NVDA
NVIDIA Corporation
AI Chips & Compute
~$216⚠
Mkt Cap ~$5.24T⚠
52W: $120.28 – $238+
Next earnings: ~Aug 2026 (Q2 FY27)
↓ ADD (PULLBACK)
Entry zone: $210–$235
Q1 FY27 beat was expected ($81.6B vs $78.8B consensus); the market sold off 3% and missed the $91B Q2 guide — 7–9% above the buy-side whisper — with zero China DC revenue in any forward number. The sharper debate: Huang conflated inference revenue growth with inference market share on the call — Anthropic was cited as a win while Anthropic's primary gigawatt-scale workloads run on AWS Trainium and Google TPUs. At 22× NTM P/E, networking ($14.8B, +199% YoY) is unpriced as a franchise and NVLink Fusion locks hyperscaler custom silicon into NVIDIA's rack architecture — the system-level moat holds even as inference-chip share erodes.
Fwd P/E (NTM)
22×
PEG 0.52 (g=43% NTM)
vs 3Y avg ~45×⚠
Rev YoY (Q1 FY27)
+85%
QoQ ↓ Q2 guide: +12%
Gross Margin
75.0%
4Q → Q2 guide flat
FCF Margin (TTM)
~57%⚠
$48.6B Q1 FY27 quarterly
EPS Revision (90d)
↑ est⚠
Post-Q1 guide raise · 62 analysts
Next Earnings
~Aug 2026
Q2 FY27 · Est
AI Rev %
92.2%
of total (DC Q1 FY27)
DC Rev YoY
+92%
Q1 FY27 · $75.2B
Thesis Durability
A
Variant perception (updated post-Q1 FY27): Q2 guide of $91B is 7–9% above the pre-earnings buy-side whisper — before Q2 has even started. This is the pattern: guidance is set conservatively, actuals beat, next guide resets above prior whisper. Structural revisions have further to run.
Networking is a separate franchise: Data Center networking $14.8B in Q1 FY27 (+199% YoY, +35% QoQ) — now 25% of DC revenue and approaching the size of a standalone major semiconductor company. Jensen Huang stated on the Q4 FY26 call that NVDA is now likely the largest Ethernet company in the world. Not priced into most semiconductor models.1
ASIC moat extension — NVLink Fusion: NVIDIA introduced NVLink Fusion, allowing third-party custom ASICs and custom CPUs to connect directly to NVIDIA's NVLink fabric. Amazon Web Services has agreed to integrate Trainium 4 with the NVLink 6 platform. $2B investment in Marvell Technology (31 Mar 2026) ties the second-largest custom ASIC co-designer into the NVLink ecosystem. The ASIC displacement thesis fails at the system level — hyperscalers using custom silicon still depend on NVIDIA's networking and rack-scale hardware.1,8
Inference via Groq LPU: $20B acquisition of Groq (Dec 2025) delivers SRAM-based LPU technology — 150 TB/s on-chip memory bandwidth vs Rubin GPU's 22 TB/s HBM4. NVDA's Groq 3 LPX rack (256 LPUs + Oberon ETL256 backplane) delivers a reported 35× higher throughput per megawatt than Blackwell NVL72 for trillion-parameter inference. Dynamo software orchestrates prefill (GPU) and decode (LPU) workloads. This neutralises pure-play SRAM inference startups (Cerebras, Etched).8
Sovereign AI: Grew >3× YoY to cross $30B in FY26 (~14% of total revenue). Highly durable, insulated from commercial cloud capex digestion cycles.
Supply commitments: $119B non-cancelable purchase obligations in Q1 FY27 (up from $95.2B in Q4 FY26), validating multi-quarter demand forward visibility. Even under a Rubin delay, commitments are convertible to Blackwell/Blackwell Ultra orders.1
Vera Rubin timeline (updated): Delay risk materially reduced. Mass production blueprints finalised; trial production June 2026; initial commercial deliveries to MSFT, GOOGL, AMZN, META, ORCL in July 2026. ODMs (Foxconn, Quanta, Wistron) commence full-scale production in H2 2026.8
Residual bear case: Custom ASIC growth remains structural at the inference tier — TrendForce projects 44.6% ASIC growth in 2026 vs 16.1% GPU market growth. NVIDIA expected to retain 90% training market share but only 20–30% inference share by 2028. NVLink Fusion partially offsets but does not eliminate this exposure. TSMC CoWoS capacity (1.15M wafers, fully committed) is a structural bottleneck. China DC revenue is zero — $50B TAM excluded.
