Jensen Huang Is Selling the Nvidia Story: Who Wins From the $3–4 Trillion AI Buildout?

AMD, Intel, Qualcomm Extend Gains: Who Leads the Next Chip Wave? 451K Views · 46 Posts In One Chart joined discussion · 17:31 Jensen Huang Is Selling the Nvidia Story: Who Wins From the $3–4 Trillion AI Buildout? Jensen Huang believes the AI infrastructure supercycle is far from over. Speaking at Goldman Sachs' Communacopia + Technology Conference on September 10, the $NVIDIA (NVDA.US)$ CEO reiterated that global AI infrastructure spending could reach $3 trillion to $4 trillion by 2030 . When the moderator referred to his earlier forecast, Huang joked that everyone should pause and acknowledge that he had been right. Later, he made his agenda even clearer: "I'm here to sell some NVIDIA stock." He described Nvidia as a rare company combining both growth and value characteristics. The comments sounded promotional, but the real takeaway was Huang's explanation of Nvidia's next phase of growth. Over the past two years, investors mainly traded the shortage of GPUs. Going forward, Nvidia wants to capture far more than the GPU market alone. 1.Nvidia Is No Longer Just Selling GPUs Huang repeatedly emphasized that the market still views Nvidia as a GPU company, even though the company now sells a complete AI factory. From GPUs and CPUs to NVLink, networking, rack-scale systems and the CUDA software stack, Nvidia is steadily expanding its share of every dollar spent on AI infrastructure. A complete GPU system can now be worth roughly $8.5 million, while monthly shipments of Grace Blackwell NVLink72 racks recently increased by about 27%. The underlying shift is that competition is moving beyond the performance of individual chips. As Moore's Law slows, AI performance increasingly depends on system-level co-design. Chips, packaging, memory, networking and software must work together to produce more tokens at a lower cost. This is also the foundation of Huang's $3 trillion to $4 trillion infrastructure forecast. Computing is moving from retrieving information to generating it in real time—and from chatbots that respond occasionally to AI agents that operate continuously. Supporting that transition will require an entirely new computing infrastructure. Nvidia is therefore not simply trying to sell more GPUs. It wants to capture a larger share of the value inside the entire AI factory. 2.Who Benefits From the $3–4 Trillion AI Buildout? If Nvidia's forecast proves correct, the opportunity will extend well beyond Nvidia itself. Larger AI factories will require more wafers, advanced packaging, memory, networking equipment, power and data-center capacity. The conference highlighted three major investment themes. Theme One: The Most Direct "Picks and Shovels"—Wafers, Packaging and Memory Huang was unusually direct about the supply-chain bottlenecks facing the industry. Wafers, advanced packaging, DRAM, LPDDR, connectors and voltage regulators are all under pressure. More importantly, Nvidia still sees supply—not demand—as the main constraint on revenue growth. Huang reaffirmed confidence in roughly 70% year-over-year growth next fiscal year , while suggesting unconstrained demand could grow by more than 100%. If AI capital spending continues to expand, companies controlling scarce production capacity should remain among the clearest beneficiaries. $Taiwan Semiconductor Manufacturing (TSM.US)$ remains the most critical link. Whether the industry uses GPUs, custom ASICs or increasingly complex AI processors, advanced nodes and advanced packaging still depend heavily on TSMC. Nvidia's reliance on technologies such as CoWoS and large-scale system integration is also increasing as it looks for ways to overcome the limits of Moore's Law. Memory suppliers are another important group. These include $SK hynix (SKHY.US)$ , $Samsung Electronics (005930.KR)$ and $Micron Technology (MU.US)$ . Investors have largely focused on HBM, but Huang's references to DRAM and LPDDR suggest that the bottleneck is spreading into a broader range of high-performance memory. As AI systems move toward rack-scale architectures and agentic inference expands, the amount and variety of memory required by each system should continue to rise. The first layer of the investment case is therefore straightforward: more computing requires more wafers, packaging and memory. These remain the most direct picks-and-shovels plays in the AI infrastructure buildout. Theme Two: The Next Scarce Resources—Power, Data Centers and