DigitalOcean Holdings (DOCN)

The sleeper story is that ARR per megawatt, the metric management leans on hardest to justify the AI pivot, is actually down from the pre-AI era's $22M+/MW anchor to $18M blended today, and when pressed for the GPU/CPU split that would prove the 'software makes megawatts more valuable' thesis, management pivoted to 'the important number is the total' rather than answer. The obvious headline, 40% revenue growth and a raise to 60%+ for 2027, is real, but the binding constraint hasn't changed: it's still Tier-1 data center capacity (only 180MW committed through 2028) against demand management itself says 'far outstrips' supply. What should worry the desk more than the numbers is what went quiet: RPO ($894M, 12x YoY last quarter) and the open-weight token share stat (15% to 75%) both disappeared with zero acknowledgment, right as top-25 concentration jumped 700bps on a conveniently switched denominator. Paddy Srinivasan's own words cut both ways: 'it's becoming increasingly hard to distinguish between who is AI, who is not' is a great pitch to investors but also an admission that AI-attributed growth numbers are becoming less falsifiable by the quarter. My view changes if RPO reappears next quarter with continued hyper-growth, or if management discloses a GPU/CPU-split ARR/MW figure that proves the blended yield decline is purely a mix-shift illusion rather than real dilution.