NVDA — The AI tax collector, hiding in plain sight as a value stock

Screening result: all 3 hard criteria pass. Trailing 3-year ROE above 15% — check. Positive free cash flow — check. Valuation reasonable relative to peers and historical range — check. Today's primary candidate, Texas Instruments, failed on valuation: its P/E of ~48× runs 70–80% above its own 5-year average of ~27×. NVIDIA Corporation (NASDAQ: NVDA) was the backup candidate and cleared the bar on every count, including the one that surprises most people: at a trailing P/E of 30.69×, it trades at a 50% discount to its own 5-year average of 61.54×.

That inversion is the thesis in a sentence. NVIDIA's stock has risen roughly 39% over the past year, yet the P/E has compressed dramatically — because earnings grew faster than the share price. When EPS triples, the multiple comes down even if the stock goes up. At $200.42 (June 10 close), the question is whether the market has finally stopped paying for the story and started pricing the cash.

What the company actually does

NVIDIA designs semiconductors — it does not manufacture them. The company's GPUs (graphics processing units, originally built for gaming) turned out to be almost uniquely suited to the matrix math at the core of modern AI model training and inference. That accidental alignment became a structural advantage when the generative AI wave hit in 2022–2023.

Today NVIDIA sells integrated computing platforms, not discrete chips. The flagship product line is the Blackwell GPU architecture (H100/H200 predecessors, now GB200/B200), which ships inside rack-scale systems called DGX SuperPODs. These systems combine GPU compute, NVLink high-bandwidth interconnects, InfiniBand networking, and the CUDA software stack into what CEO Jensen Huang describes as an "AI factory."

Revenue by segment in Q1 of fiscal 2027 (the quarter ending April 26, 2026): 1

Data Center: $75.2 billion — 92.2% of total
Edge Computing (Gaming, Professional Viz, Automotive combined): $6.4 billion — 7.8%

Gaming, once the core business, is now a rounding error.

Screening criteria: the three gates

Gate 1 — ROE track record

Return on equity measures how much profit a company generates per dollar of shareholders' equity. The channel's minimum threshold is 15% sustained over 3 consecutive fiscal years. NVIDIA's fiscal year ends in late January.

Fiscal year	ROE	Gate
FY2023 (ended Jan 29, 2023) — SEC XBRL verified: net income $4.37B ÷ avg equity $24.36B	17.93%	✅
FY2024 (ended Jan 28, 2024)	91.46%	✅
FY2025 (ended Jan 26, 2025)	119.18%	✅
FY2026 (ended Jan 25, 2026)	101.48%	✅
TTM (ended Apr 26, 2026)	114.29%	✅

FY2023 figures — net income $4.368B, average equity $24.357B — were verified against SEC EDGAR XBRL data for NVIDIA Corp (CIK 0001045810). 2 The FY2023 year was deliberately chosen as the hardest test: crypto-driven gaming demand had collapsed, and revenue barely grew (+0.22%). NVIDIA still cleared 15% ROE. The explosive expansion in FY2024–FY2026 reflects the AI demand surge, not a business reinvention.

One technical note: when net income grows faster than retained equity grows, ROE can exceed 100%. That is the arithmetic here — NVIDIA's earnings have compounded at roughly 204% per year over the past three years while equity has grown more slowly. The ROE number is real; the underlying driver is the margin structure (62.97% net margin TTM), not financial engineering.

Gate 2 — Free cash flow

正在加载图表…

FY2023: $3.81B. FY2024: $27.22B. FY2025: $60.53B. FY2026: $96.68B. TTM: $119.08B (operating cash flow $125.65B minus capex $6.57B). 3 FCF yield on the $4.85T market cap: 2.45%. Per-share FCF: $4.87.

Capital expenditure is startlingly low ($6.57B on $253.5B of TTM revenue) because NVIDIA is fabless — it pays TSMC to manufacture the chips and does not own the factories. Nearly every dollar of operating cash converts to free cash flow.

