ExploreNVDA

NVDA NVIDIA

AI Infrastructure · Accelerators / AI compute

$234.85Fri, Oct 2, 2026, 8:06 AM EDTBefore market

Price

Daily adjusted close · last point is delayed last sale · drag the brush to zoom

1D+1.7%
1W+4.3%
1M+4.7%
3M+20.5%
6M+32.5%
YTD+26.1%
1Y+24.5%
2Y+97.9%
Statusinvestigate
Researchprimary verified
Market cap$5.57T
Updated9/5/2026
Revenue (qtr)$96.22B
FCF (qtr)$21.40B
Capex (qtr)$2.68B
Net debt$10.92B

Summary

Research status: Primary-source verified through Q2 FY2027 results on 26 August 2026. Current market reference: $230.36 close on 4 September 2026.

NVIDIA decisively passed the pre-earnings evidence gate. Q2 revenue was $96.2B, +18% QoQ and +106% YoY, while Data Center revenue reached $89.0B, +18% QoQ and +117% YoY. GAAP and non-GAAP gross margins were both 75.0%, and non-GAAP diluted EPS was $2.22. Management guided Q3 revenue to $108.0B +/-2% with approximately 74.0% gross margin, and again assumed no Data Center compute revenue from China.

The operating evidence therefore strengthened materially: demand remains broad enough to support very high sequential growth even at NVIDIA's scale, platform economics remain exceptional, and the Vera Rubin transition is moving into production rather than causing an obvious air pocket. The main underwriting risks are no longer near-term demand confirmation; they are valuation, hyperscaler/customer concentration, custom-ASIC substitution, supply/commitment intensity, and the increasingly important ecosystem-financing/credit-support layer around AI infrastructure.

Current conclusion: keep NVDA at investigate/watch rather than automatically entering after a strong print. The earnings gate passed, but the portfolio already has meaningful AI-infrastructure factor exposure. Any starter should be based on a fresh post-Q2 valuation and opportunity ranking against AVGO/CRDO/CLS/VRT rather than on earnings quality alone.

Primary source verified through 5 September 2026:

Market reference: $230.36 close on 4 September 2026 from PowerFund market data.

PowerFund valuation, portfolio-ranking and entry conclusions are internal investment judgments.

Thesis

NVIDIA remains the highest-quality full-stack AI compute platform in the universe. Q2 FY2027 reinforced the core moat: revenue and Data Center both grew 18% sequentially, gross margin held at 75%, and Q3 guidance implies another large step-up despite no China Data Center compute revenue. The thesis rests on CUDA/software, accelerated-compute silicon, networking/interconnect and rack-scale systems combining into a platform whose economics are difficult to replicate. The investment case is now primarily about entry valuation, customer/custom-ASIC concentration and the quality of ecosystem-financed demand—not whether current AI demand is real.

Catalysts

Q3 revenue near the $108B guide; continued Data Center double-digit sequential growth; Vera Rubin production/ramp; networking and interconnect attachment; inference/agentic workload growth; sovereign/enterprise diversification; any compliant China upside; disciplined capital returns.

Risks

Customer concentration; hyperscaler custom ASIC substitution; AMD/merchant competition; AI-capex cyclicality; ecosystem financing/credit-support circularity; large inventory and manufacturing commitments; gross-margin normalization; export controls; architecture/yield execution; power/grid bottlenecks; valuation at very large market capitalization.

Invalidation

WARNING / FREEZE ADDITIONS — investigate before committing more capital if any of the following occurs:

  • Data Center sequential growth falls below ~5% outside a clearly identified architecture transition, or forward revenue guidance misses then-current expectations by >5% without a temporary supply explanation.
  • Non-GAAP gross margin falls below ~73% or guidance points below ~72.5% while revenue growth remains strong, suggesting structural pricing/mix pressure.
  • Inventory rises >25% sequentially while revenue guidance decelerates materially, or inventory and supply commitments grow materially faster than the next two quarters of expected revenue without matching customer-demand visibility.
  • Two major hyperscalers/model builders cut AI infrastructure capex or accelerator deployment plans by >10–15%, or utilization/ROI commentary weakens across multiple customers.
  • NVIDIA's own guarantees, investments, prepayments or other contingent support expand materially faster than free cash flow and begin to obscure independent end demand.
  • Forward valuation expands above ~30x FY28 adjusted EPS while estimate revisions flatten or turn negative.