Inference share vs inference growth (IO Fund, post-Q1): On the Q1 FY27 call, an analyst asked directly about Vera Rubin's impact on inference market share. Management responded by citing growth in inference deployments — conflating "growing share" with growing absolute revenue. The tell: Anthropic was cited as evidence NVDA is "gaining share," yet Anthropic's primary gigawatt-scale workloads run on AWS Trainium and Google TPUs. NVDA's coverage of Anthropic was "largely 0 until recently" — this is catching up from a near-zero base, not share expansion. NVDA retaining 20–30% inference share by 2028 (TrendForce) remains the working assumption; this Q&A did not resolve the concern.6,8
CPU-to-GPU attach rate shift: TrendForce and Arm commentary project the CPU-to-GPU attach ratio to move from 1:8 to 1:2 or 1:1 as agentic AI scales — CPUs handle 50–90% of total workflow latency (orchestration between inference calls). NVDA is addressing this with Vera CPU: $20B visibility in standalone Vera CPU revenue in FY27 is incremental TAM opening a new $200B TAM. Management characterised this as non-cannibalistic ("thinking runs on GPUs, orchestration on CPUs"), but it will direct a portion of compute budget toward CPU spend that would have been pure GPU spend in prior cycles.6
Networking is a separate franchise: Data Center networking $14.8B in Q1 FY27 (+199% YoY, +35% QoQ) — now 25% of DC revenue and approaching the size of a standalone major semiconductor company. Jensen Huang stated on the Q4 FY26 call that NVDA is now likely the largest Ethernet company in the world. Not priced into most semiconductor models.1
ASIC moat extension — NVLink Fusion: NVIDIA introduced NVLink Fusion, allowing third-party custom ASICs and custom CPUs to connect directly to NVIDIA's NVLink fabric. Amazon Web Services has agreed to integrate Trainium 4 with the NVLink 6 platform. $2B investment in Marvell Technology (31 Mar 2026) ties the second-largest custom ASIC co-designer into the NVLink ecosystem. The ASIC displacement thesis fails at the system level — hyperscalers using custom silicon still depend on NVIDIA's networking and rack-scale hardware.1,8
Inference via Groq LPU: $20B acquisition of Groq (Dec 2025) delivers SRAM-based LPU technology — 150 TB/s on-chip memory bandwidth vs Rubin GPU's 22 TB/s HBM4. NVDA's Groq 3 LPX rack (256 LPUs + Oberon ETL256 backplane) delivers a reported 35× higher throughput per megawatt than Blackwell NVL72 for trillion-parameter inference. Dynamo software orchestrates prefill (GPU) and decode (LPU) workloads. This neutralises pure-play SRAM inference startups (Cerebras, Etched).8
Sovereign AI: Grew >3× YoY to cross $30B in FY26 (~14% of total revenue). Highly durable, insulated from commercial cloud capex digestion cycles.
Supply commitments: $119B non-cancelable purchase obligations in Q1 FY27 (up from $95.2B in Q4 FY26), validating multi-quarter demand forward visibility. Even under a Rubin delay, commitments are convertible to Blackwell/Blackwell Ultra orders.1
Vera Rubin timeline (updated): Delay risk materially reduced. Mass production blueprints finalised; trial production June 2026; initial commercial deliveries to MSFT, GOOGL, AMZN, META, ORCL in July 2026. ODMs (Foxconn, Quanta, Wistron) commence full-scale production in H2 2026.8
Residual bear case: Custom ASIC growth remains structural at the inference tier — TrendForce projects 44.6% ASIC growth in 2026 vs 16.1% GPU market growth. NVIDIA expected to retain 90% training market share but only 20–30% inference share by 2028. NVLink Fusion partially offsets but does not eliminate this exposure. TSMC CoWoS capacity (1.15M wafers, fully committed) is a structural bottleneck. China DC revenue is zero — $50B TAM excluded.
