Microsoft’s Maia 200 is real: 10 PF FP4, 216GB HBM3e, 7 TB/s.
— Ben Pouladian (@benitoz) January 26, 2026
But NVIDIA’s Vera Rubin (H2 2026): 50 PF FP4, 288GB HBM4, 13 TB/s.
Hyperscalers build for today’s inference costs.
NVIDIA builds for tomorrow’s ceiling.
$NVDA $msft pic.twitter.com/gIwd3xlwT4