Nvidia didn’t just catch the AI wave; it effectively became the tide.

Over the past couple of years, the popular chipmaker has evolved from a tech stalwart to a market superpower, surpassing $4 trillion in market capitalization.

The stock has now turned into the S&P 500’s engine, riding a once-in-a-generation build-out of AI data centers.

Under the hood, Nvidia owns the high end of AI compute. Estimates show that it commands an 80% to 90% share of AI accelerators. Everyone from OpenAI to the biggest hyperscalers is scrambling to buy its chips.

Revenue has followed, with AI orders pushing quarterly sales to records, while software and networking have only deepened the moat.

For perspective, Nvidia’s sales in the pre-ChatGPT era (fiscal 2022) were at $26.91 billion and have now surged to $130.5 billion in fiscal 2025 (a 385% jump).

Now comes the next pivot: Vera Rubin, its brand-new CPU+GPU platform, tailor-made for powering the future of AI as we know it.

With the platform, Nvidia isn’t just refreshing chips; it’s extending the cadence while allowing hyperscalers to plan years in advance.

Vera Rubin could mark the start of “AI Wave 2.0,” extending Nvidia’s lead in the data-center race.

Image source: Morris/Bloomberg via Getty Images

What is Nvidia’s “Vera Rubin” platform?

Nvidia’s next major platform isn’t just a fancy new nickname, but it’s perhaps the next AI growth cycle. “Vera Rubin” efficiently layers a new Arm-based Vera CPU with the Rubin GPU, a direct successor to the current Blackwell chips, designed to handle the enormous AI inference and training needs.

The flagship rack, the Vera Rubin NVL144 CPX, is expected to deliver nearly8 exaflops of AI performance along with 100 terabytes of fast memory, which is about 7.5 times more performance than the current GB300 NVL72 system. 

Another major edge for Nvidia is that the new platform is tuned for handling next-generation workloads, including things like million-token coding, generative video, and autonomous agents.

Related: Former Intel CEO drops curt 2-word verdict on AI

On the hardware side, Rubin substantially improves compute density (up to 50-100 PFLOPs FP4 per GPU), memory bandwidth (with HBM4 and GDDR7 variants), and networking through NVLink 144 and Spectrum X. 

Also, Nvidia says the platform has already been taped out and is currently “in fab” at TSMC, with general availability expected in late 2026.

Could Vera Rubin spark another AI wave? 

Many AI experts agree that the industry’s next major leap won’t come from smarter algorithms, but from faster, denser infrastructure, and Rubin is right in that lucrative wheelhouse.

With its effective new rack-scale design and economics, analysts say Rubin could potentially unlock the next wave of AI adoption across hyperscalers and enterprises.

Here’s why Vera Rubin matters

  • Bigger context windows: With a whopping 100 TB of rack memory and 1.7 PB/s bandwidth, Rubin is able to handle colossal datasets and prompts on-rack, cutting latency and chatter. 
  • Better ROI math: Nvidia claims customers can effectively monetize nearly $5 billion in token revenue per $100 million invested in Rubin infrastructure. If that claim is remotely accurate, that substantially slashes inference costs while making always-on copilots commercially viable.
  • Predictable rollout: Nvidia’s steady cadence de-risks hyperscaler planning where buyers can easily deploy Blackwell now and slot Rubin in 2026 without retooling their stacks.
  • Higher pricing power: With HBM4, NVLink 144, Spectrum X networking, along with full rack-scale integration, Vera Rubin is likely to carry significantly higher ASPs than Blackwell, expanding the company’s gross margins in the process.

Nvidia’s GPU evolution at a glance

  • Celsius/ Kelvin/Rankine/Curie (1999-2004): A massive step-up from primarily fixed-function graphics to truly programmable shaders, opening the door for general compute. 
  • Tesla (2006): The CUDA era begins, laying the foundation for a unified programming model that allowed the running of non-graphics math on GPUs.
  • Fermi (2010): Beefed up double-precision and better memory hierarchy.
  • Kepler (2012): Greater efficiency and more parallelism per watt, facilitating massive power savings for data centers and in scaling core counts.
  • Maxwell (2014): Major perf/watt leap and smarter caching, enabling stronger real-world throughput.
  • Pascal (2016): P100 with HBM2 and NVLink showed up, which led to superior memory bandwidth and quicker GPU-to-GPU links, facilitating the first truly large AI training clusters.
  • Volta (2017): First Tensor Cores purpose-built for deep learning, delivering incredible speedups for training neural networks.
  • Turing (2018): Tensor + RT Cores for real-time ray tracing on desktops. 
  • Ampere (2020): Wider rollout of Tensor Cores, along with more robust efficiency across data center and client. 
  • Hopper (2022): Transformer Engine optimized for LLMs pushing generative AI into overdrive.
  • Blackwell (2025): Colossal training/inference jump backed by varied memory options and faster interconnects. 
  • Vera Rubin (2026, planned): Vera CPU + Rubin GPU as an integrated, rack-scale AI platform custom-built to handle ultra-long context windows, heavier memory footprints, and agentic/video workloads.

AMD’s answer to Nvidia’s Vera Rubin: how it stacks up

AMD’s catching up to Nvidia’s Vera Rubin quickly. 

Its powerful Instinct line has moved swiftly from today’s MI300X to MI350, then MI450 in 2026, with Oracle planning to deploy 50,000 MI450s starting Q3 2026, a major nod of approval.

AMD also announced Helios, which is a powerful rack-level system, potentially fitting in nearly 50% more memory per rack than Nvidia’s comparable Vera Rubin setup.

Related: Jamie Dimon drops surprising take on AI stocks

Nvidia’s Rubin is betting on a split design, focusing more on a compute-heavy configuration that facilitates raw speed, as well as another tuned for more demanding long-context tasks. AMD’s pitch, though, is simpler, as it aims to pack more memory per GPU and open-source software (ROCm) to lure developers away from CUDA. 

Ultimately, however, the winner comes down to throughput per dollar and software ease, where Nvidia still has an edge.

Wall Street bets on Nvidia’s next leap, the Rubin chip cycle

Wall Street is singing the praises of Nvidia’s GPU platform rhythm.

Goldman Sachs recently reiterated a solid buy rating, modeling 2026 EPS to be 10% above consensus, on the back of the Rubin ramp and resilient hyperscaler spending. Morgan Stanley also raised its price target on the stock to $200 on a 33× 2026 p/e, pointing to easing supply bottlenecks. 

Citi went even higher at $210, flagging Rubin CPX as a new growth driver while lifting its AI infrastructure forecast to $490 billion by 2026. At the same time, HSBC’s Frank Lee has a new Street-high rating at $320, arguing that mega-deals like the one with OpenAI could double Nvidia’s data-center revenue by FY27.

It’s essential to note that Nvidia’s Vera Rubin combo arrives at a critical juncture.

At GTC 2025, CEO Jensen Huang pitched the next phase as agentic AI, and Rubin is tailor-made for handling richer world models and video.

Related: Veteran trader who bought Nvidia at $13 resets stock price target