Bert Hochfeld | TickerTarget | April 1, 2026
NVIDIA is the most important company in the AI infrastructure buildout and arguably the most important technology company in the world right now. FY26 revenue reached $215.9 billion, up 65% year-over-year, with free cash flow of $96.6 billion — a 44.7% FCF margin on a hardware-centric business model that rivals the best software companies ever built. The Q4 FY26 quarter was a blowout: $68.1 billion in revenue, 75.2% non-GAAP gross margin, and $34.9 billion in free cash flow. The Q1 FY27 guide of $78 billion implies nearly 79% year-over-year growth, with consensus estimates for FY27 revenue in the $365-370 billion range. At a forward P/E of approximately 21X on FY27 consensus estimates, the PEG ratio is 0.30X — a level I would describe as extraordinary for any company, let alone one generating nearly $100 billion in annual free cash flow. I haven't covered NVIDIA in the newsletter previously — my AI infrastructure exposure has been through Nebius, Pure Storage, and Arista — but applying my framework, the valuation is genuinely attractive on a growth-adjusted basis, and the competitive position is durable enough to warrant serious consideration. This is a company where the numbers speak for themselves, and the numbers are unlike anything I have encountered in my career.
NVIDIA's business has undergone a transformation that is, in the literal sense, unprecedented. Three years ago, this was a company that did $27 billion in annual revenue, roughly evenly split between gaming and data center. Today it is a $216 billion annual revenue company with 91.5% of its revenue coming from Data Center — the picks-and-shovels provider for the entire AI infrastructure buildout that I have written about extensively.
The core competitive asset is not any single GPU architecture. It is the CUDA software ecosystem — over a decade of developer tools, libraries, and frameworks that create switching costs rivaling enterprise software. Every AI training framework, every inference library, every optimization toolkit has been built on CUDA. This is the moat. The hardware improves generationally; the software moat compounds cumulatively.
Data Center ($62.3B Q4 FY26, 91.5% of revenue): The engine of everything. Grew 75% year-over-year in Q4, accelerating from 56% in Q2 FY26. Within Data Center:
Hyperscaler concentration: "Slightly over 50%" of Data Center revenue comes from hyperscalers — Amazon, Google, Meta, Microsoft. The critical observation is that the non-hyperscaler segment is the faster-growing component: enterprise, sovereign AI deployments (>30Bannualrunrate, > 3Xyear − over − year), andphysicalAI(>6B annual run rate). This customer diversification is a structural positive that reduces single-customer concentration risk.
Gaming ($3.7B, 5.5% of revenue): Increasingly a rounding error, though still a $16 billion annual business in FY26. Supply-constrained as fab allocation prioritizes Data Center. Seasonal Q4 decline was normal.
Professional Visualization ($1.3B, record): Up 159% year-over-year. Blackwell demand in rendering and simulation. Small but high-margin.
Automotive ($604M, <1%): Long-term optionality in self-driving platforms. Not a near-term driver.
I am a numbers guy. Perhaps too much so, some might say. But the NVIDIA numbers over the past three years require no interpretation — they tell the story themselves.
Revenue Trajectory (Recent 8 Quarters):
| Quarter | Revenue ($B) | YoY (%) | QoQ (%) | Non-GAAP GM | FCF ($B) | FCF Margin |
|---|---|---|---|---|---|---|
| Q1 FY25 (Apr-24) | 26.0 | +262% | +18% | 78.9% | 14.9 | 57.3% |
| Q2 FY25 (Jul-24) | 30.0 | +122% | +15% | 75.7% | 13.5 | 44.9% |
| Q3 FY25 (Oct-24) | 35.1 | +94% | +17% | 75.0% | 16.8 | 47.9% |
| Q4 FY25 (Jan-25) | 39.3 | +78% | +12% | 73.5% | 15.5 | 39.5% |
| Q1 FY26 (Apr-25) | 44.1 | +69% | +12% | 61.0%* | 26.1 | 59.3% |
| Q2 FY26 (Jul-25) | 46.7 | +56% | +6% | 72.7% | 13.5 | 28.8% |
| Q3 FY26 (Oct-25) | 57.0 | +63% | +22% | 73.6% | 22.1 | 38.7% |
| Q4 FY26 (Jan-26) | 68.1 | +73% | +20% | 75.2% | 34.9 | 51.2% |
*Q1 FY26 non-GAAP GM depressed by $4.5B one-time H20 charge (export restriction write-down). Excluding this, the underlying margin trajectory was smooth.
