A few moments from our GDCC Happy Hour that we wanted to share! Thanks to everyone who stopped by and spent the evening with us. We had a great time hosting you! Here's a look at how the night went. #GDCC2026 #ArcCompute #AIInfrastructure

The Economics of GPU Cluster Ownership runs again Oct 29, 11AM ET. One hour on what a deployment takes from contract to running cluster, how utilization, opex and residual value hold up, how US and UK owners structured theirs, plus live Q&A: hubs.la/Q04C3Fq40 #ArcCompute

At #GDCC2026 today. Day 1 numbers: hyperscaler capex past $450B this year, roughly 3x 2023, now showing up as 120 kW to 1 MW racks, microfluidic cooling to the silicon and gigawatt campuses. BTW Join us for today for a happy hour with Intel, Dell and KTI. hubs.la/Q04BNJnb0

Our team is at GDCC Expo Canada in Mississauga, Oct 20-21. On Oct 20 we’re co-hosting a VIP Happy Hour with Intel, Dell and KTI at Tikka Lounge, 4:30 to 7 PM ET. A small room for real talk on GPUs, power and cooling. Request a seat: hubs.la/Q04BBD_60 #GDCC2026 #ArcCompute

Spent a few days at World Summit AI in Amsterdam. Agents, enterprise adoption, sovereign AI, compute efficiency. Different rooms, same underlying point: AI is hitting production faster than the infrastructure under it is getting built. #WSAI2026 #ArcCompute

Arc Compute CEO Jeffery Potvin and Account Executive Sam Samuel are attending #WSAI26 on October 7-8. If you're at World Summit AI and thinking about how to build, deploy or operate GPU infrastructure at scale, connect with JP or Sam in Amsterdam. #WorldSummitAI #AIInfrastructure

Our CEO Jeffery Potvin keynoted Regional Fiber Connect in Boston last week and met great people there. Highlight from the talk: compute has a number ($1T in Blackwell and Rubin demand by 2027), power has one (31 to 66 GW of US load), fiber rarely does.

Compute, power and fiber are converging as AI infrastructure scales. Join us at Regional Fiber Connect Boston for the keynote: “The Economics of AI Infrastructure” 📅 Sept. 24 | 2:15 PM #RegionalFiberConnect #AIInfrastructure

Highlights from the #AIInfraSummit in Santa Clara. Efficiency was the theme, and NVIDIA framed it as tokens per megawatt.

What owning a GPU cluster actually costs. Justin Ritchie walks through the ownership model, the 10 to 14 week procurement to deployment timeline, real operating economics and live Q&A. Sept 17, 11AM ET, 60 min on Teams. Register here: hubs.la/Q04xBpG30

Back from Rafay's AI Infrastructure Leadership Summit in Barcelona. Monetization, margins, sovereign AI, the operational load of scaling GPU fleets. Different rooms, same conclusion: securing GPUs is no longer the hard part. Running them well is. #AIInfrastructure #ArcCompute

GPUs get the headlines. Operations decide who wins. At Datacloud USA, our CEO Jeffery Potvin joined industry leaders to talk neoclouds, GPU-as-a-Service and moving enterprise AI into production. The real opportunity goes far beyond access to compute.

Join our live session on the economics of GPU cluster ownership with Justin Ritchie, President of Arc Compute. Signed agreement to operating cluster, the model behind high utilization, how tenancy and 24/7 ops shape returns. Sept 17, 11 AM ET. Register: hubs.la/Q04wWBys0

Jeffery Potvin, our CEO, will be in Barcelona Sept 8-10 for Rafay's AI Infrastructure Leadership Summit. Deploying GPU capacity is one thing. Running it as a service people can rely on is the harder part, and that's what we want to talk about. Going? Let's meet. #ArcCompute

Average GPU utilization across 23,000 production clusters was 5%. Most of that gap opens between the purchase order and the first workload, in power, storage and cabling decisions made too late. Our full guide to turnkey GPU clusters: arccompute.io/resources/arc-… #ArcCompute

DeepSeek V4-Pro holds ~862 GB of weights. That is 4 B300s, about $265K of hardware doing nothing else. Qwen3.6-27B fits on one GPU. V4 ships experts in FP4, so a model card number is not the real footprint. Memory math and the serving layer: hubs.la/Q04vcnB80 #ArcCompute

AI sovereignty runs past data residency into training, inference, algorithms, jurisdiction, and governance. Cloud-only answers few of those, and data gravity plus committed spend makes leaving costly. The test is whether you have an exit path: hubs.la/Q04tSYbz0 #ArcCompute

Vera Rubin is in full production. Rubin Ultra follows in H2 2027 on a rack expected to draw around 600 kW. Facilities move slower than silicon, so power and cooling matter more than purchase timing. What it means for mid-market buyers: hubs.la/Q04trbZm0 #NVIDIA #ArcCompute