01 · Equity deep-dive — the silicon desk
NVDA
$207.40 ▼ 12% off its 52-wk high
NASDAQ · SEMICONDUCTORSMKT CAP ≈ $5.0T52-WK $142.03 – $236.54AS OF JUNE 18, 2026

The whole world is wiring itself to run on AI. The only question left: does the current keep flowing through NVIDIA's silicon — or find another path?

Revenue just grew 85% to a record $81.6B and management guided next quarter to $91B — yet the stock sits 12% below its high, because the entire debate compresses to one question: is the AI build-out a durable multi-year shift NVIDIA's full-stack moat captures, or a capex super-cycle that custom silicon and digestion eventually route around? Five analyst lenses, three scenarios, four time horizons.

The verdict · TL;DR
One question decides the stock: does AI-compute demand keep flowing through NVIDIA, or route around it? Demand has gone "parabolic" — record $81.6B revenue, a $91B next-quarter guide, ~$1T Blackwell-plus-Rubin backlog, and a 75% gross margin throwing off ~$97B of free cash flow. The base case says NVIDIA stays the tollbooth on a multi-year compute super-cycle; the bear case — a hyperscaler capex air-pocket plus custom ASICs eating inference — is real, not imaginary. At ~$5T, the upside is large but the cushion is thin.
5-yr · prob-weighted
$413
+99% vs $207.40
52-week playback · where the tape sits ◷ Mid-range · room above
$207.40 · Jun 18, 2026 consensus $299 · +44%
$142.03 · 52-wk low · Jun ’25 $236.54 · 52-wk high
Price history + cone of outcomes · 2024 → 2031
HISTORICALBULLBASEBEARPROB-WTD
$700$560$420 $280$140$0 202420252026 202720282029 20302031 $142 · low Jun ’25 $237 · 52-wk high $413 $270$310$350 $680 $420 $130 TODAY · $207.40

Gray line = NVIDIA's actual split-adjusted price into today ($142 low Jun ’25 → $237 52-week high → $207.40 now); colored paths = synthesized scenario midpoints forward, probability-weighted (base 50% · bull 25% · bear 25%). Linear scale, mid-year marks. Wall Street 12-month consensus ≈ $299 (range $180–$500, “Strong Buy” from ~59 of 62 analysts). NVIDIA's fiscal year ends in late January, so a given calendar year leads the matching fiscal year by ~11 months.

Re-weight the scenarios

Those probabilities are a judgment call — so make them yours. Drag to set how likely the bear and bull cases are (base takes the remainder); the blended target below, the dotted line on the chart, and the prob-weighted row of the scenario cards all update live.

25% bear 50% base 25% bull
Blended 5-yr expected $413 +99% vs $207.40
+85%
Q1’27 Revenue ($81.6B)
+92%
Data Center Revenue (~$75B)
+109%
Non-GAAP EPS ($1.87)
~75%
Non-GAAP Gross Margin
$91B
Q2 FY27 Revenue Guide
~$97B
FY26 Free Cash Flow
~80%+
AI Accelerator Share
~$1T
Blackwell+Rubin Backlog ’27
02 · The panel — five ways to read the same tape

Five analyst lenses, five answers

The same fundamentals support wildly different conclusions depending on which framework you trust. Each lens below is a synthesized expert perspective with its own 12-month target.

Growth / Momentum PM

The Parabola

"Demand has gone parabolic," and NVIDIA is the only company selling the entire stack into a multi-decade compute super-cycle. Revenue +85% to $81.6B, a $91B next-quarter guide, ~$1T Blackwell+Rubin backlog through 2027, and hyperscaler capex running ~$600–690B (+36% YoY). The annual cadence — Blackwell → Rubin → Feynman — keeps rivals a generation behind. This earns a growth premium, not a fear discount.

12-MO TARGET $360 · ~30x fwd EPS
Value / FCF / Quality

The Cash Furnace

For a hyper-grower this is shockingly cheap on cash: ~$97B FY26 free cash flow, a net-cash balance sheet, 75% gross margins, 114% ROE, and $41B returned to holders last year. At ~24x forward earnings for 30%+ EPS growth, the PEG is below 1. As the working-capital and capex build of the Blackwell ramp normalizes, FCF conversion re-expands and the multiple looks conservative.

12-MO TARGET $300 · in line with consensus
Bear / Disruption Skeptic

The Air Pocket

This is a capex cycle, not an annuity. About 40% of revenue rests on four hyperscalers that are all shipping their own silicon (TPU, Trainium, Maia, MTIA). Inference — two-thirds of compute and the fastest-growing slice — is exactly where the CUDA moat is shallowest; some analysts model NVIDIA inference share falling from 90%+ toward 20–30% by 2028. China is zeroed out. One hyperscaler "we're pausing builds" reprices the whole complex.