The re-acceleration narrative is critical. Year-over-year growth hit a trough in Q2 FY26 at 56% — which for any normal company would be cause for celebration, but for NVIDIA created a brief sentiment panic. Since then: 56% to 63% to 73%, and Q1 FY27 is guided to approximately 79%. Revenue is re-accelerating at $216 billion in annual scale. I have never seen anything like this.
Full-Year Summary:
| Fiscal Year | Revenue ($B) | YoY (%) | GAAP GM | Non-GAAP Op Margin | FCF ($B) | FCF Margin | Rule of 40 |
|---|---|---|---|---|---|---|---|
| FY22 | 26.9 | +61% | 64.9% | 37.1% | 8.1 | 29.9% | 91 |
| FY23 | 27.0 | +0% | 56.5% | 19.3% | 3.8 | 13.9% | 14 |
| FY24 | 60.9 | +126% | 71.2% | 49.8% | 27.0 | 44.3% | 170 |
| FY25 | 130.5 | +114% | 75.0% | 62.6% | 60.7 | 46.5% | 161 |
| FY26 | 215.9 | +65% | 71.1% | 59.5% | 96.6 | 44.7% | 110 |
The Rule of 40 score of 110 at this revenue scale is not just exceptional — it is in a category occupied by NVIDIA alone. For context, AppLovin at its peak managed a Rule of 40 of approximately 130 on a $5 billion revenue base. NVIDIA achieves 110 on $216 billion. The operating leverage is staggering.
Non-GAAP gross margins require careful parsing:
| Period | Non-GAAP GM | Context |
|---|---|---|
| FY25 Q1-Q4 | 78.9% to 73.5% | Peak-to-trough on Blackwell architecture transition |
| FY26 Q1 | 61.0% | Trough — $4.5B H20 charge (one-time) |
| FY26 Q2 | 72.7% | Blackwell ramp costs |
| FY26 Q3 | 73.6% | Improving mix and cost structure |
| FY26 Q4 | 75.2% | Recovery complete; beat guidance (75.0%) |
| FY27 Q1 guide | 75.0% | Mid-70s sustainable per management |
The CFO's statement is worth noting: "The single most important lever of our gross margins is actually delivering generational leads to our customers" through performance-per-watt improvements. Translation: so long as NVIDIA maintains its generational performance gap over competitors and custom silicon, mid-70s gross margins are sustainable. This is the key assumption to monitor.
I would note that 75% non-GAAP gross margins on a hardware business at $68 billion quarterly revenue is, frankly, extraordinary. Most semiconductor companies operate at 40-60% gross margins. NVIDIA's margin structure reflects the pricing power of a company with no genuine substitute for its training-optimized compute platform.
This is where the debate gets interesting, and I want to address it directly because it is the single most important risk factor.
The bull case (which I find persuasive but not without caveats):
The bear case (which must be addressed honestly):
My assessment: The training market is NVIDIA's fortress — 90%+ share, and custom silicon has not displaced it meaningfully for frontier model training. Inference is a more competitive market, and over time custom ASICs will capture share in high-volume, well-defined inference workloads. But three factors mitigate the risk:
The inference market itself is expanding exponentially. Even if NVIDIA's share declines from 80% to 50%, the total addressable market for inference compute is growing far faster than the share loss. This is the Jevons paradox that Jensen invokes — cheaper inference creates more inference demand.
Inference is not monolithic. Frontier model inference with rapidly changing architectures (agentic AI, multi-modal reasoning) requires flexibility that GPUs provide and fixed-function ASICs do not. Commodity inference on well-defined models is where ASICs win.
NVIDIA is building the full-stack platform — not just the GPU but the interconnect (NVLink), the networking (InfiniBand, Spectrum-X), the software (CUDA, TensorRT, NIM), and the agentic frameworks. This platform approach is what Nebius invests in and what hyperscalers increasingly buy as a complete system.
I would characterize the competitive risk as real but manageable — not existential. The CUDA moat is deep, the product cadence is punishing for competitors, and the overall market is expanding fast enough to accommodate multiple winners.
I have written extensively about my three-pronged AI strategy: (1) heavy weight in AI infrastructure; (2) cybersecurity as a structural hedge; (3) software winners adapting to the agent-first world. NVIDIA sits at the very top of prong one — it is the foundational layer upon which everything else is built.
