NVIDIA hopes physical AI and edge AI will spread this agentic demand wave farther out, as AI agents help to automate and streamline operations across every industry. NVIDIA wants to spread its entire ecosystem outward, anchoring itself as the end-to-end stack powering the coming waves of demand from autonomous vehicles, robotics, smart devices, industrial automation, and smarter communication networks.

I've been entirely focused on NVIDIA's success in AI data centers in my coverage thus far. Let's finally walk through these further out waves, as its ecosystem spreads across physical and edge AI, NVLink Fusion partnerships, and the roadmap beyond Vera Rubin. It will also be of interest to the hyperscalers and neoclouds that hope to serve this physical AI demand wave.

• Physical AI is starting to fully unleash autonomous vehicles, robotics, drones, and other devices. NVIDIA is seeding each of these areas with ecosystems, software stacks, and open models.
• Physical AI, edge AI, and industrial AI will eventually build towards another inflection point for agentic AI, as compute demand increases further from the ongoing automation of enterprise & industrial workflows.
• NVIDIA is spreading GPU capacity out across multiple levels for these needs – from on-device systems, workstations, edge servers, on-premise clusters, regional or sovereign neoclouds, and hyperscalers. Users will have flexibility to choose between open models running locally, on-demand cloud inference, or dedicated AI capacity.
• They recently announced a new line of superchips designed to reinvent the modern Windows laptop. Let's hope it goes better than their first foray with DGX Spark.
• NVLink Fusion lets partners combine their custom CPU and AI chips into NVIDIA's scale-up fabric, expanding the overall AI ecosystem while preserving NVIDIA's fabric as the anchor point. These partners gain an immediate scale-up network, and it seems likely to also position them for pairing with NVL72 GPUs in disaggregated inference.
• Vera Rubin Ultra and Feynman will see a major expansion of their scale-up networking and introduce optics, resulting in larger supercomputer clusters.

Part 4:

• Expanding compute outward
• Physical AI
• Edge compute systems
• NVLink Fusion
• Future Roadmap

Now that we've covered Vera Rubin and its modular racks, let's focus on NVIDIA's strategic moves to spur downstream demand, secure upstream supply, and expand its ecosystem. This includes their push to accelerate enterprise agentic AI adoption, their ecosystem investments across the supply chain and neoclouds, and the latest wave of major AI buildouts.

• Agentic AI is now driving a major inflection in inference, and it is still just getting started.
• NVIDIA is providing open-source LLM models and new agentic software layers to help spur enterprise adoption of agentic AI.
• Agents will also thrive on data access. NVIDIA is leveraging existing CUDA-X libraries to help accelerate agentic data workloads, spurring even more GPU demand while freeing up the CPU for other agentic tasks.
• Deeper agentic AI adoption will magnify compute needs from here across long-running orchestration, spawned off sub-agents, calling or running separate AI models specific to task, and accelerated data processing.
• NVIDIA’s ecosystem investments and partnerships help secure upstream capacity in the supply chain, while seeding downstream capacity for rising neoclouds, frontier AI labs, and sovereign AI.
• The recent wave of AI buildouts shows demand coming from all directions at once, across hyperscalers, neoclouds, frontier AI labs, sovereign AI, and private capital funds.
• The largest of individual AI buildouts are projected to grow from an average 400-600MW (housing 350-600K GPUs) this year to >2GW (housing 3.5-4.5M GPUs) in 2028-2029. Mgmt expects buildout costs to rise from $50-60B per 1GW to $80-100B, given the coming increase in density.
• Frontier AI labs like Anthropic and OpenAI are seeking compute wherever they can find it. Anthropic is finally branching into GPUs, and has committed to capacity from Azure, CoreWeave, and SpaceX.
• OpenAI is snuggling up with AWS and will adopt Trainium for some coming agentic services – but seem to remain primarily focused on GPUs.
• Meanwhile, Google is building its own competing TPU ecosystem, and just announced it is heavily ramping up its planned capex spend. And Anthropic is helping fuel that new TPU ecosystem with a brand new deal this week (with hints that OpenAI might be involved).

Part 3:

• Tidbits from GTC Taipei
• Agentic AI demand
• Ecosystem investments
• AI buildouts continue
• OpenAI shift
• Google's TPU push into neoclouds

We aren't quite done with NVIDIA Week. A final part will cover their farther out waves to expand their ecosystem across physical AI, edge compute, and NVLink Fusion, as well as their roadmap after Vera Rubin.

