Lumivest
Sign in
HomeETFsWatchNewsHistorySettings
Sign in
Research history/NVDA

NVDA

NVIDIA Corp

Solid

AI-accelerator monopolist with ~105% TTM ROIC and ~75% latest-quarter gross margin; forward PE 20.8x and PEG 0.40 are undemanding for ~87% revenue growth, but undiversifiable hyperscaler concentration and binary China export risk cap conviction.

Researched 2 days ago

Earnings Aug 25· After close· in 10 weeks

Solid
Conviction76
Upside35
Risk-adj10
Holdbelow $170.43

AI-accelerator monopolist with ~105% TTM ROIC and ~75% latest-quarter gross margin; forward PE 20.8x and PEG 0.40 are undemanding for ~87% revenue growth, but undiversifiable hyperscaler concentration and binary China export risk cap conviction.

ScoresAnalyst DDAnalystSince lastRed flagsThesisConvictionQualityFinancialsValuationUpsideCycleCatalysts / RisksSizingRed-teamExpectationsFalsifiersSources

Recommendation

Hold

Quality holds but price is in the fair range — no action.

Recommendation

Hold

Conviction

76/100

solid

Upside

35/100

bull 50% · ~40% odds · +20% expected

Risk-adjusted upside

10/100

+10% after downside pressure

Thesis quality

10.0/10

Opportunity

7.1/10

Risk pressure

9.1/10

Valuation

Fair

AI fair value

$235.00

Fundamentals check

$210.69

12-24mo fair-value range

$180.00 / $235.00 / $280.00

width 43%

Buy below

$170.43

Trim above

$336.00

Implied expectations

achievable

46.2% implied revenue CAGR

Today's P/S implies ~46.2% revenue CAGR for five years versus 67.4% realized growth.

Analyst take (Substack DD)

  • damnangPartial

    Agree that CUDA plus multi-precision hardware (FP8/FP4) and HBM bandwidth scaling structurally advantage NVDA in training and frontier serving, but the post understates Cerebras as real inference-decode pressure given its public benchmark lead and >$20B OpenAI deal — a niche, not negligible, competitive datapoint. Numbers in the post are not relied upon as primary evidence.

  • damnangAgrees

    NVDA being cited as first CoPoS customer with Rubin Ultra mixing CoWoS/CoPoS is consistent with the packaging-roadmap leadership the structured margins and ROIC imply; treated as qualitative platform-expansion signal, not verified volume.

  • damnangPartial

    The Vera Rubin SOCAMM/DRAM-cut and edge-AI posts are neutral execution context; plausible direction but no structured provider in the packet confirms the configuration change, so it informs narrative only, not the verdict.

Since Last Research

Material changes

  • Price moved $205.19 → $208.49 (+1.6%); forward PE 20.8x / PEG 0.40 and core fundamentals unchanged.
  • New Mizuho TIER_A note (Jun 16) reaffirms no CPO/800V delays and raises optical-engine projections on NVDA demand — reinforces near-term catalyst.
  • New TIER_A Substack (Jun 16) names NVDA the first customer for TSMC CoPoS mass production (Rubin Ultra) — extends packaging roadmap signal.

Unchanged thesis elements

  • Verdict remains CONSIDER; wide moat, net-cash balance sheet, and AI-bottleneck position structurally unchanged.
  • AI thesis scores unchanged: direct exposure 10, pricing power 9, defensibility 9.
  • Q1 FY2027 fundamentals (74.9% gross margin, 65.6% operating margin, ~105% TTM ROIC) unchanged — no new earnings event.
  • Hyperscaler concentration and binary China export-control risk unchanged; sizing band held at 2-5%.

Red flags

  • Hyperscaler customer concentration: majority of Data Center revenue tied to a small set of hyperscalers, multiple >10% customers disclosed historically.
  • US export-control overhang: H20/China restrictions blocked material revenue; further tightening remains a binary policy risk (symmetric tail risk, itemized).
  • Insider net selling: Director Mark Stevens sold ~1.3M shares (early June 2026, ~$217-222) and directors Neal/Dabiri trimmed; only small Gawel acquisitions offset.
  • Elevated absolute multiples: ~33x trailing PE and ~21x P/S leave limited margin of safety if consensus is missed.