12-MO TARGET $150 · multiple de-rates
Moat / Competitive Strategy

The CUDA Tollbooth

The moat isn't the GPU — it's CUDA (4M+ developers, 20 years) plus the full rack-scale stack (NVLink, InfiniBand, cuDNN, TensorRT) and a ~60% lock on TSMC's CoWoS packaging. Switching costs are measured in years, not dollars; NVIDIA is now also #1 in data-center Ethernet switching and "the only platform that runs every frontier model." Even a competitive scenario leaves a 65–70% share floor. The toll holds.

12-MO TARGET $285 · durable-share fair value
Quant / Macro Strategist

The $5 Trillion Gravity

The largest stock on earth (>7% of the S&P) faces the law of large numbers, a 2.2 beta, and index-crowding gravity. It sits ~12% off its all-time high after chopping sideways for months, even great prints draw "sell-the-news," and insiders have sold ~5.9M shares in 90 days. The forward multiple is mid-range versus its own history — but at this size, slope compresses even as earnings rise.

12-MO TARGET $240 · gravity caps the slope
03 · Wall Street's read

Wall Street 12-month price targets

What the sell-side expects over the next year. Bars are sorted low to high; the dashed line is today's $207.40 — almost every desk targets well above it, and the spread from $180 to $500 is the widest of any mega-cap.

Consensus ≈ $299 (+44%) · selected names, range $180–$500
BUYHOLDSELL
Street low · cautious $180 Goldman Sachs $250 Morgan Stanley $250 JPMorgan $265 Wells Fargo $265 Citi $270 UBS $275 Wedbush $275 BofA Securities $320 Evercore ISI $352 Street high $500 TODAY · $207.40

Sell-side 12-month targets — a selection of the ~62 firms covering NVIDIA; the full consensus is ≈ $299, about +44% above today, with a near-unanimous Strong-Buy skew (≈ 59 of 62 buy, 1 sell). The dashed line marks today's $207.40: even the cautious desks cluster above it, and the Street high reaches $500. The lone $180 "street low" is the visible bear — the gap between it and the $352 Evercore high is the entire debate priced into one chart. Firms, ratings, and targets illustrative.

04 · Price scenarios — 1 / 2 / 3 / 5 years

Where the current carries it

Synthesized scenario midpoints (mid-year). Returns shown vs. today's $207.40. These are illustrative frameworks, not predictions — five-year outcomes hinge on whether AI-compute demand keeps routing through NVIDIA's silicon or around it.

1 Year

Mid-2027
Bull$340+64%
Base$270+30%
Bear$165−20%
Prob-wtd$261+26%

2 Years

Mid-2028
Bull$430+107%
Base$310+49%
Bear$150−28%
Prob-wtd$300+45%

3 Years

Mid-2029
Bull$530+156%
Base$350+69%
Bear$140−32%
Prob-wtd$343+65%

5 Years

Mid-2031
Bull$680+228%
Base$420+103%
Bear$130−37%
Prob-wtd$413+99%
Bull case — show the assumptions & math
AI-compute demand stays "parabolic": NVIDIA holds 80%+ accelerator share through the Rubin and Feynman ramps, sovereign/enterprise AI more than offsets China, gross margins hold ~75%, and the annual cadence keeps rivals a node behind. EPS compounds toward ~$22 by FY32 on a still-premium multiple.
EPS ≈ $22 by FY32 × ~31× exit multiple → ≈ $680 · 5-yr price CAGR ≈ +27%/yr
Base case — show the assumptions & math
Compute demand keeps growing but share drifts toward a 70–75% floor as custom ASICs take inference; revenue growth decelerates from triple digits to a healthy 20s%, margins normalize modestly, and the multiple compresses as the law of large numbers bites at $5T+.
EPS ≈ $17.5 by FY31 × ~24× exit multiple → ≈ $420 · 5-yr price CAGR ≈ +15%/yr
Bear case — show the assumptions & math
A capex air-pocket hits: one or more hyperscalers pause builds, custom silicon takes inference share faster than expected, China stays zeroed, gross margins slip toward the high-60s, and the multiple de-rates hard as growth normalizes and "AI ROI" skepticism sets in.
EPS ≈ $9 with a de-rated ~14× multiple → ≈ $130 · 5-yr price CAGR ≈ −9%/yr
05 · Follow the cash

Revenue, capex, free cash flow & debt ($B)

Where the current actually goes. NVIDIA is the rare hyper-grower that throws off enormous cash while barely spending on capex — the foundries carry that burden. The bull and bear both live in the gap between these bars.

