NVDA's market cap has tripled in two years on FCF that grew from $3.8B to $96.7B. Most of the bear arguments are about the multiple, not the cash. Here's what the consensus is pricing wrong.
Nvidia closed FY2026 with $215.9 billion in revenue, up 65.5% year on year. Net income reached $120.1 billion. Free cash flow climbed to $96.7 billion against a base of $3.8 billion just three years earlier. The market cap of $5.26 trillion makes NVDA the largest single-stock weight in major US indices. Every conversation about Nvidia comes back to the same question: at this scale, can the growth continue? The bears have constructed a stack of arguments saying no. Each argument has a specific data response, and each data response has been hardening rather than softening over the past four quarters.
This is not a defence of paying any price. The Signals Desk is constructive on Nvidia at the current $217 print, target $295 over 18 months, with downside risk to $145 if hyperscaler capex meaningfully decelerates. The argument here is narrower: that six specific bear theses are not supported by the data the company is producing. We walk through each.
The six points below build cumulatively. The thesis is that FCF compounds at 25-35% for the next three fiscal years, the operating margin holds in the 60-65% zone, and the forward multiple stays in the 25-30x range as the cash conversion makes the absolute earnings power harder to dismiss. That is the case the consensus has not yet fully absorbed.
The most consistent bear thesis since late 2024 has been that hyperscaler capex (Microsoft, Meta, Google, Amazon) was set to plateau, then decline, by mid-2026. The actual data has gone the other way. Combined hyperscaler capex guidance for calendar 2026, based on the most recent earnings disclosures, sits at roughly $445 billion. That is up from $310 billion in 2025 and $210 billion in 2024. The capex acceleration has continued for three consecutive years and the FY2026 guidance points to another step-up.
More informative is the breakdown by hyperscaler. Microsoft has guided to $90-95 billion in calendar 2026 capex, Meta to $108-115 billion, Google to $100-105 billion, Amazon to $130-140 billion. Each of those figures is the highest the respective company has ever guided. The mix of that capex remains heavily weighted toward GPU compute. NVDA's data centre segment guidance for FY2027 implied roughly 60% growth at the midpoint, consistent with the hyperscaler capex pace.
The bear retort is that the capex pace must eventually compress because the unit economics of AI workloads cannot indefinitely justify the spend. The retort is correct in principle and wrong on timing. Inference workloads have grown faster than training workloads through FY2025-FY2026, which is the pattern we observe across hyperscaler disclosures. Inference workloads have a different unit economics profile, and the demand curve has not yet shown deceleration.
TickerXray Report
Get the complete Nvidia report with all 12 quantitative models, AI-generated investment thesis, and real-time data.
The second bear argument: the major hyperscalers are all developing internal custom silicon (Google TPU, Amazon Trainium and Inferentia, Microsoft Maia, Meta MTIA), and the share gain by these chips will erode Nvidia's pricing power. The qualitative story is correct. The quantitative impact is overstated.
Google has been running TPUs in production since 2017. The fact that NVDA's data-centre revenue continued compounding at 60-100% during a window when TPU was ramping is the strongest evidence that custom silicon and merchant silicon coexist rather than compete on a winner-take-all basis. The reality is that hyperscaler workloads vary significantly. Internal custom silicon makes sense for predictable, high-volume inference workloads. Merchant silicon (NVDA) wins for training, for variable inference, and for emerging workload types.
The Signals Desk modelled custom silicon share assuming the most aggressive bull case for hyperscaler internal chips. Even at 35% workload share by 2028, the residual addressable market for NVDA grows from approximately $215 billion in 2026 to north of $400 billion by 2028, simply because the total compute market is growing faster than the share gain by internal silicon. The growth rate decelerates from 60%+ to 25-30%, but the absolute scale continues to compound. Custom silicon is a deceleration story, not a contraction story.
Nvidia's software and services revenue, captured in the disclosed 'Software and Services' segment plus elements of the data-centre breakdown, ran at approximately $14 billion in FY2026, up from $8 billion in FY2025. The forward run-rate is on track to clear $20 billion by FY2027 and approach $35-40 billion by FY2029.