Gate 3 — Valuation

The gate requires "reasonable valuation relative to sector peers and historical range." Both tests pass.

vs. 5-year own history: Current trailing P/E is 30.69×. Over 21 quarter-end observations from July 2021 through June 2026, NVIDIA's average trailing P/E was approximately 61.54× — the multiple peaked at 147.30× in April 2023 when EPS was $0.19. 4 Current P/E is 50.1% below that 5-year average. The compression is earnings-driven: EPS grew from ~$0.19 in mid-2023 to $6.53 TTM, a 34× increase in about three years.

vs. semiconductor peers (all data June 10, 2026): 5

Ticker	Trailing P/E	Forward P/E	PEG	EV/EBITDA	ROE
NVDA	30.69×	20.17×	0.45	29.08×	114.3%
AMD	150.86×	51.92×	0.92	98.14×	8.1%
AVGO	61.93×	23.64×	0.52	43.14×	37.3%
INTC	n/a (loss)	101.31×	1.44	38.80×	−2.9%
MU	41.96×	9.21×	0.06	27.17×	39.8%
QCOM	20.89×	19.57×	7.59	15.92×	36.1%
TXN	48.20×	34.21×	1.63	30.65×	32.4%
ADI	58.46×	28.36×	1.30	31.96×	9.6%
Peer median	58.46×	28.36×	1.29	35.38×	24.2%

NVIDIA's trailing P/E is 47.5% below the peer median. Its PEG of 0.45 — a ratio of P/E to forward earnings growth rate, where 1.0 is considered fairly priced — sits 65% below the peer median of 1.29. The one metric where NVIDIA looks expensive is P/B at 24.83×, which is 110.6% above the peer median of 11.79×. That premium is a mechanical consequence of 114% ROE: high-return businesses destroy book value efficiently (via buybacks) while generating enormous earnings, so P/B naturally inflates.

Forward P/E: two estimates are in circulation. StockAnalysis uses consensus EPS of ~$9.94, yielding a forward P/E of 20.17×. Finviz uses a higher estimate (~$12.48), yielding 16.05× — below the S&P 500's ~19× average. Barron's ran a piece on June 10 noting this unusual positioning for a hypergrowth company. 6

Revenue and earnings growth

正在加载图表…

*Q1 FY27 is a single quarter, shown for trend context only.

Revenue went from $16.7B in FY2021 to $215.9B in FY2026 — a 13× increase in five fiscal years. The compound annual growth rate over that period is roughly 67%. 7 8

The FY2023 flat spot (revenue up just 0.22% YoY) is worth flagging: gaming demand cratered post-crypto-boom, and NVIDIA's consumer GPU business fell sharply. The company survived entirely on the then-nascent data center segment, which has since grown from a supporting act to 92% of revenue. That concentration is both the source of NVIDIA's current power and one of its core risk factors.

Most recent quarter — Q1 FY2027 (ended April 26, 2026): revenue $81.6B (+85% YoY), GAAP EPS $2.39, non-GAAP EPS $1.87. Data Center revenue $75.2B (+92% YoY, +21% quarter-over-quarter). Q2 FY2027 guidance: $91.0B ±2%, above the Bloomberg consensus estimate of $87B at the time. 9

TTM margin profile: gross margin 74.15%, operating margin 64.02%, net margin 62.97%. For context, AMD's gross margin runs around 53%, Intel's around 37%. S&P Global Ratings, which upgraded NVIDIA to AA− in April 2024, noted: "NVIDIA's competitive moat is wide and sustainable." 10

Balance sheet

The balance sheet is, in a word, overpowered. 11

Balance sheet snapshot (TTM)

As of April 26, 2026

Net cash

$0.00

Debt/equity ratio

0.07

Interest coverage

544×

Current ratio

3.44

正在加载统计卡片…

Total cash and short-term investments: $80.57B. Total debt: $12.35B. Net cash position: $68.22B ($2.79 per share). The debt-to-equity ratio of 0.07 means NVIDIA is essentially debt-free relative to its equity base. Interest coverage of 544× means annual interest expense is less than 0.2% of operating income. Current ratio of 3.44 — every $1 of short-term liabilities is covered by $3.44 of short-term assets.

