NVDA: Why The King Is Not Done And The Bears Keep Being Wrong

Nvidia is the most important company of this decade. Not Microsoft. Not Apple. Not Google. Not OpenAI. Nvidia. The AI revolution as we have experienced it from 2022 onward is fundamentally a Nvidia revolution. Every meaningful frontier model has been trained on Nvidia silicon. Every meaningful inference deployment is overwhelmingly Nvidia. Every sovereign AI program from Saudi Arabia to the UAE to France to Japan is buying Nvidia. The world is, for now, dependent on a single company for the compute that powers the technology shift of our lifetimes.

That is an extraordinary position. The bear case has been wrong every single quarter since 2022. There are good reasons to think it will continue being wrong through 2026 and likely well beyond.

This is the honest 2026 bull case.

Nvidia owns the world's most defensible AI compute platform, the demand cycle is structurally longer than any prior tech cycle because AI compute is generative rather than replacement, and the moat is widening across silicon, networking, software, and sovereign relationships at the same time.

Most retail investors think Nvidia's moat is the GPU. The moat is much bigger than that. Nvidia owns CUDA, the dominant software stack for AI development. Nvidia owns NVLink, the dominant high-speed chip-to-chip interconnect. Nvidia owns Spectrum-X and Quantum, the dominant AI fabric networking platforms. Nvidia owns Mellanox heritage networking IP. Nvidia owns its DGX systems. Nvidia owns the Omniverse platform for industrial digital twins. Nvidia owns inference acceleration libraries used by every major model deployer. Nvidia has the deepest hardware-software integration in tech outside of Apple.

That moat is not one product. It is a stack. Competitors can build a faster chip. Replicating the stack is a multi-year, multi-billion-dollar effort that nobody has finished yet.

Catalyst one: Blackwell and Rubin ramp. Blackwell GB200 and B200 are now in full deployment with hyperscalers and AI clouds. Rubin is ramping. Each new generation drives faster performance per dollar, which expands the addressable market by making more AI workloads economically viable. The product cycle is accelerating, not decelerating.

Catalyst two: sovereign AI. Sovereign AI deals are the most underappreciated revenue stream in the entire Nvidia story. Saudi Arabia, the UAE, France, Japan, Korea, India, the UK, and many more countries are committing to building national AI infrastructure with Nvidia silicon. These are multi-year contracts measured in tens of billions of dollars. They expand Nvidia's customer base beyond the handful of US hyperscalers.

Catalyst three: inference scaling. Training was the first wave of AI compute spend. Inference is the second wave. Inference compute scales with usage rather than with model development. As AI is embedded in every consumer product, every enterprise workflow, and every device, inference demand compounds at a much higher base than training. Nvidia is the dominant inference platform too.

Catalyst four: enterprise adoption. Hyperscaler revenue gets the headlines. Enterprise AI is the next phase. Fortune 500 companies are deploying internal AI infrastructure. Nvidia's enterprise software stack, including NVIDIA AI Enterprise and Omniverse, is positioned to capture that wave.

Catalyst five: networking. Nvidia's networking business is one of the fastest growing inside the company. As AI clusters get bigger, networking becomes more valuable. Spectrum-X is the dominant Ethernet-based AI fabric. InfiniBand share is huge. The networking business alone is a top ten networking company in the world.

Bear claim one: customer concentration is too high. Yes, hyperscalers are a big slice of revenue. They are also still under-built relative to their AI ambitions. They are also being joined by sovereigns and enterprises. Concentration has been falling, not rising.

Bear claim two: ASICs are coming for the data center. Google TPU, Amazon Trainium, Microsoft Maia, Meta MTIA. Custom silicon is real and growing. It also still represents a minority of training compute. Nvidia's flexibility and CUDA ecosystem makes it the default for almost every new workload. Custom silicon eats specific workloads, not the platform.

Bear claim three: digestion is coming. The data center capex cycle will eventually slow. Yes. But the level of compute the world needs to run mature AI products is much larger than the level needed to build them. A digestion phase compresses the multiple but not the long-term thesis.

Bear claim four: valuation is too high. Nvidia's forward P/E has actually compressed as earnings have outpaced the stock. The multiple is reasonable for the growth. The question is whether the growth continues. The growth almost always continues.

Bear claim five: China export controls limit growth. Real. Already mostly priced. Nvidia has lost a meaningful amount of revenue to export controls and grown anyway. That tells you everything about the rest of the demand picture.

Nvidia trades around a low to mid 30s forward P/E most quarters, which sounds expensive until you compare to the growth rate, the margin profile, and the cash flow conversion. Eighty percent gross margins. Sixty plus percent operating margins. Nearly 100% free cash flow conversion of earnings. Net cash on the balance sheet. The financial profile is among the best in tech, if not the best.

The bull case is not multiple expansion. It is durable earnings growth. As long as the earnings number keeps climbing, the stock keeps working.

Nvidia is a core position. Not a trading vehicle. Sized big enough to be a real driver of portfolio returns. I add on weakness. I sit through volatility. I do not try to time the cycle.

If I am wrong about Nvidia, I am wrong about AI. If I am wrong about AI, I have bigger problems than this position.

The first risk is a true digestion cycle that compresses both growth and multiple at the same time. That would be ugly.

The second risk is a competitor breakthrough on a meaningful workload. Not a chance to take lightly.

The third risk is geopolitical escalation between the US and China that further constrains exports.

The fourth risk is regulatory antitrust pressure as Nvidia's share continues to grow.

The fifth risk is concentration in the broader market itself. Nvidia is now such a large part of US indices that any broad market drawdown punishes the stock disproportionately.

Pull Nvidia's most recent earnings transcript and read it cover to cover. Pay close attention to the data center commentary, the networking commentary, the sovereign deals, and the gross margin guidance. Then come back and decide whether this is a company you should be long or whether the bear case has changed. The transcript reads itself.

Read next: ARM: The Royalty Machine | MRVL: The Quietest AI Monster of 2026

*⚠️ Important Disclaimer: MentorSurge is not a financial advisor. This post is for educational and entertainment purposes only. Nothing on this site constitutes financial, investment, or trading advice. Semiconductor stocks are volatile and subject to industry cycles. Always do your own research and consult a licensed professional.*