NVIDIA: Durable Infrastructure in AI Leadership, but Nigh-Perfect Precision is Required

I am still fervently steadfast in the stance that NVDA is the undisputed anchor and future driver of the global AI shift. While it is an 'obvious' megacap investment, my personal analysis depict its transition into a full stack 'AI foundry' with a widened moat that the market presently fails to price in. The company has migrated from GPU vending to the crucial core of accelerated computing, software and network tooling in the training and inference workloads utilised by hyperscalers, model developers, enterprises, and sovereign buyers who are all surging in deployment. This has evolved beyond a strategic position towards exceptional financial performance, reaching $215.9 billion in fiscal 2026 revenue, expanding 65% yoy, with GAAP gross margin of 71.1% and Q1 FY2027 revenue pushed toward $78 billion ±2%. NVIDIA also mentions expectations of sequential growth through and beyond calendar 2026, supporting the case of demand being strong beyond cyclical drivers.

It's straightforward for the bull case - the demand for AI computing in hardware and software and the means to propagate them is rising faster than supply, with NVIDIA setting themselves as the default standard for developers and cloud systems, capturing beyond chip demand to the much broader market stack through networking, rack scale systems and software. CEO Jensen Huang also explicitly framed contemporary environments as an 'agentic AI inflection point', with Grace Blackwell positioned as a leader in inference economics and Vera Rubin intending to extend the lead further.

The remaining and harder question is no longer whether NVIDIA is a sustainable or even great business. It is apparent. The real question is whether the stock price reflects the fundamental strength, and leaves room still for investors once future growth, margin durability and terminal valuation are properly put to the test.

My View

My opinion is that the real 'alpha' of NVDA is not just the sheer speed of growth, but rather something rarer - the quality of its expansion coupled with speed and the concentration of it. Fiscal 2026 Data Centre revenue rose past ~$193.7 billion, dwarfing sectors of Gaming, Professional Visualisation, and automotive and confirming that the company has become a centrepiece to the AI capex cycle rather than just a story of a cyclical consumer graphics provider. Q4 FY2026 revenue was $68.1 billion alone, with management guiding Q1 FY2027 to $78 billion, implying that there is still plenty of growth to roll over yet as of early 2026.

It is important to recognise this as current investment cycles are not being driven by a narrow customer segment or any single demographic, but from a broad portion of hyperscalers, frontier model labs, enterprise AI developments and a growing number of sovereign AI initiatives. NVDA's moat is further strengthened by its software ecosystem CUDA, where switching costs are driven to the point where it's beyond a mere change of chips but also tooling, workflow, and market timing decision. The depth in ecosystem is a significant support of NVDA's premium pricing and high margins relative to most semiconductor businesses.

Catalysts

The first major catalyst arises from the ongoing ramping of Blackwell systems and transition to next generation AI Infrastructure. Management described demand for Grace Blackwell and NVLink systems to be strong and tied to the next phase of growth increasingly to inference and not just model training. If AI operating expense broadens from hyperscaler capex towards enterprise and sovereign AI buildouts, NVDA can benefit from a much longer and sustainable demand pathway than normal cyclical semiconductor implications.

The second catalyst is the scale of NVDA's near term demand visibility. Revenue in Q4 FY2026 of $68.1 billion followed $57.0 billion in Q3 and $46.7 billion in Q2, exemplifying that even beyond massive growth, businesses are still bolstering growth quarter by quarter, the same revenue staircase is what feeds market confidence to underwrite estimates spanning multiple years than just a one year spike.

The third catalyst is expansion on multiple fronts in the stack. NVDA does not only sell accelerators but increasingly selling integrated systems, networking, and software adjacent capabilities. If the market continuously rewards full stack AI infrastructure providers than just commodity component vendors, NVDA's economics may carve its existing moat wider and structurally remain stronger than many investor expectations.