REDUCE if:

  • Data Center revenue declines sequentially for two quarters outside a deliberate transition, or one quarter declines >10% and the following guide remains weak.
  • Non-GAAP gross margin remains below ~70% for two quarters because of competition, product economics or customer bargaining rather than a temporary transition.
  • A major architecture delay or quality/yield issue causes a material revenue-guide cut or customer deferrals.
  • Multiple leading customers shift a meaningful share of new workloads to internal ASICs/other accelerators and NVIDIA networking/software attach fails to offset the compute share loss.
  • Revenue/EPS expectations reset downward by >15% and the decline is demand-driven rather than timing/supply.
  • Ecosystem-financing losses, guarantees or customer credit exposures become material relative to annual free cash flow or are required to sustain order growth.

EXIT / THESIS INVALIDATED if:

  • CUDA/full-stack platform advantage demonstrably erodes for several quarters: meaningful developer/workload migration, falling accelerator share and weaker networking/software attachment occur together.
  • Data Center growth remains below ~10% YoY for two quarters with evidence of contracting customer deployment, not merely a difficult comparison.
  • Normalized gross margin structurally settles below ~65% because custom ASICs/competition commoditize accelerated compute.
  • Customer-owned accelerators reduce NVIDIA's normalized earnings power by >25% without offsetting software/networking economics.
  • Contingent financing or counterparty losses reveal that a material portion of reported demand was economically circular or non-viable.
  • Export controls or geopolitical restrictions permanently remove enough addressable market/supply access to impair the long-run platform economics.

Important change from v1: remove the static 'inventory >100 days' kill test. With rack-scale product transitions and large pre-build commitments, a fixed days threshold can false-trigger. The correct test is inventory/commitment growth relative to revenue, forward guidance, architecture transitions and independently verified end demand.

Competitive notes

NVIDIA's competitive advantage should be evaluated at the AI-factory platform level, not chip benchmark alone.

Versus AMD: AMD is the closest merchant accelerator alternative and can win where customers value second sourcing, open software or lower acquisition cost. NVIDIA's defense is broader: CUDA/library depth, developer familiarity, NVLink/NVSwitch, networking, rack integration and faster deployment. The risk rises if AMD closes software gaps enough that customers can arbitrage GPU suppliers with limited switching cost.

Versus hyperscaler ASICs: this is the more serious structural threat. Google, Amazon and other hyperscalers can optimize silicon for large captive workloads and may accept lower flexibility in exchange for better economics. NVIDIA must therefore keep improving total cost per token, time-to-deployment and developer productivity—not simply FLOPS. Networking and software attachment are key evidence of platform durability.

Versus Broadcom/Marvell/custom-silicon ecosystems: these companies enable the ASIC alternative rather than directly replacing NVIDIA as a full platform. Their growth is a useful leading indicator of hyperscaler willingness to internalize compute and networking.

Networking/interconnect: NVIDIA's Mellanox-derived InfiniBand, Spectrum-X Ethernet and NVLink/NVSwitch make compute harder to disaggregate. If Ethernet/open standards commoditize networking or third-party fabrics achieve equivalent cluster economics, NVIDIA's content-per-AI-factory advantage narrows.

Portfolio interaction: NVDA would add direct accelerator/platform exposure but little thematic diversification to VRT/NVT/CLS/NBIS. It is nevertheless different economically: those holdings monetize power, cooling, manufacturing and cloud capacity, while NVDA captures the core accelerated-compute/software economics. A measured NVDA position could improve quality of the AI sleeve but increases exposure to the same hyperscaler-capex factor.

Next diligence

  1. Rebuild the 24/60-month scenario valuation using Q2 actuals, the $108B Q3 guide and a consistent FY27/FY28/FY29 EPS definition.
  2. Map direct/indirect hyperscaler concentration and internal-ASIC programs; compare NVIDIA content growth with custom-silicon growth.
  3. Track inventory, purchase commitments, prepayments, guarantees and investments against forward revenue/FCF to test demand quality.
  4. Track Vera Rubin production/ramp and any Blackwell-to-Rubin transition pause.
  5. Measure networking/interconnect content per rack and whether platform attachment offsets compute-share pressure.
  6. Re-rank NVDA against AVGO, CRDO, CLS and VRT before any starter; the earnings gate passing is necessary but not sufficient for entry.
  7. Re-underwrite after Q3 FY2027 results or earlier on a material hyperscaler capex, export-control or financing-structure change.

