Spend versus return starts with knowing what's being sold. AMD's fiscal 2021 annual report (filed February 3, 2022) describes the Instinct MI200 series, built on "AMD CDNA" 2 architecture and "optimized for HPC and AI/ML (Artificial Intelligence/Machine Learning) workloads," naming the MI250 within the family. EdgarBeast pulled the language.
Read against last year's filing, this is a deliberate step up: FY2020 disclosed AMD's first data-center accelerator, the MI100; FY2021 discloses a second-generation architecture and explicitly attaches it to AI and machine learning, not just high-performance computing. The accelerator line is becoming a stated AI play in the company's own words.
The 10-K does not, as of this filing, isolate Instinct revenue, so the honest read is product-momentum disclosure, not a financial result. I won't conjure a number the company hasn't reported.
Forward from here, the comparison that matters is whether AMD's accelerator narrative converts into disclosed data-center economics. As of FY2021, AMD is on record with a purpose-built AI/ML accelerator generation. The 10-K on sec.gov is primary; EdgarBeast is the evidence index.