Inference share vs inference growth (IO Fund, post-Q1): On the Q1 FY27 call, an analyst asked directly about Vera Rubin's impact on inference market share. Management responded by citing growth in inference deployments — conflating "growing share" with growing absolute revenue. The tell: Anthropic was cited as evidence NVDA is "gaining share," yet Anthropic's primary gigawatt-scale workloads run on AWS Trainium and Google TPUs. NVDA's coverage of Anthropic was "largely 0 until recently" — this is catching up from a near-zero base, not share expansion. NVDA retaining 20–30% inference share by 2028 (TrendForce) remains the working assumption; this Q&A did not resolve the concern.6,8
CPU-to-GPU attach rate shift: TrendForce and Arm commentary project the CPU-to-GPU attach ratio to move from 1:8 to 1:2 or 1:1 as agentic AI scales — CPUs handle 50–90% of total workflow latency (orchestration between inference calls). NVDA is addressing this with Vera CPU: $20B visibility in standalone Vera CPU revenue in FY27 is incremental TAM opening a new $200B TAM. Management characterised this as non-cannibalistic ("thinking runs on GPUs, orchestration on CPUs"), but it will direct a portion of compute budget toward CPU spend that would have been pure GPU spend in prior cycles.6
Conviction: high, with inference trajectory as the open variable. Rubin delay risk resolved; NVLink Fusion locking Trainium 4 into NVIDIA's fabric is the clearest signal the "ASIC displaces NVIDIA" thesis fails at the system level. Networking at $14.8B (+199% YoY) is a separate franchise unpriced in most models. Watch ACIE vs Hyperscale sub-segment mix quarterly — ACIE accelerating faster than Hyperscale is the tell for durable demand breadth.1,6,8
Business Quality
B+
Q1 FY27 actuals: Revenue $81,615M (+85% YoY, +20% QoQ). GAAP gross margin 74.9%; non-GAAP 75.0%. GAAP Net Income $58,321M (includes $15.9B equity securities gains — distorts GAAP EPS to $2.39; non-GAAP EPS $1.87). Q2 FY27 guide: $91B revenue (±2%), 75.0% gross margin (±50 bps).1
New reporting structure (effective Q1 FY27): Data Center now reported as two sub-markets — Hyperscale (public clouds/consumer internet) and ACIE (AI Clouds, Industrial, Enterprise including sovereign AI). Edge Computing replaces the prior segment structure for non-DC revenue. Stock-based compensation is now included in non-GAAP measures (restated historically).8
FCF and capital returns: Q1 FY27 FCF $48,554M (59.5% margin) from $50,344M operating cash flow. $20B returned to shareholders in Q1. New $80B buyback authorisation + $38.5B remaining = $118.5B total buyback capacity. Quarterly dividend raised 25× to $0.25/share.1
ROIC: 90.0% GAAP ROIC in FY2026 (NOPLAT $116B on invested capital $129B). Projected 257%+ in FY2027 on operational leverage. WACC 12.67% — ROIC premium is structurally wide.8
Upstream cost pressures: TSMC price hike 5–10% on <5nm nodes in 2026; 2nm wafers at ~$30K/wafer (50% above 3nm at $20K). CoWoS capacity fully committed for 2026 (1.15M wafers). Non-cancelable supply commitments $119B (up from $95.2B), inventory $25.8B — execution risk if hyperscaler schedules slip. These are real margin headwinds, but Q2 guidance of 75.0% flat GM signals pricing power is passing them through.8
Rubin memory BOM escalation (Morgan Stanley): Rubin memory BOM estimated at $2M per rack — up 435% from the GB300's $374K. Memory could account for 25% of Rubin BOM vs <10% for GB300.⚠ On a $500B+ revenue trajectory, this is a material cost uplift that NVDA must offset via pricing to maintain 75%+ GM. The flat Q2 guidance signals current pass-through capability; this is the #1 metric to watch each quarter as Rubin ramps in volume.