Credit ratings: S&P AA− (stable outlook, upgraded from A+ on April 30, 2024); Moody's Aa1 (upgraded from Aa2 on January 20, 2026). 10 Both agencies are in upgrade mode, not downgrade watch.

Capital return: NVIDIA returned $41.1B to shareholders in FY2026. Following Q1 FY2027 results, the board approved an additional $80B buyback on top of the existing $58.5B authorization — a combined ~$118.5B available. The quarterly dividend was raised from $0.01 to $0.25 per share (a 25× increase), yielding 0.48% annualized at current prices. The payout ratio is 15.3%, leaving ample room for further hikes. 9

Competitive positioning and moat

NVIDIA's defensibility rests on three interlocking layers.

Layer 1: CUDA software lock-in. CUDA (Compute Unified Device Architecture) is the programming framework that lets developers write code to run on NVIDIA GPUs. It has been in active development for over 20 years and has accumulated more than 4 million registered developers across 40,000+ organizations. 12 AlphaStreet's analysis of the ecosystem noted: "CUDA's advantage is that it has had more time to accumulate production hardening, third-party integrations, and institutional knowledge across millions of developers." Switching away requires porting code, requalifying models, and retraining staff — a process measured in quarters, not weeks. Higher-level frameworks advertise backend portability, but the fast path in production AI systems runs CUDA-first.

Layer 2: Market share and scale. NVIDIA holds approximately 76–80% of the AI accelerator market by revenue as of 2026 — down from a peak of ~87% in 2024 as AMD and custom silicon scale, but still generating revenue of $193.7B from Data Center alone in FY2026. 13 Silicon Analysts put it plainly: "The real question is not whether NVIDIA 'loses' — it is how large the overall market becomes." In discrete GPU for consumer gaming, NVIDIA's market share reached 94% in Q4 2025 per Jon Peddie Research, with AMD at 5%.

Layer 3: Supply chain leverage. NVIDIA commands an estimated 60% of TSMC's CoWoS advanced packaging capacity — the technology needed to combine multiple GPU dies on a single package at scale. AMD, by comparison, holds roughly 11% of that capacity. At the gross margin level, NVIDIA's overall 74.15% margin (H100 at ~88%, B200 at ~84%) funds an R&D pipeline that competitors at 37–53% margins cannot easily replicate. CEO Jensen Huang described the current product cycle at Q4 FY2026 earnings: "Grace Blackwell with NVLink is the king of inference today — delivering an order-of-magnitude lower cost per token." 8

Upcoming platform: the Vera Rubin architecture (pairing NVIDIA's first custom CPU, Vera, with the Rubin GPU) is slated for mass market release in H2 2026. At Computex 2026, NVIDIA also unveiled DSX OS — a data center operating system layer — and MaxLPS technology, which allows AI factories to safely deploy up to 40% more GPUs within the same power envelope. 14 That translates to 40% more compute, 40% more tokens, 40% more revenue from the same physical facility — a compelling selling point for hyperscaler customers managing power constraints.

Risk factors and red flags

Investing in NVIDIA at $4.85T requires engaging honestly with the risks. Several are significant.

Regulatory and geopolitical exposure. The DOJ issued an antitrust subpoena to NVIDIA in September 2024 regarding its position in the AI chip market. Separately, China's market regulator ruled in September 2025 that NVIDIA violated conditions attached to its 2020 Mellanox acquisition — that investigation is ongoing. 15 The White House in April 2026 banned H20 and MI308 chip exports to China but subsequently reached an agreement with NVIDIA and AMD: a 15% licensing fee on China-bound chip sales in exchange for export permits. Taiwan is separately weighing tighter AI chip export controls to align with U.S. policy. 16 On June 9, Senator Elizabeth Warren invited CEO Jensen Huang to testify before the Senate Banking Committee on June 11 regarding AI and China chip sales. Huang declined citing scheduling conflicts. Warren's public response: "If Mr. Huang has time to attend a $1 million-a-head dinner at Mar-a-Lago and fly across the world to meet with President Xi Jinping of China, he should be able to find time to answer questions from Congress." 17 None of these proceedings has produced a material financial penalty yet, but the regulatory calendar is active.

