NVDA — Earnings Review Q4 FY26 (Quarter Ending January 2026)

Date: 2026-03-01 Type: Earnings Review Quarter: Q4 FY26

Prior Beliefs / Updated Beliefs

In my Q3 FY26 review, I wrote: "I'd want to see Q4 FY26 results confirm Q1 FY27 guide of ~$75B+. If that materializes, the FY27 growth trajectory is real." That was the test. Let me grade it.

Bottom line: Every dimension exceeded my prior beliefs, including the specific $75B Q1 guide threshold I set as the condition for increased conviction. The growth re-accelerated — Q2 FY26 was +55.6%, Q3 +62.5%, Q4 +73.2% — and the Q1 FY27 guide implies further acceleration to ~+77% YoY. This is not a business that's decelerating. The numbers match the theory, and then some.

Multi-Quarter Financial Table

| | Q1 FY26 | Q2 FY26 | Q3 FY26 | Q4 FY26 | Q1 FY27E | | | Apr-25 | Jul-25 | Oct-25 | Jan-26 | Apr-26 | |---|---|---|---|---|---| | Revenue (M)|44, 062|46, 743|57, 006|68, 127| 78, 000||YoYGrowth|69.2|QoQGrowth|+12.0|Non − GAAPGM|61.0|GAAPGM|60.5|GAAPOpMargin|49.1|Non − GAAPOpMargin|52.8|FCF(M) | 26,135 | 13,450 | 22,089 | 34,902 | — | | FCF Margin | 59.3% | 28.8% | 38.7% | 51.2% | — | | Data Center (M)|39, 112|41, 096|51, 215|62, 314|—||DCMix|88.8|Networking(M) | — | 7,252 | 8,187 | 10,980 | — | | GAAP EPS | — | $1.08 | $1.30 | $1.76 | — | | Non-GAAP EPS | — | — | $1.30 | $1.62 | — |

Q1 FY26 Non-GAAP GM of 61.0% reflects H20 charge ($4.5B). Not representative of normalized run rate.

FY26 Full Year: Revenue $215.9B (+65% YoY) | Non-GAAP GM 71.3% | FCF $96.6B | Non-GAAP EPS $4.77

Segment Analysis

Networking continues to be the sleeper story turned leading story. $11B in a single quarter, +263% YoY, +34% QoQ. I flagged this in Q3 as the data point that matters for the ASIC displacement argument. You can build a custom chip. You cannot easily replicate NVLink fabric at rack scale. The more networking grows as a share of Data Center — now ~17.6%, up from ~8.5% a year ago — the deeper NVDA's system-level moat becomes. The numbers keep validating this.

Data Center at 91.5% of revenue. The business is now almost entirely a data center infrastructure company. Gaming at $3.7B is a seasonal dip — supply constraints expected in Q1 FY27. Not a growth driver, but steady.

Pro Viz at $1.3B is a record quarter and worth noting — Blackwell architecture creating demand in visualization that wasn't there before. Small but growing fast.

Leading Indicators

Supply commitments almost doubling from $50.3B to $95.2B in a single quarter is the most important number in this entire report. This is committed capital from customers. It's not a forecast or a hope — it's money on the table. When I wrote in Q3 that "supply commitments at $50.3B represent locked purchase obligations from hyperscalers — real money on the table," the follow-through is even more emphatic. 95Bincommitmentsagainst 68B quarterly revenue suggests backlog extends well beyond the next few quarters.

Sovereign AI at >$30B annually, >3x YoY — this is important for the concentration risk argument. NVDA's Data Center revenue is diversifying away from pure hyperscaler dependency. Hyperscalers were "slightly over 50%" of DC revenue with growth led by non-hyperscaler customers. That's a structural improvement in the customer base.

Gross Margin Recovery — Complete

The gross margin story has a clean resolution:

The Blackwell transition margin hit was real but temporary, exactly as management said. Colette Kress's "mid-70s" guidance has been consistently accurate for three quarters. Q1 FY27 guide at 75.0% (now including SBC — an accounting change, not an economic one) confirms the structural margin profile.

