"AI Bull Market Narrative" Creates Massive Waves Again! Huang Renxun Unveils Trillion-Dollar AI Grand Vision, NVIDIA (NVDA.US) Sets Sail Toward $6 Trillion Market Cap

Bloomberg Finance APP has learned that NVIDIA CEO Jensen Huang showcased NVIDIA’s “unprecedented AI computing revenue super blueprint” at the GTC conference in the early morning of March 17, Beijing time. He informed global investors that driven by the strong demand for Blackwell architecture GPUs and the upcoming mass production of Vera Rubin architecture AI computing systems, the future revenue scale in the AI chip field could reach at least $1 trillion by 2027, far exceeding the previous blueprint of $500 billion AI infrastructure revenue by 2026 announced at the last GTC.

According to analysts at Goldman Sachs, Wedbush, and Morgan Stanley who are optimistic about NVIDIA’s stock prospects, driven by even stronger-than-expected revenue growth, NVIDIA’s market capitalization is about to surpass $5 trillion again after October last year and is highly likely to reach much higher historical highs.

For NVIDIA’s stock price, it may soon hit new all-time highs and propel the global AI computing industry chain into a new upward trajectory. NVIDIA’s trillion-dollar super AI computing blueprint is fully supporting the “AI bull market narrative” as the main theme of the capital markets. Based on the average target prices of Wall Street analysts, this implies that NVIDIA’s market cap could break through $6 trillion within the next 12 months, with the most optimistic forecasts reaching as high as $8.8 trillion.

As model sizes, inference pipelines, and multimodal/agentic AI workloads drive exponential expansion in computing power consumption, major tech giants are increasingly focusing capital expenditures on AI infrastructure. Global investors continue to anchor the “AI bull market narrative” around NVIDIA, Google TPU clusters, and AMD’s new iterations and AI cluster deliveries, making it one of the most certain growth stories in the global stock market. This also means that investments related to power supply, liquid cooling systems, optical interconnects, and other themes closely tied to AI training and inference will remain among the hottest sectors, even amid geopolitical uncertainties in the Middle East, led by industry leaders like NVIDIA, AMD, Broadcom, TSMC, and Micron.

At the annual GTC developer conference in San Jose, California, CEO Jensen Huang announced a new data center CPU (server-grade CPU) and a set of AI inference infrastructure systems based on Groq’s exclusive AI inference architecture technology. Groq is an AI chip startup that NVIDIA licensed technology from last December for $17 billion.

These initiatives are part of Huang’s efforts to strengthen the company’s position in the so-called “inference computing” field. Inference computing refers to the massive computational process of answering queries from global B2B and B2C users. In this domain, NVIDIA’s AI GPU computing system faces fierce competition from custom AI ASIC processors developed by companies like Google (TPU-led AI ASIC technology). In recent years, NVIDIA chips have dominated large AI model training, which has always been a key market focus.

NVIDIA’s AI GPU monopoly in training requires more versatile AI compute clusters and rapid iteration capabilities of the entire system, while inference emphasizes unit token cost, latency, and energy efficiency after scaling AI technology.

“The era of AI inference has arrived,” Huang said at GTC. “And the demand for inference continues to rise,” he added.

Dressed in his signature black leather jacket, Huang delivered his speech in an ice hockey arena that can hold over 18,000 people. This four-day tech conference has become one of the largest global platforms for showcasing AI technology. “I just want to remind everyone, this is a highly anticipated tech event,” he told the audience.

The surge in AI inference has elevated NVIDIA’s “AI computing blueprint” to a trillion-dollar scale.

If we condense Huang’s GTC speech into one sentence, the core message is: NVIDIA is transforming itself from a “company selling AI GPUs” into a “chip giant selling AI factories.” The official keynote opened with token as the fundamental unit of modern AI, and Huang shifted the industry focus from “training” to “inference + agentic AI,” upgrading the AI infrastructure revenue opportunity from $500 billion to at least $1 trillion for 2025-2027. This is not just demand revision but a signal to the capital markets: future compute competition will no longer focus solely on peak FLOPS but on who can produce tokens at the lowest cost, highest data throughput, and best latency.

Centered around this expanding AI compute demand narrative, Huang’s underlying business logic is clear: data centers are no longer just “storage centers” but “AI factories.” Under fixed power budgets, the most critical metrics are “tokens per watt,” “cost per token,” and “time to first production.” This is why he repeatedly emphasizes “extreme co-design”—integrating compute, networking, storage, software, power supply, and cooling into a unified optimization. Official data shows that Vera Rubin NVL72 can achieve up to 10 times the inference throughput per watt and one-tenth the cost per token compared to Blackwell, while GPU counts for large-scale MoE model training can be reduced to a quarter of the original. This is no longer just chip iteration but a rewriting of AI infrastructure economics.

On the hardware front, the most significant change at GTC is NVIDIA’s integration of CPU, GPU, LPU, DPU, SuperNIC, switch chips, and storage architecture into a platform-level system. The Vera Rubin platform includes Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, Spectrum-6 Ethernet switch, and the latest integrated NVIDIA Groq 3 LPU. The Vera Rubin NVL72 rack consists of 72 Rubin GPUs and 36 Vera CPUs, while the Groq 3 LPX rack is dedicated to low-latency inference. Huang innovatively splits AI inference into two stages: prefill handled by Vera Rubin, decode by Groq AI chips. This means NVIDIA’s answer to the inference era is no longer “GPU does everything” but using heterogeneous computing to separate high throughput from ultra-low latency.