AI-Infrastructure Thesis Fit

AI accelerators / compute platformNetworking / custom ASICs / switching siliconOptical / CPO / silicon photonicsAI servers / hardware integration
Direct exposure10/10
Pricing power9/10
Defensibility9/10

NVDA is the central bottleneck of AI infrastructure. Data Center dominates revenue; CUDA, NVLink/Spectrum-X, and the Blackwell/GB300/Rubin cadence entrench the platform. 74.9% latest-quarter gross margin and ~105% TTM ROIC confirm pricing power and value compounding. Curated TIER_A commentary flags CPO and 800V DC ramping ahead of schedule (small-volume Q3 2026) and persistent multi-year memory/optical supply constraints — platform-expansion and demand-durability signals (qualitative only). The structural 18-36mo risk is custom-ASIC displacement (Google TPU, Amazon Trainium, Meta MTIA) eroding inference share.

Conviction Assessment

Moat

CUDA + NVLink lock-in, ~105% ROIC and ~75% gross margin confirm a wide, durable moat.

Bottleneck fit

Direct exposure 10/10 — the central AI-compute bottleneck pulling demand through the whole stack.

Valuation

Forward PE 20.8x / PEG 0.40 below NVDA's own history is undemanding for the growth and margins.

Catalyst

Q2 FY27 print (late Aug) plus CPO and 800V small-volume Q3 2026 ramps are specific, dated readouts.

Why not higher

Undiversifiable hyperscaler concentration and binary China export-control risk keep this a high-quality CONSIDER rather than the rare HIGH_CONVICTION tier.

Bull

  • Supply-constrained, multi-year AI/memory/optical demand sustains Data Center growth above the deceleration the multiple implies.
  • Rubin/Rubin Ultra cadence plus CPO and 800V platform expansion widen the system-level moat.
  • Net-cash balance sheet and ~60% FCF margin fund buybacks and R&D through any air pocket.

Bear

  • Hyperscaler capex digestion or a supplier shift hits concentrated Data Center revenue hard.
  • Tighter China export controls trigger a binary revenue and margin write-down.
  • Custom ASICs (TPU/Trainium/MTIA) erode inference share over 18-36 months.

Business Quality

Description

NVIDIA designs AI accelerators (GPUs), the CUDA software stack, and networking silicon (NVLink, Spectrum-X) that together form the dominant compute platform for AI training and inference.

Value Chain

Core infrastructure/platform layer — sits at the most constrained node of the AI value chain, pulling demand through foundry, HBM, packaging, optical and power suppliers.

Moat

wide

CUDA software lock-in, NVLink/Spectrum-X system integration, and a relentless annual product cadence (Blackwell/GB300/Rubin) create deep switching costs; ~105% TTM ROIC and ~75% gross margin confirm durable pricing power.

Pricing Power

high

Latest-quarter gross margin 74.9% and operating margin 65.6% on accelerating volume demonstrate the ability to raise price within the compute bottleneck without losing share.

Customer Concentration

High hyperscaler concentration — a small set of hyperscalers drive the majority of Data Center revenue, with multiple >10% customers disclosed historically. Undiversifiable in the near term.

Financial Health

MetricValue
Revenue growth YoY86.90%
Gross margin74.93%
Operating margin65.60%
FCF margin59.53%
Cash position80501000000
Net debt / EBITDA-0.44
Share count change YoY—
ROIC105.41%

Valuation

Forward P/E

20.82

Trailing P/E

32.98

PEG

0.40

EV/EBITDA

31.77

P/S

20.76

Sector: premiumHistory: below_average
PeerMetricValue
AVGOcontextAI networking/custom-ASIC peer; lower pure-AI revenue concentration and narrower gross margin than NVDA
AMDcontextClosest GPU rival; materially lower gross margin and ROIC; CUDA software gap remains wide
MUcontextHBM memory supplier upstream of NVDA; cyclical demand-pull beneficiary, not a direct valuation comp
INTCcontextLegacy semi with minimal direct AI-accelerator share; not a competitive valuation anchor

Upside Case

Bull fair value

$320.00

Probability

40%

Horizon

Long term

Rubin/Rubin Ultra ramp sustains Data Center growth above priced decelerationSovereign AI and inference TAM expansion broaden the demand baseCPO/800V/CoPoS platform expansion deepens system-level lock-in