**Hyperscaler capex is approaching 700billionin2026. * *Amazon(200B), Alphabet (185B), Meta(115-135B), and Microsoft are all ramping aggressively. Approximately 75% of this — roughly $450 billion — is AI-specific. Goldman Sachs projects hyperscaler capex from 2025 through 2027 will reach $1.15 trillion. NVIDIA, as the primary supplier of training and inference compute, captures a disproportionate share of this spend.
Sovereign AI ($30B+ annual run rate, >3X YoY) is a genuinely new revenue stream. When nations build domestic AI infrastructure — and they are doing so at accelerating pace (Canada, France, Netherlands, Singapore, UK) — NVIDIA is the default platform. This is not a government contracting business with associated margin compression; these are full-price platform sales.
Physical AI ($6B+ annual run rate) is early but real. Robotics and embodied AI require the same GPU compute for inference and training. This extends the TAM well beyond cloud data centers.
The supply dynamics are worth emphasizing: HBM4 memory is sold out through 2026. TSMC's CoWoS packaging capacity — critical for assembling these GPU systems — is sold out through 2026. Lead times stretch 36-52 weeks. NVIDIA has $95.2 billion in supply commitments. This is not a demand problem; it is a capacity-constrained environment where the vendor is under-shipping relative to demand. As TSMC expands to 120,000-130,000 wafers per month by late 2026 (up from ~75,000 exiting 2025), supply constraints may ease, allowing revenue to grow even faster.
NVIDIA's guidance pattern is textbook sandbagging — what I would call "deliberately undemanding estimates":
| Reporting Quarter | Guide | Actual | Beat |
|---|---|---|---|
| Q3 FY25 | $37.5B | $39.3B | +4.9% |
| Q4 FY25 | $43.0B | $44.1B | +2.5% |
| Q1 FY26 | $45.0B | $46.7B | +3.9% |
| Q2 FY26 | $54.0B | $57.0B | +5.6% |
| Q3 FY26 | $65.0B | $68.1B | +4.8% |
| Q4 FY26 | $78.0B | Q1 FY27 TBD | — |
Five consecutive beats at 2.5-5.6%. The guide midpoint is the floor, not the target. If we apply the average beat magnitude of 4.3% to the Q1 FY27 guide of $78 billion, the implied actual is approximately $81.4 billion.
This is management credibility of the highest order. Jensen Huang and Colette Kress are not in the business of setting expectations they cannot exceed. The conservative approach to China revenue — zero Data Center compute revenue from China assumed in Q1 FY27 guidance — is another form of de-risking: any H200 revenue is pure upside to the guide.
The balance sheet is a fortress:
Shareholder returns: 41.1billioninFY26(40.1B buybacks + $974M dividends), representing 42.5% of FCF. Remaining buyback authorization of $58.5 billion. Share count declining from ~24.9B to ~24.4B (-2.0%) despite $6.4B+ annual SBC. This is precisely the capital allocation discipline I look for: returning cash to shareholders while investing aggressively in R&D and capacity.
SBC analysis: $1.6 billion in Q4 FY26, representing just 2.4% of revenue and 4.7% of FCF. This is negligible at current scale. The policy change — including SBC in non-GAAP starting Q1 FY27 — is worth noting for comparability, but the absolute levels are immaterial.
I would be remiss not to address the specific risk factors:
Hyperscaler capex cyclicality: If the hyperscalers collectively decide to slow or pause AI spending — as they did with cloud infrastructure in 2022-2023 — NVIDIA's revenue would decelerate materially. The counter-argument is that $1 trillion in backlog and multi-year contracts provide visibility. But hyperscaler CFOs can stretch timelines. The $95.2B in supply commitments cuts both ways: it secures capacity but also commits capital.
Custom silicon displacement in inference: As discussed above. Real in commodity inference, not real in training or frontier inference. Monitor NVIDIA's inference market share disclosures.
China export restrictions: Currently de-risked in guidance (zero assumed). But the regulatory environment is volatile. Any tightening could impact broader supply chain; any loosening would be upside.
Product transition execution (Blackwell to Rubin): Every generational transition carries risk. Blackwell Ultra (later CY2026) and Vera Rubin (production H2 CY2026) must execute on the claimed performance improvements or the generational premium compresses. So far, the track record is flawless — but that does not mean it always will be.
Concentration in a single theme: NVIDIA is essentially a single-thesis company: AI infrastructure demand grows exponentially. If that thesis breaks — if AI spending proves to have been pulled forward, if inference costs drop faster than demand grows, if the "AI winter" crowd proves correct — the stock reprices violently. I do not believe that thesis will break, but the risk is non-zero.