Now that we covered the major Vera Rubin announcements, let's dive into the finer details on those new modular rack systems across GTC in March and GTC Taipei this week.

• Grace Blackwell was their push into inference at scale. Vera Rubin is their push into Agentic AI at scale, by disaggregating its needs across different chips (GPU, CPU, DPU, networking) and rack systems (Vera, SPX, Spectrum-6, Groq LPX).
• NVIDIA is drastically ramping up its capacity, by doubling its supply chain for Vera Rubin and massively speeding up assembly time across it. They redesigned & simplified their MGX modular architectures, taking tray assembly from over 2 hours to now 5 minutes and greatly improving the ongoing maintainability.
• They are shifting into disaggregated inference with a new Groq-powered LPX rack that works in tandem with a Vera Rubin NVL72 rack to drastically improve latency and per-user interactivity. This leverages the strengths of LPU chips while circumventing their shortcomings.
• These new Groq racks allow GPUaaS and AI providers to greatly expand the type of premium AI workloads they can offer, enabling higher pricing tiers.
• They have silently dropped the Rubin CPX rack systems they announced in September in lieu of this new LPX rack. Rising memory prices and TSMC bottlenecks likely shifted their focus towards Groq. (Those plans may be only temporarily shelved – we may see it return once DRAM prices settle down.)
• NVIDIA is getting more serious about its Arm-based CPU. Their redesigned next-gen Vera CPU will now be sold as a standalone CPU and rack designed for agentic orchestration.
• They are leveraging the new Vera & BlueField-4 chips in new AI storage (STX) and shared AI memory (CMX) systems, which will greatly improve data & cache access speeds from Vera Rubin clusters.
• Spectrum-X Ethernet networking is now at 800GbE, with co-packaged optical (CPO) ports. This will be leveraged to increase the scope of scale-out and scale-across networking, with cluster sizes expected to exceed 500K in Vera Rubin. This week at GTC Taipei, they announced that it is now in production.
• They have a new DSX suite of reference designs and tooling for the design and operation of AI data centers. They are stressing how DSX can help AI factory build outs maximize the number of GPUs for a given power load, to help smooth ongoing power demand.

Part 2:

• Vera Rubin rack & POD
• Groq LPX rack
• Silently dropping Rubin CPX plans
• Vera CPU rack
• Networking
• BlueField-4/Storage
• DGX software stack

Welcome to NVIDIA Week! It's time to catch up on the strategic moves from our favorite AI infrastructure provider. This will soon be followed by a Neocloud Week, to catch up on CoreWeave, Nebius, and IREN.

Now that we've looked at NVIDIA's stellar Q127, let's peek at what comes later this year with Vera Rubin. NVIDIA debuted this new era of AI systems at CES in January, greatly expanded it at GTC in March, and refined their agentic message at GTC Taipei this week. This will also be of interest to AI chip competitors, hyperscaler clouds, neoclouds, and AI providers like Anthropic and OpenAI.

This will be a multi-part series. This first post focuses on the strategic and financial impacts of NVIDIA's expansion of Vera Rubin into a modular AI factory platform. A second post will go deeper into the individual rack systems, and a third post on the agentic software stack and other major moves.  

• The rise of agentic AI is creating another scaling law that will not only drive GPU demand, but also CPU demand for agentic orchestration, tool use, service calls, data access, and running code generated by the AI agent.
• NVIDIA initially announced the Vera Rubin line at CES in January, built around 6 new frontier chips designed in tandem across GPU, CPU, DPU, and 3 layers of networking.
• They "acquired" Groq just 2 weeks before CES. While not part of the initial announcements, it soon became a major part of the roadmap by GTC as a 7th chip in the Vera Rubin mix.
• Groq is not replacing GPUs, but rather, enhancing them in the stages of inference where its SRAM-based serialized architecture can lower latency to drastically improve per-user interactivity. They are positioning this as allowing inference providers to add premium-priced tiers for the lowest latency.
• At GTC in March, Vera Rubin was then expanded into a complete line of 5 new rack systems powered by those 7 chips. Customers can now mix in these new modular racks to improve performance in different areas across disaggregated compute, networking, storage, and agentic orchestration, depending on their needs and use cases.
• Vera Rubin sales are expected to start showing up in October (at the very end of their Q327), so will really start contributing heavily in Q427 and Q128.  
• These new rack types are being framed as incremental to that $1T in expected GPU sales through 2027. The CEO's rough expectations (Jensen Math_™) are for 25% uplift from Groq, 20% uplift from AI storage, and 5% uplift from Vera CPUs.
• As for Vera going standalone, mgmt believes it will add $200B in TAM, and expects Vera systems to be $20B in FY27 (~5% of the mix).
• Going forward, Rubin Ultra and Feynman will continue to be offered as Oberon-based NVL72 racks. This assures that past AI data centers can continue to upgrade existing facilities into these systems without a complete power & cooling overhaul, as long as they can deliver the per-rack power needed.