Cycle Position

YTD

—

1Y

43.35%

Vs sector

—

From 52w high

-11.86%

Valuation: cheaperAnalysts: neutralInsiders: net_selling

Catalysts & Risks

Near-term catalysts

  • Q2 FY2027 print (~late Aug 2026) — first clean read on Rubin ramp and sequential Data Center growth; watch gross-margin and China-revenue commentary
  • CPO switch mass production beginning with small-volume customer deliveries in H2/Q3 2026 (per TIER_A SVP commentary) — validates optical platform expansion
  • 800V DC small-volume shipments pulled to Q3 2026 ahead of schedule (per TIER_A commentary) — next-gen power architecture execution proof

Medium-term catalysts

  • Rubin / Rubin Ultra generation ramp using a mix of CoWoS and CoPoS (NVDA cited as first CoPoS customer) through 2026-2027
  • Resolution or further tightening of US-China export controls on advanced accelerators
RiskSeverityExplanation
Hyperscaler customer concentrationhighA small set of hyperscalers drive the majority of Data Center revenue with multiple >10% customers disclosed historically; a capex pause or supplier shift would hit revenue sharply. Undiversifiable.
US-China export-control overhang (binary policy tail risk)highH20/China restrictions have blocked material revenue; further tightening remains a binary policy risk. Itemized symmetrically rather than left implicit in prose.
Custom-ASIC displacement in inferencemediumGoogle TPU, Amazon Trainium, Meta MTIA and open scale-up standards (per TIER_A Substack on ALAB/NVLink) could erode NVDA inference share over 18-36 months.
Valuation compression on a consensus missmedium~33x trailing PE and ~21x P/S leave limited margin of safety if revenue or margin guidance disappoints.
Inference competition from wafer-scale designslowCerebras WSE-3 posts higher raw decode speed on public benchmarks and holds a >$20B OpenAI inference deal, a niche but real competitive datapoint (per TIER_A Substack).

Position Sizing

2%–5%

High-quality CONSIDER: wide moat, ~105% TTM ROIC, net-cash balance sheet, forward PE 20.8x / PEG 0.40 against ~87% revenue growth. Band held at 2-5% because hyperscaler concentration is undiversifiable and China export risk is binary; entry at ~12% below the 52W high supports scaling toward the upper half on Q2 confirmation.

Adversarial Review

Thesis survives

Strongest bear case

NVDA is a peak-cyclical capex bet dressed as a durable platform. The ~$5T valuation and the consensus ~$10 forward EPS (implied by 208.49/forward PE 20.8x) rest entirely on hyperscaler capex continuing to accelerate — yet the packet states a small set of hyperscalers drive the majority of Data Center revenue, and those same customers (Google TPU, Amazon Trainium, Meta MTIA) are building in-house silicon. Both revenue AND margins are at all-time peaks (74.9% GM, 65.6% OM, 105% ROIC), so earnings are doubly exposed to mean reversion: a capex digestion or ASIC shift compresses volume, price, and margin simultaneously. The 'cheap' PEG 0.40 is therefore circular — it only looks cheap if 52% EPS growth holds; if growth decelerates as the bull admits the multiple prices, forward EPS is overstated and the discount evaporates. Director Stevens dumping ~1.3M shares (~$280M) near the 52W high corroborates the peak read.

Challenges to the thesis

Load-bearing assumptionWhy it might be wrongSeverity
Forward PE 20.8x / PEG 0.40 is 'undemanding' for the growthThe multiple is cheap only because forward EPS ~$10 embeds ~52% growth. That EPS is consensus-derived (no income statement in packet — FMP 403'd), and it assumes the very Data Center re-acceleration at risk. On deceleration, EPS and multiple compress together, so the cheapness is circular.high
Gross margin is 'stable' at ~75%The series contradicts this: annual GM fell from 74.99% (FY25) to 71.07% (FY26), and the 2025-04-27 quarter printed 60.5% — almost certainly the H20/China write-down. Margins have already proven volatile to the exact China risk flagged, so peak-margin durability is not established.high
Multi-year supply-constrained demand (CPO/800V/memory) sustains growth above priced decelerationNearly all of these TIER_A demand signals come from a single anonymous handle (@aleabitoreddit), uniformly bullish. The +2 conviction adjustment rests on this promotional confluence, not on disclosed orders or guidance.medium
Hyperscaler concentration is a near-term-undiversifiable but contained riskThe biggest customers are simultaneously the biggest emerging competitors (in-house ASICs), so concentration and displacement are the same risk, not two; a single >10% customer reallocating capex hits revenue and the displacement thesis at once.medium

