AMD will significantly expand its AI chip production capacity, with Chief Executive Lisa Su reporting that customer demand for the company's artificial-intelligence accelerators remains strong.

Speaking via a report carried by financial platform Moomoo, Su confirmed AMD's intent to scale manufacturing output for its AI processors, framing the move as a response to continuing customer demand. Her remarks position AMD among the semiconductor suppliers publicly committing additional capacity to the data-center AI buildout.

What did Su actually say?

"AI chip demand remains robust," Su stated, signaling that order books have not weakened despite recent volatility in broader semiconductor markets. The phrasing — ongoing, not peaking — implies AMD expects multi-quarter visibility into orders rather than a tapering cycle.

The Moomoo report did not specify whether Su's comments originated from an earnings call, investor event, or interview. AMD has not published a transcript or press release through the channels covered in the original aggregation.

How does AMD fit the AI accelerator supply picture?

AMD competes in the AI accelerator market against Nvidia's H100, H200 and Blackwell families, supplying alternatives built around its Instinct MI300 series and planned follow-on products. Hyperscale operators including Microsoft, Meta and Oracle have publicly disclosed MI300-series deployments.

Su's emphasis on ongoing demand reflects a broader market dynamic. While Nvidia commands the largest share of AI-GPU shipments, customers have publicly diversified procurement to second-source suppliers — a pattern that has supported AMD's commercial momentum through 2024 and into 2025.

What production levers can AMD pull?

Expanding capacity typically routes through three constrained inputs. First, contracted wafer allocations from foundry partner TSMC. Second, advanced-packaging capacity for chip-on-wafer-on-substrate (CoWoS) interconnects — the bottleneck that has constrained AI-GPU supply industry-wide. Third, HBM memory allocation, where AMD's accelerators compete for stacks alongside Nvidia and a wave of custom AI silicon programs at hyperscalers.

Each input is currently allocated across competing customers. Securing additional allocation generally requires multi-quarter purchase commitments and prepayment structures — a financial pattern that has become standard across the AI supply chain.

What was not disclosed?

The Moomoo report did not include specific dollar amounts, capacity volumes, capex figures, manufacturing-location details, or timeline targets. AMD's prior public guidance pointed to AI accelerator revenue rising to a multi-billion-dollar run rate, and additional output capacity will be required to meet that trajectory.

What to watch next

Traders and analysts will look to AMD's next quarterly earnings call for concrete capacity disclosures, capex implications, and customer-shipment breakdowns. Updates on next-generation accelerator products and on packaging capacity at TSMC will materially influence sell-side forecasts.

Until AMD provides firmer figures, Su's characterization of demand as robust is the only quantitative anchor the report offers.