⚠ Insider selling cluster (Form 4, through ~Apr 2026): 15 insiders sold, zero purchased since Dec 2025; all under pre-scheduled 10b5-1 plans. Jensen Huang ~$2.9B sold over 18 months. EVP Ajay Puri ~$148M, Director Mark Stevens ~$100M. Plan-based and proportionally small vs Huang's stake, but sell-to-buy ratio is stark. [Form 4 data: GuruFocus, through ~Apr 2026 — 45-day lag on 13F data]
New reporting structure (effective Q1 FY27): Data Center now reported as two sub-markets — Hyperscale (public clouds/consumer internet) and ACIE (AI Clouds, Industrial, Enterprise including sovereign AI). Edge Computing replaces the prior segment structure for non-DC revenue. Stock-based compensation is now included in non-GAAP measures (restated historically).8
FCF and capital returns: Q1 FY27 FCF $48,554M (59.5% margin) from $50,344M operating cash flow. $20B returned to shareholders in Q1. New $80B buyback authorisation + $38.5B remaining = $118.5B total buyback capacity. Quarterly dividend raised 25× to $0.25/share.1
ROIC: 90.0% GAAP ROIC in FY2026 (NOPLAT $116B on invested capital $129B). Projected 257%+ in FY2027 on operational leverage. WACC 12.67% — ROIC premium is structurally wide.8
Upstream cost pressures: TSMC price hike 5–10% on <5nm nodes in 2026; 2nm wafers at ~$30K/wafer (50% above 3nm at $20K). CoWoS capacity fully committed for 2026 (1.15M wafers). Non-cancelable supply commitments $119B (up from $95.2B), inventory $25.8B — execution risk if hyperscaler schedules slip. These are real margin headwinds, but Q2 guidance of 75.0% flat GM signals pricing power is passing them through.8
Rubin memory BOM escalation (Morgan Stanley): Rubin memory BOM estimated at $2M per rack — up 435% from the GB300's $374K. Memory could account for 25% of Rubin BOM vs <10% for GB300.⚠ On a $500B+ revenue trajectory, this is a material cost uplift that NVDA must offset via pricing to maintain 75%+ GM. The flat Q2 guidance signals current pass-through capability; this is the #1 metric to watch each quarter as Rubin ramps in volume.
⚠ Insider selling cluster (Form 4, through ~Apr 2026): 15 insiders sold, zero purchased since Dec 2025; all under pre-scheduled 10b5-1 plans. Jensen Huang ~$2.9B sold over 18 months. EVP Ajay Puri ~$148M, Director Mark Stevens ~$100M. Plan-based and proportionally small vs Huang's stake, but sell-to-buy ratio is stark. [Form 4 data: GuruFocus, through ~Apr 2026 — 45-day lag on 13F data]
Solid with margin headwind watch. FCF quality strong; ROIC trajectory outstanding. The TSMC pricing and CoWoS capacity constraint are the key variables — flat Q2 GM guidance shows current pass-through power but this will need to be re-checked each quarter.1,5,8
Entry Price Discipline
B+
Post-earnings multiples: Forward P/E 26× (CY26), 19× (CY27) — 60% discount to Magnificent Seven average of 49× (CY26) and 42× (CY27). EV/FCF 28× (CY26), 20× (CY27) — 66% discount to peer average of 83× and 59×. PEG 0.41× vs peer avg 2.61× and S&P 500 avg 1.3×.8
Q1 FY27 beat: $81,615M revenue vs $78.8–79.2B consensus (+3.5%). Non-GAAP EPS $1.87. Q2 FY27 guide: $91B (vs pre-earnings buy-side whisper ~$83–85B, +7–9% surprise). Stock fell ~3% post-earnings — limited negative reaction to a decisive beat, consistent with elevated pre-positioning but not a thesis break.1
Valuation targets: Bear $184–215 (16–18× CY27); Consensus/Fair value $260–278 (Morningstar $260, WS avg $278.73); Bull $315–352 (WF $315, BofA $320, Evercore ISI $352).8
At ~$231: Downside to bear midpoint ($200) is ~13%. Upside to consensus ($269) is ~16%. Modest absolute asymmetry, but the thesis is now cleaner post-catalyst — no binary overhang, Rubin timeline confirmed, $91B Q2 guide sets a new floor for analyst revisions. The pre-earnings entry zone of $185–205 was priced for a catalyst-resolution discount that is no longer available. Entry zone shifts to $210–235 — current price is within zone.
Q1 FY27 beat: $81,615M revenue vs $78.8–79.2B consensus (+3.5%). Non-GAAP EPS $1.87. Q2 FY27 guide: $91B (vs pre-earnings buy-side whisper ~$83–85B, +7–9% surprise). Stock fell ~3% post-earnings — limited negative reaction to a decisive beat, consistent with elevated pre-positioning but not a thesis break.1
Valuation targets: Bear $184–215 (16–18× CY27); Consensus/Fair value $260–278 (Morningstar $260, WS avg $278.73); Bull $315–352 (WF $315, BofA $320, Evercore ISI $352).8
At ~$231: Downside to bear midpoint ($200) is ~13%. Upside to consensus ($269) is ~16%. Modest absolute asymmetry, but the thesis is now cleaner post-catalyst — no binary overhang, Rubin timeline confirmed, $91B Q2 guide sets a new floor for analyst revisions. The pre-earnings entry zone of $185–205 was priced for a catalyst-resolution discount that is no longer available. Entry zone shifts to $210–235 — current price is within zone.