Annual revenue, capex, FCF & total debt · FY23 → FY26 (Jan year-end)
REVENUECAPEXFREE CASH FLOWTOTAL DEBT
$0$60$120$180$240 FY23FY24FY25FY26

NVIDIA's fabless engine in one view: revenue went from $27B to ~$216B in three years (+8×) while free cash flow scaled from ~$4B to ~$97B — the core of the "cash furnace" thesis. Capex (clay) stays tiny because TSMC owns the fabs; that's why FCF conversion is so high. Total debt (black) is trivial — under $9B against a net-cash position north of $50B — so the $41B of buybacks-plus-dividends is funded entirely by cash, not leverage. The bear's worry isn't the balance sheet; it's whether the revenue bar can keep climbing once hyperscalers digest. Figures illustrative; FY ends late January.

06 · Earnings power

EPS path underpinning the targets ($)

The price targets aren't pulled from the air — each is an EPS estimate times an exit multiple. Here's the earnings ladder the scenarios are built on.

Non-GAAP EPS · reported vs. estimated, FY24 → FY31E (split-adjusted)
REPORTEDESTIMATE
$0$4.50$9.00$13.50$18.00 FY24FY25FY26FY27EFY28EFY29EFY30EFY31E $1.30 $2.99 $4.93 $8.80 $11.50 $13.80 $15.40 $17.00

Non-GAAP EPS, split-adjusted for the 10-for-1 (2024). Gray = reported (FY24–FY26), olive = consensus estimates assuming growth decelerating from ~80% toward the low-teens by FY31. The base case's ~$17.5 of FY31 EPS at a ~24× exit multiple ≈ the $420 base-case 5-year target — this is the ladder underneath those prices. Note Q1 FY27 onward, NVIDIA's non-GAAP EPS definition now includes stock-based comp. Estimates illustrative.

07 · Growth scorecard

The business is still compounding

Q1 FY27, year-over-year — read these against a stock sitting only ~12% off its all-time high, not at a low.

Year-over-year growth by metric · Q1 FY27
COREFRONTIER
Gross profit +83% Revenue +85% Data center +92% Operating income +98% Non-GAAP EPS +109% Sovereign + enterprise +135% Networking +263%

Every line compounds — revenue +85%, data center +92%, operating income +98%, earnings +109%, with sovereign/enterprise and networking (now #1 in data-center Ethernet) sprinting off smaller bases (clay). Yet the stock has chopped sideways for months. That gap — business accelerating while the multiple compresses — is the bull's whole case. Frontier/select figures illustrative.

08 · The debate

Bull vs. Bear

The entire valuation argument compresses into one disagreement: does AI-compute demand keep flowing through NVIDIA's silicon, or route around it?

▲ THE BULL CASE

  • Demand has "gone parabolic." Q1 FY27 revenue +85% to a record $81.6B and a $91B next-quarter guide — that guide assumes zero China data-center revenue, so it's all the rest of the world.
  • Profit outgrows revenue. Operating income +98% and non-GAAP EPS +109%, with ~75% gross margins intact even through the Blackwell ramp.
  • A cash furnace. ~$97B FY26 free cash flow, a net-cash balance sheet north of $50B, and $41B returned to holders last year via buybacks and a 25×-raised dividend.
  • The CUDA moat compounds. 4M+ developers, 20 years of software lock-in, the full rack-scale stack (NVLink, InfiniBand), ~60% of TSMC's CoWoS capacity, and now #1 in data-center Ethernet — switching costs measured in years.
  • ~$1T backlog, annual cadence. Blackwell + Rubin orders booked through 2027; the Blackwell → Rubin → Feynman roadmap keeps rivals a generation behind.
  • Mid-range multiple. At ~24× forward earnings for 30%+ EPS growth, the PEG is below 1 — the fear of disruption is doing the de-rating, not the fundamentals.