Now in Part 1:

• A new scaling law (agentic AI) in inference
• Groq "acquisition"
• CES Announcements / initial look at Vera Rubin line
• GTC Announcements / expansion of Vera Rubin line
• Modular inference
• Future iterations

Part 2:

• Groq LPX rack
• Silently dropping Rubin CPX plans
• Vera CPU
• Networking
• DPU/BlueField-4/Storage
• Software stacks
• Roadmap changes

Part 3:

• Agentic AI
• AI buildouts
• Ecosystem investments
• Partnerships & NVLink Fusion
• Other announcements of interest

Axon reported Q126 results this week. The stock jumped over 10% the next day, but that is now mostly getting erased today. Axon remains down -29% YTD.

As always, their shareholder letter held a wealth of details, including additional product moves out of Axon Week that weren't visible to non-attendees.

• Axon continues to show durable >30% growth. However, this has been coupled with continued margin impacts from tariffs & strategic improvements in inventory & supply chain.
• Mgmt is suggesting that deals are landing faster due to the urgency to adopt new technologies (RTCC, drone, counter-drone) and AI.
• Axon's execution remains lumpy due to hardware sales. Weakness in Q1's sequential growth isn't surprising after the impressive strength here in Q4. Given the improvements in sales cycles, I think (in retrospect) that some pending deals may have been pulled forward last quarter.
• International saw a massive acceleration to +116% growth, rising to 20% of the mix.
• A mere 18 months after acquisition, Dedrone is riding significant momentum, accelerating to +300% revenue and +500% bookings growth. This drove the Platform Solutions segment to accelerate to +95%.
• Their enterprise segment rose +50%, with rising adoption from Fusus, Dedrone, and the ABW Mini front-line wearable still in beta.
• Both Dedrone and AI features are seeing strong appeal across all market segments.
• ARR grew +34%, and Future Contracted Bookings rose +44%. FY26 revenue guidance was raised to +32%.
• Adj FCF TTM margins sank further to 1%, putting them at a Rule of 35 TTM. However, mgmt is guiding FY26 Adj FCF to over 12%, suggesting a massive rebound over the next 3 quarters. This would bring them back above a Rule of 40.
• In the ongoing Q2, they just landed major new deals with the City of Baltimore and a major enterprise telco. It was exciting to see these new deals include Axon Vision to automate the monitoring of camera feeds in Fusus.
• They also let us know about other new product moves out of Axon Week, including Axon Guardian and Dedrone enhancements.

NVIDIA continues to impress. The stock is up +78% TTM, but was down YTD on macro worries. Thankfully, it recovered by +20% in April, bringing it back to +5% YTD.  It remains my largest position.

• They continued to reaccelerate in Q4, and guided Q1 to further reacceleration – all without sales to China. And after hitting a record in net new revenue last quarter, they topped it again.
• Gross margin rose back to 75% after its drastic dip during the ramp-up of Blackwell NVL72 racks. FY27 was guided to stay in the mid-70%s.
• FCF margin continued to rebound, and the CFO expects 50% of it to be returned to shareholders going forward.
• Total supply chain commitments rose a massive +89% sequentially, but is of little worry – the upcoming Rubin is again magnifying their supply chain complexity.
• Mgmt is now projecting a vision into $1T in Blackwell & Rubin sales through 2027, increasing their projection from 6 months ago (at GTC DC and Q3 earnings) of $500B through 2026. This previously implied $350B of sales in 2026, and now implies $500B of sales in 2027. And that estimate does not include any of the new rack varieties.
• Agentic AI is now driving a surge in inference demand from the new frontier of enterprise coding tools and personalized AI assistants.

The next year (and likely 2) of sales is visible, so now it's down to the execution of their supply chain and the real-world performance of their next-generation Rubin stack – plus we likely have upside from several new complementary rack forms that enhance inference performance in several directions. In part 2, I'll cover the new chips and rack-level systems announced at CES and GTC, and where they are going with Groq LPUs, BlueField-4, and Vera CPUs.