In zone. Accumulate at current price or on weakness. The catalyst binary has resolved, the thesis is validated, and multiples remain materially compressed vs growth and vs sector peers. Pre-earnings binary discount ($185–205) is no longer the right target; fair entry is now $210–235 post-clarity.1,8
Verdict — Add (Pullback)
Q1 FY27 resolved the binary: $81.6B beat ($78.8B consensus), Q2 guide $91B (vs buy-side whisper $83–85B), gross margin 75.0% maintained, $119B supply commitments, $80B new buyback authorisation. The thesis is validated across all three dimensions. At 19× CY27 forward P/E — a 55% discount to the Mag-7 peer average — the stock is not priced for the demand durability that the Q2 guide implies.
The 3% post-earnings sell-off reflects "sell the news" positioning unwind, not a thesis break. At ~$231, the stock is within the revised entry zone of $210–235. The pre-earnings $185–205 zone was a binary-resolution discount; that discount has been collected by those who positioned earlier. New entry at current price is within fair range given the Q2 guide resets the forward multiple lower.
Note on MAGS overlap: MAGS (Roundhill Magnificent 7 ETF) holds ~14% NVDA by weight. Any MAGS allocation carries indirect NVDA exposure — factor this in when sizing a direct position.
The 3% post-earnings sell-off reflects "sell the news" positioning unwind, not a thesis break. At ~$231, the stock is within the revised entry zone of $210–235. The pre-earnings $185–205 zone was a binary-resolution discount; that discount has been collected by those who positioned earlier. New entry at current price is within fair range given the Q2 guide resets the forward multiple lower.
Note on MAGS overlap: MAGS (Roundhill Magnificent 7 ETF) holds ~14% NVDA by weight. Any MAGS allocation carries indirect NVDA exposure — factor this in when sizing a direct position.
Entry zone: $210–235 (current ~$231 is in zone). Size new direct position to 8–10% of total portfolio NLV (medium-high conviction; net of ~1.4% indirect NVDA via MAGS if MAGS = 10% portfolio). Accumulate at current price or on any weakness toward $210. 20% single-stock hard cap.
Opportunity-cost check
▾
vs CSPX
NVDA's forward PEG of 0.35 on 75% projected EPS growth is a materially different risk/reward profile than CSPX's ~22× fwd P/E on ~15% blended earnings growth — the asymmetry is clear enough to justify active capital, subject to entry discipline.
vs current holdings
NVDA stacks directly onto MAGS (~14% NVDA by weight in the Roundhill Magnificent 7 ETF). A direct position adds pure AI Chips & Compute factor exposure on top of existing indirect exposure. MU is held at a meaningful weight — NVDA at 19× CY27 P/E is structurally different from MU's cyclical exposure; not competing for the same allocation thesis. Active picks allocation is materially underweight target (~20% vs ~50% target) — dry powder is the binding constraint, not allocation between names.
Thesis-breakers
▾
- 01 Data Centre revenue growth decelerates below 40% YoY for two consecutive quarters (currently +75% YoY Q4 FY2026; signals structural hyperscaler capex pullback, not a quarter-specific mix issue) — by Q3 FY2027 (Nov 2026 report).
- 02 Non-GAAP gross margin falls below 70% in any quarter (signals either Blackwell cost-of-goods surprise, TSMC CoWoS pricing power shift, or customer pushback on ASP) — any quarter in FY2027.
- 03 Hyperscaler in-house ASIC revenue (Google TPU, Amazon Trainium, Meta MTIA) collectively exceeds 25% of total AI chip silicon spend in annual capex disclosures — by end of CY2027; this would confirm Beth Kindig's thesis that custom silicon is taking training share, not just inference.