Neoclouds The more important message from the conference, however, was that the industry's bottleneck is shifting. The question is no longer only whether enough GPUs are available. It is increasingly whether the industry has enough power, land and data-center capacity to deploy them. Huang said Nvidia is tracking available land, electricity and data-center shell capacity by the gigawatt across the world. That reflects a broader transition: AI infrastructure is entering the gigawatt era . At the conference, Huang highlighted an Australian AI infrastructure project targeting about 2 gigawatts of capacity by 2027. He estimated that a project of that scale could represent roughly $80 billion of investment. This helps explain why Nvidia has been supporting Neocloud providers such as $CoreWeave (CRWV.US)$ , $NEBIUS (NBIS.US)$ , $IREN Ltd (IREN.US)$ , Nscale, Lambda and Firmus. These companies are no longer offering simple GPU rentals. They combine GPUs, land, power, data centers and financing into complete computing infrastructure. Huang described Neoclouds as important future distribution channels for Nvidia's architecture because they can secure power and land beyond the capacity already controlled by traditional cloud providers. That points to an important shift in the AI trade. Investors previously focused on GPU shipment volumes; going forward, they may pay increasing attention to gigawatts of available capacity. Companies with reliable, low-cost power, suitable land and deployable data-center capacity could become the next major beneficiaries. Data centers, electrical equipment, grid infrastructure and Neoclouds may therefore emerge as the next AI infrastructure trade after GPUs and HBM. Theme Three: From AI Factories to Applications—Cybersecurity and Physical AI As infrastructure expands, investors must eventually ask a more important question: what will all this computing power be used for? Huang's first answer remains coding. But he believes cybersecurity could become the next major AI application after coding. The logic is simple. AI can write code continuously, but it can also search for vulnerabilities and automate attacks. In response, red-team agents could probe systems around the clock, while blue-team agents continuously detect and repair threats. This always-on, high-frequency workload is well suited to AI and could create sustained demand for computing. Huang specifically highlighted $CrowdStrike (CRWD.US)$ . Nvidia and CrowdStrike are using Nemotron models to build an automated system in which offensive and defensive agents continuously compete and improve. Companies such as CrowdStrike, $Palantir (PLTR.US)$ and $Cisco (CSCO.US)$ represent the next stage of the AI value chain: the expansion from capital spending to measurable returns. Nvidia, TSMC and memory manufacturers sell what is needed to build AI. Cybersecurity and enterprise software companies help answer why businesses will continue paying for it. Beyond cybersecurity lies another major opportunity: Physical AI . Huang believes autonomous driving will be the first mature killer application for Physical AI, with meaningful progress likely over the next two to three years. The technology could then spread to autonomous mobile robots, warehouse logistics and intelligent manipulation systems. The long-term destination of AI infrastructure is therefore not limited to data centers. Once models can understand and interact with the physical world, vehicles, robots, logistics networks and even 6G infrastructure could become new sources of computing demand. Conclusion Huang's message was clear: the AI trade is moving beyond the question of whether enough GPUs are available. The next question is whether the entire AI infrastructure ecosystem can continue expanding. As Nvidia evolves from a chip supplier into an AI factory platform, the potential beneficiaries are also broadening—from GPUs to wafers, packaging, memory, power and data centers, and eventually to cybersecurity, autonomous driving and Physical AI. For investors, the next stage is not about finding every company with an AI label. It is about identifying the businesses that control the scarcest and hardest-to-replace parts of the AI buildout. If the $3 trillion to $4 trillion infrastructure opportunity continues to materialize, the next sources of outperformance may emerge from these new bottlenecks. Disclaimer: Moomoo Technologies Inc. is providing this content for information and educational use only. Read more 30 2 32 50 111K Views