What matters about this revenue line is the gross margin. Software and services run at 85-90% gross margins versus the 75% gross margin on the hardware mix. As the software mix grows from roughly 7% of revenue today to a modelled 15-18% by FY2029, the consolidated gross margin trajectory has structural support to remain in the high 70s rather than mean-reverting to the 60-65% historical hardware-only profile. That gross margin stability matters because it is the engine of FCF compounding.
The components driving the software line are diversified. CUDA enterprise licenses, Omniverse, NIM microservices, NVIDIA AI Enterprise subscriptions, and the developer-platform monetisation (NGC, NIM, etc.) each scale with developer adoption rather than with hardware sales. The bear argument that NVDA is just a hardware company has been progressively invalidated by each quarterly print. The software line is now meaningful, and the take-rate is climbing.
The fourth point requires unpacking. Sovereign AI, the term Nvidia uses for AI infrastructure built by national governments rather than commercial hyperscalers, has emerged as a structural demand tail that the consensus model is not fully pricing. France, the UAE, India, Singapore, Saudi Arabia, the UK, Japan, and most recently Korea have each announced multi-billion-dollar sovereign AI compute programmes through 2025-2026. The cumulative announced spend is approximately $135 billion, with the actual procurement landing across FY2026-FY2028.
This demand has different characteristics than the hyperscaler demand. Sovereign buyers prioritise full-stack solutions (compute, networking, software, services) rather than discrete components. They tend to make larger initial commitments with longer build-out cycles. Pricing pressure is lower because the buyer is not optimising for the same unit economics as a commercial hyperscaler. The Signals Desk estimate is that sovereign AI demand contributes 8-10% of NVDA revenue in FY2027 and 12-15% by FY2029.
What makes this tail different from the hyperscaler line is that it is not yet visible to most public-equity analysts because the procurement cycles are longer and the disclosed figures are noisy. The procurement pipeline disclosures Nvidia provided on the most recent earnings call confirm the trajectory. Each new sovereign announcement (Germany, France, Korea over the past six months) adds 2-3 quarters of additional revenue visibility to the FY2027-FY2028 model.
The Spectrum-X and InfiniBand networking businesses, which Nvidia disclosed at greater detail starting in FY2025, ran at approximately $14 billion of revenue in FY2026, up from approximately $8 billion the prior year. That growth rate of 75% is faster than the consolidated data-centre line. Networking economics are different from GPU economics, with longer software-defined moats, stickier customer contracts, and gross margins in the 70-72% zone.
The strategic logic is that AI clusters at scale are increasingly bottlenecked by networking rather than by raw compute. The shift from training-only clusters (which can tolerate slower interconnect) to large-scale inference and training-blended clusters (which cannot) drives a structural shift in spend toward networking. The Signals Desk model points to networking revenue clearing $30 billion by FY2028 and approaching $50 billion by FY2030.
This line is the cleanest answer to the bear concern that compute pricing might deteriorate. Networking is harder to commoditise because the software stack (NCCL, DOCA, BlueField DPU integration) compounds the moat rather than eroding it. Most bear models do not have a separate line for networking; they assume it follows the GPU economics. That is one of the modelling errors that creates the gap between bear targets and the actual operational data.
Six. The forward PE of 26.6x sounds like a high multiple in absolute terms. It is not, given the cash flow trajectory. Run the comparison against the seven other large-cap technology franchises with comparable growth profiles. Microsoft trades at 22x forward, growing low-to-mid teens. Apple at 31x, growing high single digits. Meta at 22x, growing 20%. Amazon at 32x, growing 13%. Alphabet at 20x, growing 13%. Broadcom at 35x, growing 17%.
Nvidia's PEG ratio, computed as forward PE divided by FCF growth rate, sits at roughly 0.85x against the modelled 30%+ FCF growth through FY2028. Microsoft's PEG sits at 1.6x. Apple's PEG is 3.3x. Broadcom's is 2.0x. Among large-cap tech, Nvidia is the cheapest growth-adjusted franchise in the cluster.
The bear retort: at $5.2 trillion, the law of large numbers eventually breaks growth. That is mathematically inevitable but timing-dependent. Nvidia at $5 trillion compounding FCF at 25%+ for three more years adds another $5 trillion of cash flow value. Beyond FY2029, the deceleration to 12-15% growth is reasonable, and the multiple compresses to 18-20x at that point. The fair value path therefore lands at $295-$340 within 18 months and $400-$480 by FY2028.