Key risks
▾
- 01China export controls (structural zero): Q2 FY27 guide also assumes zero Data Center compute revenue from China. The ~$50B addressable TAM (pre-ban run-rate estimate) is fully excluded. Any re-escalation covering additional SKUs adds further downside; any regulatory softening is pure upside on top of a $91B run-rate. Net: asymmetric optionality but not a near-term catalyst.1
- 02Custom ASIC structural shift (inference tier): TrendForce: ASIC deployments growing 44.6% in 2026 vs 16.1% for GPUs. Custom ASICs expected to handle 37% of DC inference workloads by 2026. By 2028, ASIC shipments projected to surpass GPU shipments (70–75% of production inference). NVDA expected to retain 90% training share but only 20–30% inference share by 2028. NVLink Fusion and Groq LPU mitigate but do not eliminate this structural shift — hyperscalers still deploy custom silicon for high-volume inference.8
- 03TSMC pricing and CoWoS capacity: TSMC is implementing 5–10% price hikes on <5nm nodes in 2026. 2nm wafers are expected at ~$30K each — 50% premium over 3nm at $20K. TSMC's CoWoS capacity for 2026 is 1.15M wafers, fully committed; some customers offering 100% premium for residual capacity. These are direct margin headwinds — NVDA's dual-die Rubin chiplet design mitigates yield risk but does not eliminate cost inflation. Q2 GM guidance of 75.0% flat implies current pass-through; monitor each quarter.8
- 04HBM4 supply concentration: Micron excluded from flagship Rubin HBM4. SK Hynix (70%) and Samsung (30%) are sole sources. Samsung has started mass production and cleared NVDA's 10 Gb/s and 11 Gb/s qualification tiers. SK Hynix still qualifying at the premium 11 Gb/s tier; SK Hynix HBM4E samples not until H2 2026, mass production 2027. A SK Hynix qualification slip or production ramp issue would concentrate Rubin memory supply entirely on Samsung — execution risk.8
- 05Non-cancelable commitment execution risk: $119B in non-cancelable purchase obligations and supply-related commitments (Q1 FY27, up from $95.2B in Q4 FY26). Inventory $25.8B. If hyperscalers adjust procurement schedules or experience delays in data center power availability, NVDA faces downside from pre-committed obligations. Power infrastructure constraints (800V DC architecture discussions with Korean power equipment firms) add a secondary bottleneck to delivery timelines.8
- 06Rubin execution risk (residual): Trial production confirmed June 2026; initial deliveries July 2026 to top CSPs. Main residual risk: SK Hynix HBM4 qualification lag could slow full-volume ramp. A multi-quarter Rubin slip accelerates hyperscaler ASIC evaluation timelines — Beth Kindig (IO Fund): "the product road map is the second line of defense should the CUDA moat be breached." Delay risk is reduced but not eliminated. SemiAnalysis noted Feynman NVL1152 at "early hints" stage — Rubin slip would compress the architecture cadence buffer into 2027.6
DCF scenarios
▾
Discount rate: 11% — high-growth profitable tech (per framework). Terminal growth: 5%. Q1 FY27 FCF: $48,554M (59.5% margin). Q2 FY27 guide: $91B revenue — implies annualised run-rate of ~$360B+ before the second half ramps. 24.3B diluted shares. Scenarios anchored to institutional price targets with scenario-consistent FCF modelling.
FY27 Rev est.
~$400B
5Y FCF CAGR
~32%
Fair value / share
$315–$352
vs current ~$231
+36–52%
Bull path (WF $315 / BofA $320 / Evercore $352): NVLink Fusion locks in hyperscaler custom silicon into NVIDIA rack architectures; sovereign AI >$45B; Groq LPU racks capture inference market. FY27 revenue ~$400B (Q1 = $81.6B, Q2 guided $91B, H2 ramps >$100B/quarter). FCF margin 55–60%; FY27 FCF ~$220B. NVLink Fusion networking moat sustains even as compute ASP normalises. Implies 28–31× CY27 P/E — justified by structural platform economics.
FY27 Rev est.
~$305B
Implied CY27 P/E
16–18×
Fair value / share
$184–$215
vs current ~$231
−8 to −20%
Bear path: Hyperscaler capex digestion H2 2026; inference workloads shift faster to custom ASICs; TSMC 2nm cost inflation compresses gross margins to 70–72% despite pass-through attempts. Rubin ramp delayed to H1 2027 — architecture cadence gap creates 2 quarters of revenue uncertainty. FY27 revenue ~$305B vs guided run-rate; FCF margin compresses to 45%. Implies 16–18× CY27 P/E — multiple de-rate from current 19× on deceleration evidence.
Position: At ~$231, downside to bear midpoint ($200) is ~13%; upside to consensus ($269) is ~16%; upside to bull midpoint ($333) is ~44%. The risk/reward is modestly positive at current price. The prior entry zone of $185–205 would have offered 2:1+ asymmetry; at $231 post-catalyst, the skew is more moderate but the catalyst overhang is gone. Accumulate at current price; size up on any weakness toward $210.
Catalyst timeline
▾
-
20 May 2026 ✓Q1 FY2027 earnings — resolved. $81,615M revenue (beat $78.8B consensus), non-GAAP EPS $1.87. Q2 FY27 guide: $91B (vs buy-side whisper ~$83–85B). Gross margin 75.0%. $80B new buyback. Stock fell ~3% on "sell the news" — thesis intact.
