Investors shift focus downstream from Nvidia

Nvidia still sits at the center of the AI buildout, but the investor conversation is widening. The chipmaker’s latest reported quarter showed $62.3 billion in data-center revenue out of $68.1 billion in total sales, underscoring how much of the company’s business still comes from supplying the hardware behind large AI systems. At the same time, companies that rent out Nvidia-based compute, manage inference workloads or package AI into production services are drawing more attention from investors and executives. (investor.nvidia.com) (investors.coreweave.com) Jensen Huang has helped frame that next phase himself. At Nvidia’s GTC conference in March, Huang said he saw $1 trillion in orders for Blackwell and Vera Rubin systems through 2027, according to CNBC’s report of the keynote. In a January statement announcing a deeper partnership with CoreWeave, Huang said AI was entering “its next frontier,” while CoreWeave Chief Executive Michael Intrator said Blackwell offered Nvidia’s “lowest cost architecture for inference.” (nvidianews.nvidia.com) (investor.nvidia.com) ### Why are investors looking beyond Nvidia if Nvidia still dominates? Nvidia’s own numbers explain part of the answer. The company reported full-year fiscal 2026 revenue of $215.9 billion and fourth-quarter data-center revenue of $62.3 billion, making the data-center segment about 91% of quarterly sales. (cnbc.com) That concentration leaves Nvidia as the clearest way to invest in AI infrastructure, but it also pushes investors to ask which companies capture spending after the chips are installed. March 16, 2026, added another marker. Huang said Nvidia saw $1 trillion in orders for Blackwell and Vera Rubin through 2027, a figure CNBC reported as a sharp increase from the company’s earlier $500 billion opportunity estimate. (nvidianews.nvidia.com) A market of that size implies room for more than one winner, especially among cloud operators, model-serving platforms and software vendors that turn compute into recurring services. (investor.nvidia.com) ### What does “downstream” mean in this AI trade? CoreWeave offers one example. The company describes itself as an AI cloud provider and said on May 7 that first-quarter revenue rose to $2.1 billion, up 112% from a year earlier, while contracted revenue backlog reached $99.4 billion. Those figures point to investor interest in businesses that buy Nvidia systems, deploy them at scale and resell access to model builders and enterprises. (cnbc.com) Baseten offers another. The company said in January that it raised $300 million at a $5 billion valuation, with IVP, CapitalG and Nvidia as anchor investors, and described itself as an “AI inference company.” Fireworks AI said in October 2025 that it raised $250 million at a $4 billion valuation to expand what it called an AI inference cloud for enterprise customers including Uber and Shopify. (investors.coreweave.com) ### Why is inference getting so much attention now? Inference is the stage where trained models are actually used to answer prompts, generate content or run software agents. Nvidia’s January announcement with CoreWeave said the two companies were expanding their relationship as AI systems moved into “large-scale production,” and Intrator said Blackwell was designed to lower the cost of inference. (businesswire.com) Morgan Stanley’s summary of a March conference appearance by Huang said he described AI as shifting toward agentic systems and argued that efficiency measured in “tokens per watt” had become a CEO-level decision. (businesswire.com) That language matters for investors because it shifts the discussion from buying chips to operating services efficiently over time. ### Does this mean Nvidia is losing ground? Nvidia’s partnerships suggest the opposite. The company said on January 26 that it had invested $2 billion in CoreWeave at $87.20 a share and was expanding collaboration across CPUs, storage systems, software and future chip generations including Rubin. (nvidianews.nvidia.com) Rather than stepping away from infrastructure, Nvidia is tying itself more closely to the companies building AI clouds on top of its hardware. (morganstanley.com) CoreWeave’s disclosures also show how dependent many downstream players remain on Nvidia’s ecosystem. The company said it had surpassed 1 gigawatt of active power and posted nearly $100 billion in backlog, figures that reflect demand for cloud capacity built around accelerated computing. The downstream trade, in other words, still runs through Nvidia’s supply chain. (nvidianews.nvidia.com) ### What are investors watching next? May 20, 2026, is Nvidia’s next major checkpoint. The company’s investor relations page lists fiscal first-quarter 2027 results as the next earnings release, giving investors a fresh read on Blackwell demand and data-center growth. May 13, 2026, is one of CoreWeave’s next public appearances. Its investor relations page says the company will participate in the J.P. Morgan Global Technology, Media and Communications Conference after reporting first-quarter results on May 7. (investors.coreweave.com) Those events are likely to provide the next public signals on whether investor attention keeps moving from the chip layer to the cloud and inference companies selling AI as a service. (investors.coreweave.com) (investor.nvidia.com)