Customer concentration and AI CapEx dependency. Five hyperscalers — Microsoft, Meta, Google, Amazon, Oracle — account for the vast majority of Data Center revenue. Microsoft has publicly committed to doubling its AI infrastructure over the next two years, but that plan depends on forecasts made during the current AI buildout cycle. Any single hyperscaler reducing spend would produce a visible impact on NVIDIA's quarterly results. All five are also building proprietary silicon alternatives (Google TPU v6, AWS Trainium 2, Microsoft Maia 100, Meta MTIA 2) that could reduce their dependence on merchant GPUs over time.

Competition. AMD's MI300X is the most credible external challenge today. Its forthcoming MI400 (432GB HBM4, 40 petaflops FP4) is expected in H2 2026, with S&P Global projecting ~$15B in revenue for that product. Broadcom designs custom ASICs (application-specific integrated circuits) for Google's TPU line and is reportedly pursuing similar custom-silicon contracts with other hyperscalers. Neither AMD's ROCm software ecosystem nor Broadcom's custom-only model replicates CUDA's breadth, but competition at the margins of NVIDIA's market share is real and growing. 18

TSMC concentration and supply chain. NVIDIA has no manufacturing of its own — everything runs through TSMC's fabs in Taiwan. Geopolitical risk to Taiwan is a single-point-of-failure scenario. Separately, HBM (high-bandwidth memory) supply is concentrated at SK Hynix and Samsung. NVIDIA signed a multi-year HBM partnership with SK Hynix in June 2026, and Jensen Huang traveled to Seoul to meet with Samsung on June 10. The supply chain is managed — not eliminated as a risk.

Insider selling. Over the past six months, every recorded insider transaction has been a sale — none have been purchases. The most notable: board member Mark Stevens sold 500,000 shares on June 2 ($111.2M at $222.38/share) and another 500,000 shares on June 4 ($109.9M at $219.83/share), totaling ~$221M in eight days. EVP Ajay Puri sold 300,000 shares in March for $54.7M. 19 Jensen Huang, notably, has no insider transactions in the Finviz record for the past six months. Insider selling at a stock near all-time highs by executives with large compensation grants is not unusual, but the absence of any buying at current prices is worth noting.

Short interest: 1.23% of float (284.7M shares short, days-to-cover 1.72). This is extremely low — the market is not heavily betting against NVIDIA. 19

Litigation. At least 20 active federal cases as of 2026, including five patent suits, a securities class action on appeal at the Ninth Circuit, and a copyright claim. None appear material relative to NVIDIA's cash position, but the litigation volume reflects the company's profile. 20

Near-term catalysts and market context

52-week range: $140.85 – $236.54. At $200.42, the stock is 15.3% below its 52-week high and 42.3% above its 52-week low. RSI 14-day: 41.07 — approaching, not yet in, oversold territory.
Next earnings: Q2 FY2027, expected around August 20, 2026. Guidance issued: $91.0B ±2%.
Dividend: $0.25/quarter ($1.00 annualized). Ex-date was June 4, 2026; payment date June 26. Yield: 0.48% at current price.
Buyback: ~$118.5B authorized and available — roughly 2.4% of market cap.
Analyst consensus: Strong Buy. 62 analysts, average price target $298.42 (StockAnalysis) / $309.93 (Finviz), implying 48.9–54.6% upside. Targets range from $255 (Deutsche Bank, Hold — the lone dissenting voice) to $500 (Robert W. Baird, Outperform). Evercore ISI sits at $413. 5 BofA reiterated Buy post-earnings and raised its target to $350, citing "strong Blackwell ramp." 19
Recent catalysts: NVIDIA confirmed its chips (via Google Cloud) power Apple Intelligence processing. NVIDIA joined the $1.4B funding round for humanoid robotics company Neura Robotics alongside Amazon and Tether. Collaboration with Nebius Group on a Physical AI Living Lab in the UK and Europe using RTX PRO 6000 Blackwell GPUs. SK Hynix multi-year HBM technology partnership signed.
YTD / 1-year return: +7.46% YTD as of June 10, 2026; +39.22% over 1 year; +416.95% over 3 years. Beta: 2.20.