-
Jun 2026Vera Rubin trial production begins. Mass production blueprints finalised. Trial production at TSMC N3P node. CoWoS-L dual-die chiplet architecture locks in yield parameters before full ramp. Key watch: SK Hynix HBM4 qualification at 11 Gb/s tier — any slip delays volume.
-
Jul 2026Initial Vera Rubin commercial deliveries to top-5 CSPs. First NVL72 Rubin racks delivered to MSFT, GOOGL, AMZN, META, ORCL. ODMs (Foxconn, Quanta, Wistron) commence full-scale production of rack systems. This is the product cycle that underpins FY27 consensus $380–420B revenue. CPO NVLink (scale-up photonics) shipping 2026 — COHR, LITE, MRVL supply-chain read-throughs.
-
~Aug 2026Q2 FY2027 earnings — the $91B validation test. NVDA guided $91B (±2%) revenue, 75.0% GM. Consensus will reset significantly above prior $83–85B whisper. Watch for: Rubin revenue contribution, networking progression beyond $14.8B, ACIE sub-segment (enterprise/sovereign) growth, and whether gross margin holds above 74% despite TSMC pricing headwinds.
-
H2 2026China regulatory update + SK Hynix HBM4E samples. Q1 and Q2 assume zero China DC compute. Any US export control softening restores an estimated $12–15B+ annual revenue opportunity. SK Hynix expects HBM4E samples in H2 2026 — successful qualification would reduce Rubin memory supply concentration risk (currently Samsung-dominated).
-
2027 H1Feynman (NVL1152) architecture reveal. SemiAnalysis noted "early hints" at GTC 2026. Official reveal + supply chain confirmation would extend the competitive moat timeline 18–24 months and re-anchor FY2028 consensus. SK Hynix HBM4E mass production targeted for 2027 — Feynman platform likely relies on HBM4E.
Named analyst commentary
▾
"Heavy is the head that wears the crown. The catalysts on the horizon are imperfect, whether it's custom silicon gaining traction as inference scales or the attach rate on CPUs-to-GPUs shifting as agentic AI requires more orchestration. Nvidia delivered a near perfect quarter — but I am turning my focus to other opportunities across the AI trade. The I/O Fund is a top tier team in AI research. We can do better than hold the most well-known name in the AI trade."
— Beth Kindig / IO Fund, io-fund.com (NVDA FQ1 FY27 post-earnings analysis), 22 May 20266
"Every two years NVIDIA is tripling or quadrupling the performance while only 2X-ing or 50% increasing the price. That's ultimately what makes NVIDIA's position so hard to challenge. Even if a competitor shows up with better technology, if they can't secure steady supply, they can't gain market share."
— Dylan Patel / SemiAnalysis (@SemiAnalysis), Dwarkesh Patel Podcast, 2026-03-13
"$NVDA actively pushing to 800V [DC power infrastructure], and today was a pretty big signal. On 22 Apr 2026, NVIDIA has initiated discussions with major South Korean power equipment companies to explore designing DC infra based on an approximately 800V direct current system."
— Serenity (@aleabitoreddit), X post, 2026-04-22 · Supply chain read: NVDA locking in power infrastructure supply chain ahead of Rubin Ultra rack deployments
Wells Fargo PT: $315 · Bank of America PT: $320 · Evercore ISI PT: $352 · Morningstar fair value: $260 · Wall Street average target: $278.73
— Institutional price targets post-Q1 FY27 earnings, May 2026 · Morningstar FV: $260 (22–24× CY27); bull-side WF/BofA/Evercore at 28–31× CY27 on NVLink Fusion + sovereign AI + Groq LPU thesis
Note: Dylan Patel's quote is from a paid newsletter/podcast; quoted from free-tier transcript.
