Nvidia reported $96.2B in quarterly revenue, up 106% year over year, and for the first time gave investors a year-ahead forecast: roughly 70% revenue growth in the next fiscal year, stacked on top of what it just delivered. The stock added $442B in a single session pushing Nvidia's valuation to $5.5T, with CEO Jensen Huang telling analysts demand is "much greater than 70%" and the binding constraint is supply, not appetite. (CNBC)
Private Companies
Voltaother
JPMorgan is leading a $5B debt package to fund the AI data-center buildout of Volta Infra Holdings, an eight-month-old company founded by two former Brookfield infrastructure executives. Volta's first site — a 121MW campus in Norway leased from bitcoin miner Bitdeer — anchors the $10B, six-year compute deal it signed with Anthropic this month, and the new debt sits alongside a separate $5B non-dilutive infrastructure program Volta struck with Spanish asset manager Azora in early August. A startup this young raising bank debt on the strength of one hyperscaler-grade contract is what compute financing looks like once equity checks stop being the binding constraint. (Bloomberg)
Public Markets
Marvell posted record fiscal second-quarter revenue of $2.74B, up 37% year over year, with data-center revenue accelerating to 46% growth and Q3 guidance landing roughly $116M above consensus — as clean a beat-and-raise quarter as the stock has had all year. Shares fell as much as 7.6% in after-hours trading anyway: after a 184% year-to-date run, investors wanted a bigger number, and a beat that would have been celebrated in January now reads as merely in-line. (SiliconANGLE)
Emerging
Real-time robotics: A new paper, Deltoris, accepted at MICRO 2026, tackles the latency wall in diffusion-based vision-language-action models — the models that let robots plan and act in one pass — by exploiting how little consecutive control-loop frames actually change. Its temporal-aware bit-level sparsity and speculative-inference techniques cut redundant computation and amortize memory traffic, delivering up to 34.2x speedup and 822x energy savings over mobile GPUs, the kind of headroom needed to run sophisticated embodied AI at the 50-200Hz control frequencies smooth robot motion requires. (arXiv)
And before we wrap..
The Compute 100 will be taking the week off! We’ll be back on September 8th for a fall covering the stories that define the compute market. See you then!