Bull / bear framework

Bull case

Earnings growth has structurally outrun the multiple. At a PEG of 0.45 and a trailing P/E 50% below NVIDIA's own 5-year average, the market is not paying a premium for expected growth — it is paying a discount relative to the growth rate already being delivered. Q2 FY2027 guidance of $91B would represent sequential growth of roughly 12% on top of an 85% year-over-year comp.
The moat is software, not hardware. CUDA's 20+ year development lead and 4M+ developer ecosystem cannot be replicated by shipping faster silicon. Switching costs accrue quietly — each additional CUDA-dependent production workload makes migration more expensive. As long as new model development happens on CUDA-first toolchains, NVIDIA collects a royalty on AI progress without owning a patent.
The balance sheet funds the transition. $68.2B net cash, $119B TTM FCF, ~$118.5B authorized buyback. If AI CapEx were to soften in 2027, NVIDIA could continue returning capital while waiting for the next demand cycle. The FCF generation alone would fund the entire $118.5B buyback authorization in roughly 14 months at current run rates.
Product roadmap maintains the cadence. Vera Rubin in H2 2026, then the next generation after that. Each cycle resets the competitive clock — by the time AMD or custom silicon catches Blackwell, Rubin is shipping.
New verticals are additive. Automotive (2.3B in FY2026), robotics (Neura investment, Isaac platform), and the enterprise edge (RTX Spark, DGX Station for Windows) represent demand that did not exist in FY2023. Data Center won't be 100% of revenue forever — growth from adjacent verticals provides optionality.

Bear case

Hyperscaler CapEx is a cycle, not a law of physics. Microsoft, Meta, Google, and Amazon have all signaled multi-year AI infrastructure buildouts, but capital spending plans get revised when macroeconomic conditions change. NVIDIA's FY2023 flat year — when gaming CapEx collapsed overnight — is a reminder that demand can shift faster than the company's quarterly revenue would suggest. If the five major hyperscalers slow order rates simultaneously, there is no diversified revenue base to buffer the impact.
Custom silicon is a slow-moving substitution, not a scenario to dismiss. Google's TPU v6, AWS Trainium 2, and Microsoft Maia 100/200 are purpose-built for specific workloads at those companies. They won't replace NVIDIA for training frontier models anytime soon, but they can reduce GPU utilization per dollar of inference at scale — which is where the dollar volume is heading. Broadcom's custom ASIC business benefits every time a hyperscaler decides to go proprietary, and Broadcom's market cap reflects the growing opportunity.
Regulatory risk is accumulating, not resolving. An active DOJ antitrust investigation, a Chinese anti-monopoly ruling still in progress, Congressional scrutiny, and a 15% tax on China chip exports — no single item is a binary threat, but the combined overhang represents an unusual regulatory environment for a company of this size. If the DOJ investigation moves toward a structural remedy rather than a fine, the thesis changes.
China revenue is uncertain in both directions. FY2024 China was ~13% of total revenue. After the export ban and licensing deal, the exact current number is unclear. A short-seller report claimed over 20% of FY2026 compute revenue came through China via intermediaries — that figure has not been independently verified. If it is materially true, the 15% licensing fee already negotiated could escalate. If it is false, the China risk is already priced in.
Pervasive insider selling at current prices. Every insider trade in the past six months has been a sale, including $221M of stock sold by a single board member in two days in early June. Jensen Huang has not purchased on the open market either. This does not predict anything by itself, but executives with full information about the business are exiting at $200–$222, not adding.

This article is for informational purposes only and does not constitute investment advice. All financial data sourced from SEC filings, StockAnalysis.com, Finviz, and public earnings releases. Prices as of June 10, 2026 close. Investors should conduct their own due diligence before making any investment decisions.