On the software and ecosystem side, Huang’s stance is equally aggressive. Dynamo 1.0 is defined as the inference operating system for AI factories, claiming up to 7x inference performance improvements on Blackwell. For agentic AI, NVIDIA launched Agent Toolkit, OpenShell, NemoClaw, elevating OpenClaw to a “personal AI OS” platform, with strategies for enterprise deployment including control, privacy routing, and security boundaries. Meanwhile, NVIDIA expanded its open large model family with Nemotron, Cosmos, Isaac GR00T, Alpaymayo, BioNeMo, Earth-2, and previewed the Feynman architecture roadmap: next-generation platforms will introduce Rosa CPU, LP40 LPU, BlueField-5, CX10, and Kyber, pushing copper interconnects and integrated optical packaging further into the next AI factory.

Extending further, GTC 2026 is not just about data centers. NVIDIA is also bringing “physical AI” and “spatial computing” to the main stage: IGX Thor is now generally available for industrial, medical, robotics, and edge computing; the Open Physical AI Data Factory Blueprint accelerates data generation, enhancement, and evaluation for robots, visual AI agents, and autonomous driving; Space-1 Vera Rubin Module extends the Vera Rubin architecture to orbital data centers, claiming up to 25x AI compute power over H100 for space inference. This indicates NVIDIA has expanded the “AI factory” from cloud data centers to a unified infrastructure paradigm across cloud, edge, endpoint, automotive, robotics, and even space.

The core theme of GTC 2026 is not just new product launches but NVIDIA integrating GeForce, data center infrastructure, networking, storage, inference systems, agent platforms, robotics, and space computing into a unified narrative—“upgrading from a single GPU supplier to an AI infrastructure integrator.” This is why the most noteworthy aspect of this conference is not a specific AI chip parameter but NVIDIA’s system-level products locking in future token economics, inference monetization, and infrastructure bargaining power for years to come.

NVIDIA’s dominance in AI compute infrastructure solidifies, with its stock price heading toward new all-time highs.

“Investors previously worried that the massive AI infrastructure spending by tech giants might be unsustainable,” said Jacob Bourne, an analyst at Emarketer. “But with Huang outlining a $1 trillion revenue opportunity through 2027, investors are beginning to believe that the demand for NVIDIA’s AI infrastructure remains long-lasting.” “As the AI industry moves from early experimentation to large-scale deployment, NVIDIA continues to maintain its leadership in AI compute market,” he added.

When Huang raised NVIDIA’s AI chip and infrastructure opportunity scale to at least $1 trillion by 2027 at GTC, the market no longer sees a chip company selling better GPUs but an infrastructure empire trying to define the next-generation “AI factory” production function: transitioning from training dominance to inference dominance, from single-chip competition to system-level dominance of entire racks, networks, and software stacks. From Blackwell and Vera Rubin to low-latency decoding with Groq technology, NVIDIA is rewriting valuation language around token throughput, per-watt revenue, and inference monetization.

At GTC, Huang used the $1 trillion opportunity to demonstrate ongoing demand expansion and explained that NVIDIA’s entire platform—CPU, GPU, LPU, high-performance networking, software ecosystem, and agent tools—is no longer just about individual chips but about building an entire AI factory.

The “inference inflection point” Huang mentions essentially signals to the capital markets: AI capital expenditure has not peaked; large-scale deployment has just begun. When NVIDIA integrates CPU, GPU, LPU, networking, agent software, and data center economics into one narrative, it is not just a new product cycle but a super vessel heading toward a $50 trillion valuation space. According to TIPRANKS’ compiled Wall Street analyst average target prices, analysts generally expect NVIDIA’s stock to surge to $273, implying a 51% upside over the next 12 months, with the most optimistic target reaching $360. The $273 target corresponds to roughly a $6.6 trillion valuation. As of Monday’s close, NVIDIA’s stock closed at $183.22, with a market cap of about $4.45 trillion.

Huang raised the revenue opportunity for AI chips and infrastructure to at least $1 trillion by 2027, significantly higher than the previous $500 billion target based on Blackwell and Rubin architectures. After GTC, Goldman Sachs stated that the new trillion-dollar revenue outlook provides a longer-term demand endorsement, easing investor concerns about a potential peak in AI capital spending in 2026. In other words, Goldman sees this presentation not just as a product showcase but as a re-anchoring of NVIDIA’s order ceiling and performance sustainability for the next two to three years.

Goldman emphasizes that NVIDIA is not only launching another highly powerful AI GPU but also commercializing inference in NVIDIA’s unique way, fully upgrading its AI compute infrastructure into the core equipment of the next phase of the global AI arms race. As mentioned, Huang splits inference into prefill and decode stages: prefill handled by Vera Rubin, decode by Groq LPX/LPU. This indicates NVIDIA is expanding from a “training champion” to a “comprehensive AI inference infrastructure provider.” Goldman highlights that the official data exceeds market expectations: Vera Rubin + LPX can achieve up to 35x inference throughput per megawatt, offering up to 10x revenue opportunities for trillion-parameter models.

Goldman states that NVIDIA is not just defending its training market but also presenting a stronger monetization framework and more complete heterogeneous compute solutions for the latency-sensitive inference era. Their more optimistic stance is mainly because GTC simultaneously addressed two investor concerns: whether demand has peaked and whether inference dominance might be diluted by CPUs, custom ASICs, or other chips.

Goldman notes that the $1 trillion forward guidance far exceeds expectations, confirming that hyperscalers’ demand remains strong and durable. Based on optimistic outlooks for potential catalysts in the coming months, Goldman reaffirms its “Buy” rating on NVIDIA and maintains a 12-month target price of $250, emphasizing that the capital expenditure plans of super cloud providers and new models based on Blackwell and Rubin architectures will continue to reinforce NVIDIA’s performance leadership.