The difference between 73% and 75% gross margins at $300B revenue is $6B in annual gross profit. It matters. The fact that margins are expanding, not compressing, during the fastest revenue ramp in semiconductor history is remarkable.

Non-GAAP Accounting Change — Worth Flagging

Starting Q1 FY27, NVIDIA includes SBC in non-GAAP results. This is a policy alignment with how many tech companies already report, and it's the right thing to do — SBC is real expense. But it breaks the historical non-GAAP series. SBC was $1.6B in Q4 FY26. The Q1 FY27 guide of $7.5B non-GAAP OpEx includes $1.9B of SBC.

For comparability: Q4 FY26 non-GAAP op margin was 67.7%. Under the new policy, it would have been roughly 65.3%. Not a material change to the investment thesis, but investors need to adjust their models. I'll track both series going forward.

Valuation Assessment

As of 2026-03-01:

• Market cap: ~$4,506B
• TTM Revenue (Q1-Q4 FY26): $215,938M = $216B
• TTM P/S: 4,506 / 216 = ~20.9x
• FY27 Revenue (Q1 guided at 78B, assume 85-90B for Q2-Q4 avg): ~$330-340B
• FY27 P/S: 4,506 / 335 = ~13.5x (midpoint estimate)
• P/E (annualizing Q4 Non-GAAP EPS 1.62): 6.48 annualized → PE ~28x
• FY27 Non-GAAP EPS (street ~$5.50-6.00): PE **~30-33x forward**

Context update from Q3: In my last review, I wrote that "at $4.5T market cap, NVDA is priced for continued dominance through the FY27-FY28 period." Q4 results and Q1 FY27 guidance are confirming that dominance is continuing. The math I ran in Q3 — 300Brevenue, 73110B net income, 40x forward PE = $4.4T — is now looking conservative. Revenue could reach $330-340B in FY27, gross margins are 75% not 73%, and the forward PE at current prices is ~30x.

In other words, at current levels, you're paying ~30x forward earnings for a business growing revenue 73% with improving margins. My revenue-PEG heuristic: 73% growth at ~21x P/S or ~30x PE is not unreasonable. It's not cheap in absolute terms — every business is a sell at some price — but the growth rate justifies the multiple in a way that few other companies at this market cap can claim.

The ASIC question update. Jensen's framing on the call — custom silicon as net positive for NVIDIA because ASICs handle inference while NVIDIA dominates training — is self-serving but not wrong. Anthropic is building on NVIDIA (1GW). OpenAI is building on NVIDIA (10GW). The hyperscalers are ordering $95B in supply commitments. Whatever ASIC projects exist in parallel, they are not displacing NVIDIA spend. Not yet. The 2027-2028 timeline for meaningful ASIC impact hasn't changed, but the evidence that hyperscalers are doubling down on NVIDIA in the meantime is stronger than ever.

I could be wrong about the durability. But the data this quarter makes the bear case harder, not easier.

Management Commentary Assessment

Jensen Huang: "The agentic AI inflection point has arrived." This is the most committal forward-looking statement I've seen from him. Previous quarters had "demand exceeds supply" — present tense, backward-looking. "Inflection point has arrived" is a claim about the future demand trajectory. He's telling you to expect inference demand to compound on top of training demand.

"Compute equals revenues." Jensen is simplifying the NVIDIA investment thesis to a one-line equation: every dollar of CapEx = every dollar of NVIDIA revenue. If hyperscaler aggregate CapEx reaches $600B in CY2026 (which the data supports), and NVIDIA captures 40-50% of the compute layer, that's $240-300B. The math works.

Colette Kress: Consistent and measured. "Mid-70s" gross margins — she's said this for three straight quarters and delivered each time. The SBC accounting change was disclosed proactively and transparently. No sandbagging, no surprises.