Missed / under-weighted red flags

  • Gross_margin_trend labeled 'stable' despite annual decline 74.99%→71.07% and a 60.5% quarter — the China-margin vulnerability is understated.
  • Forward EPS ~$10 is inferred from price/forward-PE, not independently sourced; FMP income/balance/cash-flow all 403'd, so no statements, no guidance, no actual segment/customer disclosure backs the concentration or growth claims.
  • Beta 2.24 — high drawdown amplitude in an AI-capex unwind is not reflected in the 2-5% sizing rationale.
  • +2 conviction adjustment is sourced almost entirely from one anonymous bullish X handle, an evidence-quality weakness in the process.
  • Bull-case FV $280 double-counts (28x multiple AND elevated forward EPS) while the bear $150 does not symmetrically compound multiple-and-EPS compression, understating downside skew.
  • Insider selling is broad (Stevens, Neal, Dabiri all trimming near highs); only token Gawel buys offset — net-selling signal near peak is real.

Fair-value critique

Fair value $235 = 23.5x forward EPS — both inputs are market/consensus-derived, so '$235 base' mostly says 'modestly above consensus,' not an independent anchor. The forward EPS (~$10) embeds the 52% growth the bear disputes, and the 23.5x is a premium to today's 20.8x; the method assumes multiple and EPS are independent when they are correlated. In a deceleration both fall together, so true downside runs below the $180 'low' (which only flexes the multiple to 18x while holding EPS). The range is right-skewed by construction and understates the peak-earnings × peak-multiple tail.

Independent red-team pass · claude-opus-4-8 · 2026-06-17

Implied Expectations

stretched

What the price implies

  • Forward PE ~20.8x against ~52% implied EPS growth (PEG 0.40) embeds a sharp deceleration and/or margin compression within 2-3 years
  • Trailing multiple (33x PE, 20.8x P/S) sits below NVDA's own 5yr average, implying the market treats current earnings as near a cyclical peak
  • EV/EBITDA ~31.8x assumes Data Center capex normalizes rather than re-accelerates

Our variant view

Price implies a faster Data Center deceleration than the order book, CPO/800V acceleration, and multi-year HBM/optical supply constraints support; we expect FY27 growth to decelerate more slowly than priced, supporting a higher multiple.

Thesis Falsifiers

Pre-committed, dated checks that would disconfirm the thesis — a review is flagged automatically as each date passes.

If we observe…By…the thesis is wrong because
Q2 FY27 print shows Data Center revenue growth decelerating sharply or gross margin below 70%, with weak forward guidance2026-08-31Demand/margin durability thesis weakening — downgrade toward WATCH and re-evaluate the multiple.
CPO switch and 800V DC H2/Q3 2026 deliveries slip materially versus the on-schedule TIER_A commentary2026-11-30Platform-expansion catalyst broken — remove from near-term catalysts and reassess execution credibility.
A disclosed >10% hyperscaler customer publicly cuts GPU orders in favor of in-house ASICs2027-02-28Concentration risk materializing into the displacement thesis — cut sizing and downgrade.