Reviews

Completed catalysts for this name. The full archive is on the Calendar past list.

  • Assess late-September PCE/GDP and deployment conditionsPrevious belief → new evidence → updated belief. Previous belief: after the 16 Sep FOMC, PowerFund was SLOW / SELECTIVE because the policy-rate path and real-yield hurdle had worsened, so correlated AI-infrastructure deployment required both fresh company evidence and greater valuation margin. New evidence: BEA's 30 Sep releases show August PCE prices +0.3% m/m and +3.4% y/y, core PCE +0.2% m/m and +3.0% y/y, while real PCE rose 0.6% m/m. Q2 real GDP was revised up to 2.2% annualized from 1.5%, with real final sales to private domestic purchasers +4.6%; consumer spending and investment were important contributors. Inflation is therefore improving at the margin, especially core monthly PCE, but demand/growth remain too firm to treat this as a clean rate-relief signal. Updated belief: MAINTAIN SLOW / SELECTIVE deployment rather than accelerate correlated AI-capex risk. The macro valuation hurdle has eased modestly but has not disappeared; price weakness alone remains insufficient for adds. Independent-factor opportunities remain eligible if their own evidence and valuation gates clear. No planned trade is created by this macro review. Sources: https://www.bea.gov/news/2026/personal-income-and-outlays-august-2026 ; https://bea.gov/news/2026/gdp-third-estimate-industries-corporate-profits-state-gdp-and-state-personal-income-2nd
  • Review September FOMC and deployment stanceCompleted 17 Sep 2026 after the Sep 15–16 FOMC. The Fed unanimously raised the target range 25 bp to 3.75–4.00%, its first hike since 2023. The September SEP shifted materially hawkish versus June: median appropriate fed funds is 4.1% at end-2026 (vs 3.8% in June), 4.1% at end-2027 (vs 3.6%), 3.9% in 2028 and 3.6% in 2029; the longer-run median rose to 3.2%. Sixteen of 18 submitted rate paths call for at least one further hike in 2026. The macro mix is stronger growth/labor plus stickier inflation: 2026 GDP median 2.3% vs 2.2% in June, unemployment 4.1% vs 4.3%, PCE inflation 3.7% vs 3.6%, core PCE 3.4% vs 3.3%; 17 of 18 participants see PCE inflation risks tilted upside and 15 of 18 see core-PCE risks tilted upside. Chair Warsh emphasized that underlying inflation trends have not meaningfully improved and declined to give forward guidance. Financial conditions remain restrictive: Sep 16 2-year Treasury yield rose ~7.5 bp to 4.738%, 10-year reached ~5.00%; the latest 10-year TIPS real yield before the meeting was ~2.62% on Sep 15; the dollar index rose ~0.63%. Equities weakened but AI/tech showed some resilience: S&P 500 -0.44%, Nasdaq essentially flat and semiconductors +0.6%. The implementation framework continues ample reserves and full rollover/reinvestment, so the tightening impulse is principally the higher policy-rate path rather than renewed balance-sheet runoff. PowerFund conclusion: MAINTAIN A SLOW / SELECTIVE DEPLOYMENT STANCE; do not accelerate correlated AI Infrastructure deployment. The long-term AI-capex thesis remains intact—Fed itself describes capital investment as robust and the growth outlook strengthened—but the discount-rate/real-yield hurdle has worsened and the 10-year near 5% means valuation discipline must tighten. Existing thesis-intact holdings remain holds; NBIS remains strict no-add without contract/financing evidence; SNDK remains deferred. Do not average down solely because AI names have fallen. New AI adds require both fresh company-specific evidence and a re-ranked expected return that compensates for the higher cost of capital. VRT's drop into the ~$235–240 area is now economically interesting but should trigger a refreshed dossier/opportunity ranking rather than an automatic add, especially given UIG acquisition integration/capital-allocation risk and the recent sector-wide debate about AI buildout pacing. Independent Defence/Robotics/Energy opportunities remain eligible if they clear their own gates. The late-September PCE/GDP review remains the next scheduled macro deployment gate. No trade or planned action is created by this macro review alone.