DeepSeek framing: Management presented Chinese AI efficiency models as demand-expanding via Jevons paradox — lower inference costs = more inference demand = more total compute. This is the right framework. Shows what I know, but the concern about Chinese efficiency models disrupting demand was backwards. Cheaper inference doesn't reduce GPU demand; it expands the addressable market.

Promises to track:

Capital Returns

• FY26 buybacks: $40.1B + dividends: 974M = **41.1B total**
• Q4 FY26: $3.8B buybacks + $243M dividends = $4.1B
• FY26 shares reduced: 24,706M → 24,430M = -276M shares (-1.1%)
• Remaining buyback authorization: $58.5B
• Cash & marketable securities: $62.6B

FCF of $96.6B in FY26 against $41.1B returned to shareholders = 43% payout ratio. The business is generating cash faster than it can responsibly return it. $62.6B cash pile + $58.5B buyback auth = massive optionality. At $4.5T market cap, even $40B/year buybacks only reduce share count ~1% annually — the sheer size of the company limits buyback impact. But it's a luxury problem.

Thesis Assessment

Thesis: NVDA is the dominant infrastructure provider for the AI era, with a software-like margin profile in a hardware business, protected by the CUDA ecosystem, NVLink interconnect fabric, and continuous product cycle (Hopper → Blackwell → Rubin).

Status: Intact / Strengthening

In Q3 I said I was watching four things:

1. GM pressure below 73% — Resolved. GM at 75.2%, guided to 75.0%. No pressure.
2. Hyperscaler CapEx guidance turning flat — Not happening. CSP CapEx at ~$600B CY2026, up $200B since year start. Supply commitments nearly doubled QoQ.
3. ASIC with actual capacity numbers — Still aspirational. No production data from Broadcom/OpenAI custom silicon or Meta MTIA T1. Hyperscalers are ordering more NVIDIA, not less.
4. Rubin execution — Samples shipped. Production H2 CY2026. On track.

All four watchpoints came in favorable. The Q4 results didn't just maintain the thesis — they strengthened it materially. Growth re-accelerated for the third consecutive quarter. Margins expanded to the highest level since the Hopper peak. FCF margin hit 51%. Q1 FY27 guidance at $78B implies the ramp has legs into FY27.

What I'm still watching:

1. FY27 GM sustainability — 75% is great but input costs are rising. If Rubin transition causes another margin hit like Blackwell did (Q1 FY26), that's the dip to watch for.
2. Hyperscaler CapEx CY2027 — The critical year. If spending flattens, NVIDIA's growth decelerates sharply. No signs yet, but it's the structural risk.
3. ASIC production evidence — Still 12-18 months out. When Broadcom or Google report actual ASIC revenue at scale, that's when the competitive threat becomes quantifiable.
4. Concentration risk — Despite diversification, top 4 hyperscalers probably account for 35-40% of total revenue. Any single customer pullback would be felt.

Would I own this? Here's where my Q3 conditionality resolves. I wrote: "I'd want to see Q4 FY26 results confirm Q1 FY27 guide of ~$75B+." The guide came in at $78B — $3B above my threshold. The condition was met. Not on hope, but on data.

At ~21x TTM P/S and ~30x forward PE, this isn't a bargain. But applying my revenue-PEG heuristic: 73% growth at 21x P/S is well below the threshold where growth exceeds the multiple. For comparison, many software companies I track trade at 15-25x P/S with 25-40% growth. NVDA's growth-to-multiple ratio is actually more favorable than most of them.

I'd take a position here — modest, because every business is a sell at some price and $4.5T is a lot of market cap to compound from. I'd size it at maybe 5-8% and build from there if Q1 FY27 results (May 2026) confirm the $78B guide. The time to be greedy was when NVDA was at $1T; at $4.5T, the opportunity is real but the upside is measured in "how long can they compound" rather than "how much can the multiple expand."

I don't invest based on hope. But $68B in revenue, $35B in quarterly FCF, $95B in committed supply obligations, and a $78B Q1 guide — that's not hope. That's a business delivering at a scale that has no precedent in the semiconductor industry, and probably in any industry.

Summary

Bear