Sources
1 NVIDIA 8-K Q1 FY2027 earnings press release — nvidianews.nvidia.com and SEC EDGAR, filed 2026-05-20; NVIDIA Q1 FY27 Quarterly Presentation (official), 2026-05-20
2 NVIDIA 8-K Q4 FY2026 earnings press release, filed 2026-02-25
3 Morningstar — NVDA fair value estimate $260, accessed 2026-05-21 (global.morningstar.com)
4 SemiAnalysis free-tier — GTC 2026 recap "The Inference Kingdom Expands" by Dylan Patel et al., 2026-03-24; Dwarkesh Patel podcast transcript, 2026-03-13
5 GuruFocus — Form 4 insider transactions, data through ~Apr 2026 45-day lag on 13F data
6 IO Fund / Beth Kindig — io-fund.com "Perfect Quarter, Imperfect Catalysts" (NVDA FQ1 FY27 post-earnings analysis), paywalled; published 2026-05-22; captured 2026-05-23
7 Serenity (@aleabitoreddit) — X post 2026-04-22; supply chain read on 800V DC power infrastructure
8 Gemini Deep Research synthesis — "Accelerated Computing Hegemony: An In-Depth Financial and Strategic Audit of NVIDIA's Q1 FY27 Performance", 2026-05-21. Aggregates: TrendForce ASIC/packaging reports; Goldman Sachs DRAM supply-demand forecast; TSMC pricing data; Samsung/SK Hynix HBM4 shipment reports; Marvell NVLink Fusion press release; Aragon Research Groq acquisition analysis; TradingKey Vera Rubin production confirmation; Finbox ROIC data; institutional price targets (WF $315, BofA $320, Evercore ISI $352). Full source list in research/nvda/2026-05-21-gemini-deep-research-q1-fy27-earnings-valuation.md.
⚠ = figure not verifiable from a second open-access source or sourced from a single provider.
⚠ FY27 consensus revenue (~$370–400B) and CY27 EPS (implied ~$12) are aggregator-derived. The $91B Q2 FY27 guide is the highest-confidence forward data point — all other FY27 estimates are extrapolations.
⚠ Rubin memory BOM ($2M/rack, 435% uplift) from Morgan Stanley estimate via IO Fund research — unverified from a second open-access source.
⚠ TTM FCF margin (~57%) estimated from Q1 FY27 quarterly FCF ($48.6B) as full trailing four quarters not available in captures. Q1 FY27 quarterly margin is 59.5%.
⚠ 3Y avg P/E (~45×) from Macrotrends.net — includes pre-AI-supercycle era; use directionally only.
1 NVIDIA 8-K Q1 FY2027 earnings press release — nvidianews.nvidia.com and SEC EDGAR, filed 2026-05-20; NVIDIA Q1 FY27 Quarterly Presentation (official), 2026-05-20
2 NVIDIA 8-K Q4 FY2026 earnings press release, filed 2026-02-25
3 Morningstar — NVDA fair value estimate $260, accessed 2026-05-21 (global.morningstar.com)
4 SemiAnalysis free-tier — GTC 2026 recap "The Inference Kingdom Expands" by Dylan Patel et al., 2026-03-24; Dwarkesh Patel podcast transcript, 2026-03-13
5 GuruFocus — Form 4 insider transactions, data through ~Apr 2026 45-day lag on 13F data
6 IO Fund / Beth Kindig — io-fund.com "Perfect Quarter, Imperfect Catalysts" (NVDA FQ1 FY27 post-earnings analysis), paywalled; published 2026-05-22; captured 2026-05-23
7 Serenity (@aleabitoreddit) — X post 2026-04-22; supply chain read on 800V DC power infrastructure
8 Gemini Deep Research synthesis — "Accelerated Computing Hegemony: An In-Depth Financial and Strategic Audit of NVIDIA's Q1 FY27 Performance", 2026-05-21. Aggregates: TrendForce ASIC/packaging reports; Goldman Sachs DRAM supply-demand forecast; TSMC pricing data; Samsung/SK Hynix HBM4 shipment reports; Marvell NVLink Fusion press release; Aragon Research Groq acquisition analysis; TradingKey Vera Rubin production confirmation; Finbox ROIC data; institutional price targets (WF $315, BofA $320, Evercore ISI $352). Full source list in research/nvda/2026-05-21-gemini-deep-research-q1-fy27-earnings-valuation.md.
⚠ = figure not verifiable from a second open-access source or sourced from a single provider.
⚠ FY27 consensus revenue (~$370–400B) and CY27 EPS (implied ~$12) are aggregator-derived. The $91B Q2 FY27 guide is the highest-confidence forward data point — all other FY27 estimates are extrapolations.
⚠ Rubin memory BOM ($2M/rack, 435% uplift) from Morgan Stanley estimate via IO Fund research — unverified from a second open-access source.
⚠ TTM FCF margin (~57%) estimated from Q1 FY27 quarterly FCF ($48.6B) as full trailing four quarters not available in captures. Q1 FY27 quarterly margin is 59.5%.
⚠ 3Y avg P/E (~45×) from Macrotrends.net — includes pre-AI-supercycle era; use directionally only.