  • Assess August inflation before September FOMCCompleted 15 Sep 2026 using the Aug PPI/CPI releases and post-CPI market reaction. Aug PPI rose 0.4% m/m and 5.4% y/y; final-demand goods rose 1.1%, services 0.1%, and final demand less food, energy and trade services rose 0.3% m/m. Aug CPI rose 0.4% m/m and 3.4% y/y; core CPI accelerated to 0.3% m/m while easing only to 2.4% y/y. Shelter rose 0.3% m/m, services less energy services rose 0.3% m/m, transportation services 0.5%, and gasoline 3.9%. The print therefore did not meet the pre-release benign threshold needed to clear the rate/valuation gate. Market confirmation strengthened that conclusion: September Fed hike odds rose from roughly 72% before CPI to near 90% after the release; the 10-year Treasury subsequently crossed 5% on 14 Sep, while the dollar strengthened and oil remained above $100/bbl. Equities initially rallied on 11 Sep as oil eased, showing no immediate fundamental break in AI demand, but the broader discount-rate backdrop worsened and AI/tech remained vulnerable to valuation and factor pressure. PowerFund conclusion: SLOW new correlated AI Infrastructure deployment into the FOMC rather than accelerate. Keep existing thesis-intact holdings; do not average down solely because prices have fallen; require company-specific evidence and valuation margin for adds. NBIS remains no-add without contract/financing evidence. With cash ~90% of NAV there is no need to force deployment ahead of the 16 Sep FOMC. This review does not change the long-term AI-capex thesis; it raises the near-term cost-of-capital hurdle. The September FOMC review is the next macro deployment gate. No trade is created by this review alone.
  • Assess August jobs report and AI valuation pressureCompleted 5 Sep 2026 using the 4 Sep August Employment Situation and market reaction. Payrolls rose 162k versus ~56k expected; June was revised +11k to +31k and July +44k from -23k to +21k, lifting the two-month total by 55k. Unemployment held at 4.1%; participation edged up to 61.6%; the workweek rose 0.1 hour to 34.4. Average hourly earnings rose 0.3% m/m and 3.1% y/y, the latter easing from July and limiting the wage-inflation signal. Markets nevertheless treated the report as reducing near-term rate relief: the 2-year Treasury yield rose roughly 4–5 bp to ~4.37–4.38% after briefly reaching ~4.42%; the 10-year finished near 4.78%; the dollar index rose ~0.2%; S&P 500 fell 0.38% and Nasdaq 0.29%. September Fed hike odds rose intraday into the ~60% area before easing somewhat. PowerFund conclusion: labor resilience is positive for end-demand and does not weaken the AI-capex thesis, but it raises the discount-rate hurdle and leaves CPI/PPI as the decisive near-term macro gate. Maintain, do not accelerate, AI Infrastructure deployment. Existing positions remain unchanged; do not chase rebounds; keep SNDK deferred and require company-specific valuation/fundamental evidence for any adds. The already-scheduled Aug CPI/PPI review remains the next macro decision gate.
  • Review AVGO Q3 FY26 earnings and AI read-throughAVGO Q3 FY2026 AI read-through completed 3 Sep 2026. Broadcom reported record Q3 revenue of $29.6B (+86% YoY), semiconductor revenue of $20.8B (+127%), and AI semiconductor revenue of $16.7B (+221% YoY, +54% QoQ). Management guided Q4 AI revenue to ~$21.7B (+236% YoY), FY2026 AI revenue to ~$58B, and said supply has been secured to support approximately $115B of FY2027 AI revenue, with continued growth expected into FY2028. XPU shipments were >3.5x YoY; AI networking revenue was >2.5x YoY, and Broadcom said AI networking should grow roughly as fast as XPUs over the next few years. Read-through: hyperscaler/frontier-model AI capex and networking demand remain exceptionally strong, so the current PowerFund AI-infrastructure drawdown is not supported by evidence of an AI-capex demand collapse. This is supportive for CRDO's high-speed connectivity thesis, CLS's AI networking/compute program demand, and VRT/NVT power/cooling demand. It also confirms that custom accelerators are scaling alongside GPUs, which is positive for total infrastructure spend but raises architecture/share-shift risk for NVDA and supplier-specific content risk. MRVL/ANET remain beneficiaries of the broad networking/custom-silicon build, though Broadcom's strength increases competitive pressure. Portfolio implication: classify the current AI-infrastructure weakness primarily as valuation/factor/rates pressure unless company-level evidence says otherwise; do not average down solely on price, but keep thesis-intact dislocations eligible for the opportunity-ranking process.