Valuation sensitivity: At $4.2+ trillion market cap, even modest growth disappointments could trigger 20-30% corrections. The stock sold off from its 2024 highs on mere whispers of ASIC competition (DeepSeek). High-conviction investors must have the stomach for volatility commensurate with the scale.
Now the critical question: is NVIDIA attractive at current levels?
Current metrics (as of late March 2026):
| Metric | Value |
|---|---|
| Market cap | ~$4.25 trillion |
| Enterprise value | ~4.19trillion(netcash 60B) |
| TTM revenue (FY26) | $215.9 billion |
| EV/S (TTM) | 19.4X |
| FY27E consensus revenue | ~$365 billion |
| EV/S (forward, FY27E) | 11.5X |
| TTM Non-GAAP EPS (FY26) | $4.77 |
| Trailing P/E (Non-GAAP FY26) | ~36.5X |
| FY27E consensus EPS | ~$8.25 |
| Forward P/E (FY27E) | ~21X |
| FY27E growth rate | ~70% |
| PEG ratio (FY27E) | 0.30X |
| TTM FCF | $96.6 billion |
| FCF yield (TTM) | ~2.3% |
| EV/FCF (TTM) | ~43X |
My assessment using the GARP framework:
For the valuation cohort, NVIDIA is growing at approximately 65-70% CAGR. My benchmark table for >50% growth companies shows a typical EV/S range of 12-20X. At 19.4X TTM EV/S, NVIDIA trades at the upper end of that range. However, this is TTM — which understates forward revenue by a significant margin given the growth trajectory. On a forward basis at 11.5X FY27E, NVIDIA trades at a substantial discount to cohort average for its growth rate.
But the more compelling valuation lens for a company this profitable is PEG. At a forward P/E of 21X on approximately 70% growth, the PEG ratio is 0.30X. My threshold for "exceptional value" is PEG < 0.5X. NVIDIA clears that threshold with room to spare. To put this in context: AppLovin — which I have highlighted as one of the best values in the growth universe — traded at a PEG of approximately 0.5X when I recommended it. NVIDIA's PEG is lower.
The FCF multiple is also instructive. At $96.6 billion TTM FCF, the EV/FCF is approximately 43X. If FCF grows proportionally with revenue in FY27 (which is conservative given the operating leverage), FY27 FCF could reach $160-170 billion, implying a forward EV/FCF of 25-26X. That is a reasonable multiple for a company growing at 65%+.
One must also consider: there is no directly comparable company. NVIDIA is the most profitable hardware company ever built, at the fastest growth rate ever sustained at this scale, with the deepest competitive moat in the semiconductor industry. Normal valuation heuristics strain under the weight of these numbers. The relevant question is not "is it cheap relative to Arista?" but "is the growth priced in or not?" At a PEG of 0.30X, I would argue it is not fully priced in.
I have not held NVIDIA in the High Growth Portfolio — my AI infrastructure exposure has been through Nebius, Pure Storage, and Arista, which offered more asymmetric risk/reward at the time of purchase. But applying my framework objectively, NVIDIA's valuation on a growth-adjusted basis is genuinely attractive. The forward P/E of 21X for 70% growth, the PEG of 0.30X, the Rule of 40 score above 100 at $216 billion in revenue, the $96.6 billion in FCF, the re-accelerating growth trajectory, the $1 trillion backlog, the mid-70s gross margins on a hardware business — these are not statistics that describe an overvalued company.
The risks are real — custom silicon displacement in inference, hyperscaler capex cyclicality, China regulatory headwinds, and the omnipresent risk that the AI infrastructure buildout proves less durable than the demand signals suggest. But the question for an investor at today's prices is whether NVIDIA can sustain 40-50% growth for the next three years. Given the hyperscaler capex trajectory ($700B in 2026, projected $1.15T cumulative through 2027), sovereign AI demand, physical AI emergence, and the Vera Rubin cycle, I believe the answer is yes.
Verdict: Buy for new positions. A 3-5% allocation in the High Growth Portfolio is appropriate, with a concentration limit that recognizes the absolute market cap and the single-thesis concentration risk. The beer might not get any colder — but it is still quite cold.
Condition for being wrong: If NVIDIA's year-over-year revenue growth decelerates below 40% for two consecutive quarters without a generational transition as the explanation, the growth premium unwinds and the thesis weakens. If non-GAAP gross margins fall below 70% on a sustained (not one-time) basis, the pricing power thesis is impaired. Monitor both quarterly.