Armenia and Kazakhstan with @FirebirdCloudAI are building the infrastructure to create intelligence at home, powered by the NVIDIA DSX platform. 📽️ Hear from @JensenHuang on this important milestone for the CIS region.

Today in Armenia, @FirebirdCloudAI announced the launch of the CIS region's largest AI factory, designed to bring global demand for AI infrastructure, local development opportunities and new economic growth to the region. Powered by NVIDIA accelerated computing with @Dell and @SchneiderElec, the facility will provide developers, enterprises, universities and public institutions with the infrastructure to build and deploy AI at scale. Read more:

Weekly NVIDIA Update SPACEX GOES ALL-IN ON NVIDIA $SPCX will build its AI services exclusively on $NVDA systems, Elon Musk said on SpaceX's first earnings call. "We think the Vera Rubin architecture is the best AI computer, and we're exclusive to Nvidia." SpaceX expects to exit 2026 above 2 gigawatts of compute and approach 10 gigawatts by the end of 2027, deploying Vera Rubin NVL72 racks on the ground and in orbit through its Starmind satellite program launching next year. CIS REGION'S LARGEST AI FACTORY OPENS IN ARMENIA Firebird opened its Hrazdan AI factory built on $NVDA accelerated computing and $DELL PowerEdge servers, going from plan to operating capacity in just over six months. The company plans more than 70,000 Rubin and Blackwell GPUs and 300MW of capacity in Armenia by the end of 2027, part of a roughly 2-gigawatt roadmap spanning Armenia, Kazakhstan and additional markets. NVIDIA OPENS ITS ROBOTAXI MODEL TO COMMERCIAL USE $NVDA released Alpamayo 2 Super, a 34-billion-parameter open reasoning model for autonomous driving, for commercial use. Weights are on Hugging Face under the Linux Foundation's OpenMDW-1.1 license, covering fine-tuning, derivative models and commercial redistribution. Developers can train on their own fleet data, distill into in-vehicle compute and ship, all under one license. The Alpamayo family has passed 500,000 downloads and ranks first on the LingoQA driving-reasoning benchmark.

$CRWV Goldman Fireside Chat Takeaways (And what this means for $NBIS) 🌲 Shift to Greenfield Datacenters At the end of 2025, $CRWV still explicitly disclosed that all of its existing datacenters were leased. Land, power, and the physical shell run roughly 20% of all in AI infrastructure capex. The remaining 80% goes into GPUs. Having 3P fund the buildout means the same dollar of upfront CAPEX supports roughly 25% more deployed GPUs. The problem is that capital efficiency comes at the expense of long-term margins. The developer needs to earn a return on the infrastructure it funded, leaving lease payments embedded in $CRWV’s cost structure for years after the facility comes online. Now with power and land becoming scarce, developers have more pricing power, making long duration leases more margin dilutive. Intrator does a lot of hand waving, but the reason for pivot is that as you scale into bulk of active capacity through 2030, doesn't make sense to have 90+% in colo margin. Even so, $CRWV’s disclosed greenfield pipeline still appears to be under 1GW, and a material chunk of that is held through a JV rather than wholly owned. This is of course a validation of $NBIS's model of anchoring long-term capacity in owned facilities. Similar to how $IREN long eschewed the software layer but eventually acquired Mirantis. 🗺️ Geographic Diversification When pressed on current political climate around datacenters, Intrator states they are very "forward thinking" with over 1GW contracted abroad. $NBIS of course is doing precisely this. Having secured 3GW+, with the majority in owned GW scale greenfield sites in the US, the remaining ~4.5GW of active power by 2030 (forecast by BofA) is expected to be in Europe, Asia, and EMEA. ☁️ Asset Lite Model Intrator first introduced $CRWV Omni on the Q1 call: deploy and operate the full $CRWV cloud stack in customers' own data centers, on their GPUs. First Omni agreement signed in Q2, expected to scale in 2027. This is an identical arrangement to $NBIS's Asset Lite Model. Instead of squirreling away debt in SPVs, you now put leverage entirely on other people's balance sheets. Effectively, you are decoupling your growth trajectory from CAPEX constraints. This trend should accelerate as the AI infrastructure market broadens beyond the handful of larger operators that paved the road and built the first generation of AI clouds and neoclouds. This is only possible because of the strength of the demand curve for compute. There are a shocking number of newer, non-technical entrants with capital and power but little expertise in building and serving enterprise-grade infrastructure. They need someone else’s blueprint, operating layer, and credibility. That is something companies like $NBIS and $CRWV can provide. Some of these firms include Firmus, Firebird, Yotta, Neysa etc. 💽 Managed Inference When asked where the higher-margin software opportunity sits, Intrator's answer was memory, storage, networking, and observability. Not inference. That omission is intentional. $CRWV does have a managed inference product, but it wasn't built in-house. $CRWV Serverless Inference came via the Weights & Biases acquisition. For reference, their catalog serves 18 open models against Token Factory's 60+. $CRWV's managed inference first surfaced on the Q2 earnings call at a $100M ARR run rate. Small in the grand scheme of things, and clearly not a point Intrator pushes vigorously like $NBIS for Token Factory. 💴 GPU Financing Evolves $CRWV lenders are beginning to finance beyond the life of the underlying customer contract. Historically, infrastructure debt was largely matched against contracted revenue. 5YR debt for 5YR contract. The latest facility has an 5YR maturity against customer contracts averaging 3YRs. That means lenders are now taking on renewal risk and willing to underwrite $CRWV's ability to renew or remarket those GPUs after the initial contract expires. The unlock is that $CRWV can use long duration debt to fund shorter term, potentially higher-priced contracts, instead of needing a 5YR hyperscaler or frontier lab commitment every time it wants to lever a GPU deployment. As this financing structure matures, I hope we will see the death of bare metal hyperscaler deals soon.