Assumptions

My base case DCF is not built on the assumption of 'forever growth' but instead modelled with a gross margin compression down to 72% by FY31 to account for the inevitable rise of custom silicon from hyperscalers like Google and Amazon. But even with this compression, cash generation is the proprietary powerhouse. My revenue forecasts indicate to it growing from $215.9 billion in FY2026 to ~$340.9 billion in FY2027, then to roughly $689.2 billion by FY2031, before fading towards a more mature but solidified and exceptional profile through to FY2036. The path is anchored by a quarterly FY2027 bridge initiating from Q1 guidance then to a slower, more normalised long term trajectory.

I took the assumption that gross margin moderates from 74% in FY2027 to high 60s over time, reflecting both NVDA's continued pricing power and a more competitive future as custom silicon, TPUs, ASICs and alternative accelerators expand. I also assumed EBIT margin remains unusually strong, settling at levels still implying NVDA keeps meaningful ecosystem power than just being a normal cyclical semi vendor. In terms of cash generation, I estimated unlevered free cash flow increasing from about $147.7 billion in FY2027 to >$330 billion by FY2031, further expanding during the fade period. The discount rate in my model was about 10.56%, reflecting a business that is still high in quality but exposed to volatility and risks technologically, competitively, and geopolitically.

Risks

The biggest risk is the durability of hyperscaler spend. If my forecast for Data Centre growth - fading by 12% by FY31 proves too optimistic from a capacity 'digest' period, the downside is potentially significant. A good majority of NVDA's current valuation rests on the idea that hyperscalers, model developers and enterprises will keep expanding compute budgets at very high levels for years. Should training and inference economics progress faster than expectations or if major customer digest capacities for longer than expected, revenue growth may face a downturn.

Secondly, another risk is customer vertical integration. Despite the market lead, customers are not static. Google keeps up the pressure by building TPUs, hyperscalers are exploring custom silicon and Broadcom-style ASIC models are becoming more credible in selected workloads. NVDA's danger is not losing overnight but that long term pricing power and wallet share gradually compress.

Thirdly, the greatest geopolitical risk is China and its export controls. NVDA's own FY2027 guidance assumed no Data Center compute revenue from China, which painted the picture that first, management is acknowledging geopolitical constraint, and second, that China is no longer safely embedded in near term outlook. This may reduce downside to estimates if restrictions remain, but also caps a meaningful growth opportunity in a high demand market and increases dependence on the rest of the world to absorb incremental supply.

Fourthly, arguably the greatest risk is its own valuation. Despite strong fundamentals, even great businesses may be a mediocre investment if too much of the future is already priced in. NVDA is at the stage where the stock may be more sensitive to disappointments in Data Centre growth, gross margin trajectory or next generation product ramp than macro questions of whether AI remains credible.

Fair Value

Ultimately, my DCF suggests a fair value of approximately $171 USD per share, and under an exit multiple cross check, ~US$230 per share was received. With the stock priced at around $173.75 USD in the model, NVDA is evaluated to be roughly fair to slightly overvalued in a stricter cash flow basis, but undervalued if the market continues to assign a premium multiple to its long duration AI leadership. The real insight is in the spread. NVDA is not a hard stock to love, but a hard stock to value. It's obvious the company is of pristine quality, but the debate is whether its long term economics deserve to remain so exceptional that a premium terminal framework continues to hold.

I assign my fair value around the middle of that range, but closer to the DCF than to the multiple driven cross check. To put it simply, I think NVDA is a strong and even great business, but less cheap than the enthusiasm of AI potential denotes.

Bottom Line

NVDA remains to be the strongest platform company in the AI infrastructure industry, with splendid revenue scale, high margins and an authentic ecosystem advantage. Its results are undeniable, Fiscal 2026 and FY2027 guidance support the idea that demand is still strong into 2026 and is an ongoing deployment cycle than just a one-off success.

However, this strength is also a trap, as despite the company's performance, the stock relies on investors continuing to believe in AI capex durability extending to multiple years, sustained dominance in ecosystems and longer term economics remaining far above normal semiconductor standards.