Sources

ClaimSourceURLRetrieved
Current price $208.49, market cap ~$5.02T, +0.52% on dayfinnhub:quote—retrieved 2026-06-17T13:47:11.232Z
Forward PE 20.82, PEG 0.40, trailing PE TTM 32.98, EV/EBITDA TTM 31.77, P/S TTM 20.76finnhub:basic-financials—retrieved 2026-06-17T13:47:11.057Z
Latest-quarter gross margin 74.93%, operating margin 65.6%, net margin 71.46%, FCF margin 59.53% (2026-04-26)finnhub:basic-financials—retrieved 2026-06-17T13:47:11.057Z
ROIC TTM 105.41%; net cash (net debt/EBITDA ~ -0.44)finnhub:basic-financials—retrieved 2026-06-17T13:47:11.057Z
Sales/share QoQ-YoY ~86.9% (3.346 vs 1.790); FY-YoY ~67.4% (8.809 vs 5.261)finnhub:basic-financials—retrieved 2026-06-17T13:47:11.057Z
52W high 236.54 (2026-05-14), 52W low 142.03; 1Y price return 43.35%; ~11.9% below 52W highfinnhub:basic-financials—retrieved 2026-06-17T13:47:11.057Z
Cash/share quarterly 3.3265 × ~24.2B shares ≈ $80.5B cashfinnhub:basic-financials—retrieved 2026-06-17T13:47:11.057Z
Analyst ratings Jun-2026: 24 strong buy, 39 buy, 4 hold, 1 sell (broadly stable vs Mar)finnhub:recommendations—retrieved 2026-06-17T13:47:10.990Z
Insider net selling: Stevens sold ~1.3M shares early Jun 2026 (~$217-222); Neal/Dabiri trimmed; small Gawel acquisitionsfinnhub:insider-transactions—retrieved 2026-06-17T13:47:11.034Z
Peers: AVGO, MU, AMD, INTC, TXN, MRVL, QCOMfinnhub:peers—retrieved 2026-06-17T13:47:11.001Z
Most recent 10-Q filed 2026-05-20sec:recent-filingsLinkas of 2026-05-20 · retrieved 2026-06-17T13:47:11.047Z
Macro backdrop: 10Y 4.47%, Fed funds 3.63%, HY OAS 2.71% (2026-06-15)fred:macro-snapshot—as of 2026-06-15 · retrieved 2026-06-17T13:17:20.784Z
NVDA SVP refutes analyst delay reports; H2 CPO product delivery on schedule, mass production ramping (qualitative)x:@aleabitoredditLinkretrieved 2026-06-17T13:47:10.821Z
Mizuho (Jun 16): no CPO/800V delays, optical engine projections revised up on NVDA demand (qualitative)x:@aleabitoredditLinkretrieved 2026-06-17T13:47:10.821Z
800V DC pulled to Q3 2026 with small-volume shipments ahead of schedule (qualitative)x:@aleabitoredditLinkretrieved 2026-06-17T13:47:10.821Z
Cerebras WSE-3 inference-decode competitive analysis; CUDA/precision flexibility argued as NVDA structural edgesubstack:damnangLinkretrieved 2026-06-17T13:47:10.825Z
NVDA cited as first customer for TSMC CoPoS mass production; Rubin Ultra to mix CoWoS and CoPoSsubstack:damnangLinkretrieved 2026-06-17T13:47:10.825Z

Analyst DD (Substack)

  • 6Substack · damnang·BEARISH / MOAT·19 days agopost

    The author frames NVIDIA's GPU architecture as structurally advantaged over Cerebras's wafer-scale design: hardware support for multiple precisions plus the CUDA software ecosystem lets rapid software innovation convert directly into performance, while Cerebras's fixed FP16 hardware is 'forever chasing one generation behind.' Even where Cerebras wins on raw decode speed, NVIDIA retains workload portability and dominates training and frontier-scale serving.

    Evidence
    • NVIDIA published 1,038 tokens/s for DGX B200 on Llama 4 Maverick vs Cerebras's 2,522 tokens/s, but under different conditions (batch size, precision, speculative decoding)
    • HBM bandwidth climbs each generation: H100 ~3.35 TB/s, H200 4.8 TB/s, B200 8 TB/s
    • HBM capacity scaling fast: H200 141GB, B200 ~180GB per GPU, projected past 1TB by 2028
    • NVIDIA added dedicated FP8 Tensor Cores starting with H100 and extended to FP4 with Blackwell, cutting compute and power
    • Major inference frameworks (vLLM, FlashAttention, Triton, CUTLASS, xformers) are all built on CUDA; any model/framework runs on NVIDIA
    • All major frontier labs train on NVIDIA GPUs, Google TPUs, or their own chips; the $20B OpenAI Cerebras contract is inference only
    • On AWS Bedrock, Trainium handles prefill and Cerebras is used only as a decode accelerator
    Catalysts
    • NVIDIA already moving to FP4 (Blackwell) while Cerebras would need 2-3 years to add FP8 in WSE-4
    • HBM capacity projected past 1TB by 2028, widening the capacity gap vs Cerebras's ~44GB SRAM
    Risks
    • Cerebras sits at the top tier for inference decode speed on public benchmarks (2,522 vs 1,038 tokens/s)
    • Cerebras signed a >$20B multi-year OpenAI compute deal and an AWS Bedrock deployment
    • Benchmark comparison conditions differ, so the speed gap may not be apples-to-apples