  • Assess AI/semi correction after Jackson HoleCompleted after Jackson Hole 2026 and the Aug. 28 market reaction. Chair Kevin Warsh said the 2% PCE target is firm and that the Fed must be confident underlying inflation is moving clearly and sufficiently quickly toward target or 'we have work to do.' He described labor markets as stable, output as solid, AI-related capex as a major driver of investment, and medium-term inflation expectations as broadly anchored, but emphasized that inflation remains too high and avoided dovish forward guidance. Markets interpreted the speech as hawkish: rate-hike expectations increased and U.S. equities finished lower. Conclusion for PowerFund: the AI/semiconductor fundamental demand thesis is strengthening, but the valuation/rates headwind has not cleared. Maintain—not accelerate—AI Infrastructure deployment. Keep existing CLS, VRT, NVT and NBIS positions at current sizing; no averaging down in NBIS without contract-level evidence. Favor selective entries/adds only when company-specific evidence is strong and valuation offers a margin of safety. CRDO remains gated to its own earnings; ANET/NVDA remain high-quality but valuation-sensitive. The current AI Infrastructure sleeve is only ~5.6% of NAV and cash is ~92%, so there is ample capacity to deploy later without forcing purchases into a still-hawkish rates backdrop.
  • Review Marvell AI interconnect read-throughCompleted after Marvell Q2 FY27 results and call. Marvell materially strengthened the AI-interconnect demand thesis: Q2 revenue was $2.739B (+37% YoY), Data Center revenue $2.172B (+46% YoY), FY2027 Data Center growth outlook rose to ~60%, and FY2028 Data Center revenue is expected to grow >60%. 800G remains strong while 1.6T ramps rapidly; 51.2T switching and scale-up/optical opportunities are expanding. Read-through: positive for CLS, ANET and CRDO demand and mixed-positive for NVDA because stronger custom-XPU activity validates both AI capex and alternatives to merchant GPUs. Do not chase MRVL on the beat; keep MRVL watch/investigate. Keep CRDO entry gated to its own next earnings because Marvell's expanding optical/switching stack raises the competitive bar despite a stronger TAM.
  • Review NVIDIA Q2 FY27 earnings and reassess AI deploymentNVIDIA Q2 FY27 materially supports the view that the recent AI/semi selloff has been driven primarily by factor/valuation compression rather than a break in AI infrastructure fundamentals. Q2 revenue was $96.2B (+106% YoY, +18% QoQ), Data Center revenue $89.0B (+117% YoY, +18% QoQ), and Q3 revenue guidance is $108B ±2% despite assuming no Data Center compute revenue from China. Vera Rubin is now in full production, with racks running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius; Spectrum-6 systems are arriving at gigascale AI factories. These points reinforce ongoing demand for compute, networking, power/cooling and data-center systems relevant to VRT, NVT, CLS, ANET, CRDO and NBIS. The principal caution is valuation/expectation risk: Q3 gross margin is guided to 74.0% ±50bp versus 75.0% in Q2, while investors remain sensitive to the sustainability and financing of AI capex. NVIDIA also announced financing partnerships intended to mobilize >$500B of third-party capital, which supports buildout capacity but increases the need to distinguish genuine end-demand from financing-supported expansion. Portfolio conclusion: lift the temporary blanket pause on correlated AI/semi deployment that was imposed pending NVIDIA, but do not mechanically average down. AI/semi names are again eligible for normal opportunity ranking and the pre-buy integrity gate; CRDO remains gated until after its 1 September earnings, NBIS remains not addable pending post-financing per-share analysis, and any SNDK/VRT/NVT/CLS action still requires current prices plus company-specific evidence. Existing holdings remain HOLD; no individual thesis invalidation or trim signal is triggered by NVIDIA's results.