gigawatts is enough to power roughly seven to ten million high-performance AI accelerators, depending on each chip's thermal design. The shipment plan suggests Broadcom expects to supply a significant share of the custom silicon that cloud providers use for training and inference. The company already designs chips for some of the largest AI operators, though it didn't name specific customers in the announcement.
Scaling up for 2027
Reaching 10 GW by 2027 won't be cheap or easy. Broadcom will need to secure enough wafer capacity from its manufacturing partners and win new design contracts in a market dominated by Nvidia. The CEO didn't reveal the exact revenue or unit forecasts tied to the target, but the scale implies a multi-billion-dollar bet on the long-term demand for AI compute.
The company's stock rose slightly after the news, though much of the street was already pricing in strong AI-chip growth. The bigger question — whether Broadcom can deliver that much silicon while maintaining margins — depends on how quickly the ecosystem shifts from training to inference, where lower-power chips often suffice.
Broadcom has not yet set a public deadline for its next quarterly update, where analysts expect more detail on the 2027 roadmap. Until then, the 10-gigawatt figure stands as the most concrete long-term target from any chip supplier in the AI arms race.