This is the cumulative thesis. Bear arguments individually are reasonable. Bear arguments cumulatively are weaker than each individual point because each has been progressively invalidated by quarterly data. We are constructive at $217, target $295 within 18 months, downside risk to $145 only on a major hyperscaler capex compression that the data does not currently show.
Nvidia at $5.2 trillion is the highest-stakes consensus position in global equity markets. The bears have been right about the multiple risk and wrong about the operational trajectory for three consecutive fiscal years. Each successive quarterly print has confirmed that the FCF compounding is accelerating, not decelerating. The six points above are why we expect the same pattern to hold for at least another four to six quarters.
The risks are concentrated, not diffuse. The single biggest tail risk is a coordinated hyperscaler capex compression that is not in the current guidance set. The probability is non-zero (roughly 20-25%) but not the central case. The second-largest risk is a regulatory action that meaningfully restricts NVDA's ability to sell into China, which represents approximately 12-15% of revenue. The third is a major supply chain disruption at TSMC that delays production volumes for two or more quarters.
Our target of $295 within 18 months reflects a 26-28x forward multiple on FY2028 EPS of $11. The bull case to $400+ requires sovereign AI and networking each running ahead of our central forecast. The bear case to $145 requires the hyperscaler capex line to compress to flat-to-down by mid-FY2027.
The trade is to own the franchise here. Trim modestly above $260, add aggressively on weakness toward $175. The compounding has not finished. The cash line is the moat. The bears keep being wrong because the data keeps confirming the bull setup. Stay constructive.
A final historical anchor on multiple. The previous large-cap technology franchise that compounded FCF at this scale was Microsoft from 2014 to 2020, where Azure-driven cash compounding lifted Microsoft's market cap from $300 billion to $1.4 trillion. The forward multiple expanded from 14x to 30x across that span, and the absolute earnings power grew 3.5x over the period. Nvidia's setup is structurally similar but on a larger scale and faster timeline. The 26.6x forward multiple Nvidia trades at today is comparable to where Microsoft sat in mid-2018, three years into the cycle and four years before the multiple peak.
That is the historical pattern the Signals Desk anchors the bull case on. Multi-year FCF compounding cycles in dominant tech franchises tend to last 6-8 years from inflection to peak. Nvidia is roughly three years in. The remaining runway is the trade.
For portfolio managers running benchmark-aware mandates, an underweight on NVDA at this market cap is a different kind of bet than an underweight on a normal stock. The position size means relative-performance risk is asymmetric. Underweighting and being wrong costs more than overweighting and being right. The trade implication is that even sceptical investors should hold close to benchmark weight unless they have a high-conviction, data-anchored view that the FCF compounding is breaking. We do not see that signal in the current operational data. Position accordingly.
Full forensic analysis of Nvidia
+ 6 more models included
150,000+ stocks covered
Global coverage across 60+ exchanges. Every report includes all 12 quantitative models and AI analysis.
View plansEvery report runs 12 quantitative models and generates an AI investment thesis. From Piotroski scores to manipulation detection -- get the full picture in seconds.
12 forensic models
Piotroski, Altman, Beneish, DuPont & more
AI investment thesis
Synthesized outlook on every stock
Manipulation detection
Spot red flags before they hit the news
150,000+ tickers
Global coverage across 60+ exchanges
Expected return
Forward return projections for every stock
Real-time data
Live prices, insider trades, news sentiment
Free accounts get 1 report per month. Pro gets unlimited.
Our earlier 'Five Trillion' narrative argued Nvidia was being underpriced for the AI capex cycle. Since publication, free cash flow has reached $96.7 billion, the multiple has held, and Wall Street's biggest AI bull just got louder. We update the thesis.
A $5.2 trillion market cap with revenue still compounding above 70% year over year is a profile the public market has never seen at this scale. The signals are telling a specific story.
The combined hyperscaler capex disclosures from the Q4 2025 earnings cycle landed at $385 billion for 2026, well above the consensus model. Nvidia's $216 billion FY2026 revenue print is now the floor rather than the ceiling, and the institutional positioning data shows the bid building rather than fading.