▼ THE BEAR CASE

  • This is a capex cycle, not an annuity. Hyperscaler AI spend (~$600–690B in 2026) is the revenue line; one "we're pausing builds" headline reprices the whole complex overnight.
  • Customer concentration is a loaded gun. ~40% of revenue rests on four hyperscalers — Google, Amazon, Microsoft, Meta — that are all shipping their own silicon (TPU, Trainium, Maia, MTIA).
  • Inference is the soft underbelly. Two-thirds of compute and the fastest-growing slice is exactly where CUDA matters least; some analysts model NVIDIA inference share sliding from 90%+ toward 20–30% by 2028.
  • Custom ASICs are compounding faster. The custom-accelerator market is growing ~45% a year vs. ~16% for merchant GPUs; Broadcom alone sits on a $73B AI backlog.
  • China is structurally lost. Blackwell and Rubin remain banned; the June 2026 rule even closes the subsidiary-abroad loophole. A ~25%-of-data-center market is now zeroed.
  • $5 trillion fights gravity. The largest stock on earth (>7% of the S&P), a 2.2 beta, heavy insider selling, and "sell-the-news" reactions even to great prints — size compresses the slope.
09 · Risk map

Risk map — likelihood × impact

Where each risk sits, not just how big it is. The hot upper-right corner — likely and high-impact — is the one that matters; for NVIDIA the single hottest cell is a digestion air-pocket in AI capex.

Low impact
Medium impact
High impact
Likely
  • Gross-margin normalization
  • Custom-ASIC inference share loss
  • Multiple de-rate at $5T
  • AI-capex digestion air-pocket
Possible
  • AMD share gains
  • Rubin-ramp margin dip
  • Hyperscaler in-sourcing
  • China permanent loss
Tail
  • AI-ROI reckoning / "AI winter"
  • Taiwan / TSMC supply shock

AI-capex digestion air-pocket

Likely × High

Hyperscalers pause or stagger builds after a torrid spending burst; orders lump and a quarter or two of flat-to-down data-center revenue de-rates the whole stock.

Hyperscaler in-sourcing

Possible × High

Google, Amazon, Microsoft and Meta shift more training and inference to their own TPU/Trainium/Maia/MTIA silicon, eroding the ~40% of revenue they represent.

China permanent loss

Possible × High

Export controls stay or tighten; a market that was ~25% of data center stays at zero while Huawei and local champions entrench.

Custom-ASIC inference share loss

Likely × Medium

Inference — the majority of compute — migrates to cheaper purpose-built ASICs where CUDA's lock-in is weakest, capping NVIDIA's share of the fastest-growing slice.

Multiple de-rate at $5T

Likely × Medium

The law of large numbers and index-crowding gravity compress the forward multiple even as earnings keep rising.

AI-ROI reckoning / "AI winter"

Tail × High

Enterprises fail to monetize AI fast enough, the capex super-cycle stalls, and demand resets hard — low odds near-term, but it would reprice the entire complex.

Taiwan / TSMC supply shock

Tail × High

A geopolitical or natural disruption to TSMC's advanced packaging (CoWoS) chokes supply regardless of demand — an existential, low-probability tail.

AMD share gains

Possible × Medium

MI400/MI450 wins at OpenAI, Oracle and Meta chip away at the merchant-GPU lead at the margin.

Rubin-ramp margin dip

Possible × Medium

Each new architecture transition temporarily pressures gross margin before yields mature — a recurring, manageable wobble.

Gross-margin normalization

Likely × Low

Mix and competition pull ~75% gross margins toward the low-70s over time — a gentle headwind, not a break.

10 · Plain-language glossary

The jargon, decoded

Hover the dotted terms in the metrics, or scan the desk's working definitions here.

Data center revenue
NVIDIA's biggest segment (~92% of sales) — the AI GPUs, networking and systems sold to clouds, sovereigns and enterprises.
CUDA
NVIDIA's 20-year software platform that AI runs on. Millions of developers build on it, which is why switching to a rival chip is measured in years, not dollars — the real moat.
Hyperscaler
The giant cloud builders (Amazon, Microsoft, Google, Meta). They're NVIDIA's biggest customers — and, increasingly, its rivals via in-house chips.
Custom ASIC
A chip built for one job (e.g. Google's TPU). Cheaper at scale for inference, where it threatens NVIDIA's share most.
Inference vs. training
Training teaches a model; inference runs it. Inference is the larger, faster-growing workload — and the one most exposed to custom silicon.
Gross margin
Revenue minus the direct cost of making the product, as a %. NVIDIA's ~75% is extraordinary for hardware and central to the cash story.
Free cash flow
Cash left after running and investing in the business — the fuel for buybacks and dividends. ~$97B in FY26.
Forward P/E
Price divided by the next year's expected earnings. NVIDIA's ~24× is mid-range versus its own history despite 30%+ growth.
Exit multiple
The P/E assumed at the end of the forecast. Multiply it by projected EPS to get a target price.
Prob-weighted
Each scenario's price × its probability, summed into a single expected value across bear, base and bull.