Nscale

In AI infrastructure, capacity is the easy part to quantify. What it produces is harder. Teams stall before production when they can't get a first workload running, properly factor in what it will cost, or scale it without needing to start over. That abandoned work is capacity that never gets consumed. Daniel Bathurst, Chief Product Officer, on where growth in this market actually gets decided: https://lnkd.in/eftZWytE

Time to first token (TTFT) has become a useful signal of full-stack AI performance. At RAISE Summit, Nscale's Kristin Zwez joined NVIDIA's Rod Evans and VAST Data's Jeff Denworth to discuss why TTFT now sits at the center of AI infrastructure design. The conversation explored how inference is reshaping the relationship between compute, storage, networking, and memory, and why improving TTFT depends on optimizing the entire AI stack rather than individual components. Read the key takeaways: https://lnkd.in/gAbKgcTH

Ray Summit 2026 saw Nscale connect with developers and engineers across the entire Anyscale community. Our speakers Eyal Lantzman and Oscar Savolainen, PhD explored some of the critical decision making that goes into serving frontier models, as well the role vertical integration plays in determining the cost of AI workloads. Check out some of our highlights below, and stay tuned for more insights from our week in San Francisco.

Nscale is deepening its commitment to West Virginia with a foundational $500,000 gift to Mason County Public Schools, funding premier STEM initiatives and skilled-trades programs. The gift comes alongside a new report by EY on our work at Monarch Compute Campus, which projects annual tax revenues from Phase One of the development to reach $105 million, including $87 million for Mason County Public Schools. We’re committed to delivering a positive academic and economic impact across the district and wider region. Read the full announcement: https://lnkd.in/eVqpWw-Y

Nscale Founder & CEO Josh Payne is on the 2026 TIME100 AI list. TIME’s list recognizes the top 100 global innovators, shapers, and thinkers who have a stake in the future of artificial intelligence.  Read more: https://lnkd.in/eBRG44zy #TIME100AI

AI infrastructure has a new scorecard. As inference scales, performance is about how effectively the entire stack turns compute into results. This month in Full Stack AI, we look at why time to first token brings that shift into focus. Read the latest edition.

AI-native infrastructure separates what should stay stable from what needs to continuously evolve. New models, GPUs, and frameworks arrive every few months. Your infrastructure shouldn't need to be rebuilt every time. Read more about designing an AI stack that can evolve while the underlying infrastructure remains consistent: https://lnkd.in/dxJq8fCt

What makes an AI system feel fast? A big part of the answer is time to first token: how quickly the first output token appears. TTFT hinges on more than any single component. Every layer of the stack plays a role. We break it down here: https://lnkd.in/eXTVZ55e

At Nscale, we built AI natively from the beginning. Now one of our greatest opportunities is giving every customer the means to also build in a sovereign way. Watch more from Lauren R Hurwitz, Chief Digital and Information Officer at Nscale.

The headline numbers in AI infrastructure are supply numbers: megawatts secured, GPUs deployed, capacity sold out. They only describe half the machine. The other half is whether a customer can turn that capacity into something that works. Every initiative abandoned between experiment and production is capacity that never gets consumed. The product layer is what closes that gap: It turns infrastructure into workloads, and workloads into growth. Read more from Daniel Bathurst, Chief Product Officer at Nscale: https://lnkd.in/eftZWytE