Generate Your Investor Mosaic with Kilby
New /scuttlebutt command quickly generates investor mosaics
Point Kilby at your Company notes & files to generate your investor mosaic.
Kilby does not store your data. Our new “/scuttlebutt” command is a read-only function.
Here is a brief demo where we pointed Kilby at our NVDA folder which included files from NVDA’s most recent earnings call.
Here is Kilby’s output from pointing it at files from NVDA’s most recent EPS call:
Kilby Scuttlebutt Synthesis
## NVIDIA Q4 FY2026 — February 25, 2026
---
## Core Read
The central question embedded in this material is not whether NVIDIA can grow. It has grown data center revenue 13x in three years. The real question is whether the growth mechanism is durable or whether it is a synchronized CapEx wave that eventually breaks. The evidence from these documents — read together across the press release, transcript, CFO commentary, quarterly trend table, and balance sheet —points firmly toward durability, but with three specific risks the numbers themselves surface that management’s narrative consistently underweights.
Start with the financials as ground truth. Q4 revenue of $68.1B was up 73% year-over-year and added more sequential revenue in a single quarter — $11B in data center alone — than NVIDIA’s entire annual revenue was in many prior years. Non-GAAP gross margin expanded 160 basis points sequentially to 75.2% as Blackwell ramped and its cost structure improved. Free cash flow for the quarter was $34.9B; for the full year, $96.6B on $102.7B of operating cash flow, an 80% FCF conversion rate. Net cash on the balance sheet stands at $54.1B against $8.5B in total debt. These are not soft metrics dressed up to look good — they are verified numbers from primary sources, cross-checked across the press release, CFO commentary, and investor presentation.
The 8-quarter data center revenue trend tells the more important structural story. DC revenue decelerated in Q1–Q2 FY26 (Blackwell transition, H20 charges) before re-accelerating sharply: +24.6% QoQ in Q3 and +21.7% QoQ in Q4. The reacceleration is not explained by seasonality. It is Blackwell ramping through its cost curve while demand broadened beyond hyperscalers. Hyperscalers held at ~50% of data center revenue in Q4 while non-hyperscaler revenue — sovereigns, AI model makers, enterprises — grew faster. The revenue mix is diversifying. Networking, meanwhile, has grown from $3.0B in Q4 FY25 to $11.0B in Q4
FY26 (+263% YoY) and now represents 17.6% of data center revenue. Full-year FY26 networking was $31.4B, up 142% year-over-year. This is not a GPU company with a networking attachment anymore.
One balance sheet signal that was invisible in prior analysis: goodwill jumped $15.6B year-over-year (from $5.2B to $20.8B), and non-marketable equity securities surged from $3.4B to $22.3B. The cash flow statement shows $13B paid to Groq in Q4 investing activities, plus $17.5B in purchases of non-marketable equity securities for the full year (up from $1.5B in FY25). NVIDIA is deploying capital aggressively into ecosystem positions — Anthropic ($10B investment announced on the call), Groq, Intel, and others. The Groq payment alone in Q4 is larger than the company’s entire annual CapEx for property and equipment. This is not balance sheet management; this is ecosystem acquisition at scale, following the Mellanox playbook but with significantly more capital and ambition.
---
## What The Mosaic Says
- **The data center revenue trajectory has re-accelerated, not plateaued.** The 8-quarter trend shows a mid-cycle deceleration in Q1–Q2 FY26 (H20 write-downs, Blackwell ramp costs, China shutdown) followed by a sharp re-acceleration in Q3–Q4. Q1 FY27 is guided at $78B (+$10B sequential), continuing the pattern. This is not a company entering the back half of a cycle.
- **Networking is now a structurally attached business, not an accessory.** At $10.98B in Q4 (17.6% of data center) and $31.4B for the full FY26 year (+142% YoY), networking has become too large to model as an attachment rate. Every GB200/GB300 rack ships with NVLink switch infrastructure embedded. The revenue is not separable from compute at this scale. AWS is now integrating NVLink with its custom silicon — the moat is extending into even the most competitive custom-chip environments.
- **The Groq acquisition is the most underappreciated move in the quarter.** $13B in cash paid in Q4 alone, explicitly framed by Jensen as the Mellanox playbook: license the technology, absorb the team, extend CUDA architecture over it. Mellanox cost ~$7B in 2020 and is now $31B/year in revenue. If Groq’s low-latency inference technology follows even a partial version of that trajectory, the market has not priced it. GTC (March 16) is the next Groq disclosure point.
- **The balance sheet is being weaponized as an ecosystem tool.** Non-marketable equity securities went from $3.4B to $22.3B in a single year. Combined with the Groq payment and the Anthropic $10B investment, NVIDIA spent more on ecosystem positions in FY26 than it spent on its own physical infrastructure. The remaining $58.5B buyback authorization is real, but the capital allocation priority is clearly ecosystem lock-in first, returns to shareholders second.
- **Supply commitments of $95.2B extending into calendar 2027 are the most concrete demand signal in the pack.** This is a binding financial position, not guidance language. NVIDIA does not lock in $95B of supply commitments unless demand visibility is genuine. The simultaneous inventory buildup ($21.4B, up from $10.1B) signals they are managing for a Rubin ramp alongside continued Blackwell shipments.
- **Physical AI and Automotive are the emerging third revenue leg, already real.** Automotive crossed $2.3B for FY26 (+39% YoY), Professional Visualization hit $3.2B (+70%), and physical AI contributed over $6B across robotics, autonomous vehicles, and industrial AI. These are small relative to data center but growing faster and diversifying the revenue base in ways that reduce the binary “hyperscaler CapEx or bust” framing.
- **The SBC-into-non-GAAP accounting change introduces a clean comparability break starting Q1 FY27.** Q4 FY26 had $1.633B of SBC. Going forward, $1.9B/quarter is embedded in non-GAAP opex guidance. Any analyst still running FY27 estimates on the old non-GAAP basis is working off a misstated earnings base — roughly $0.25–$0.30/share of annual EPS overstated. The FY26 non-GAAP diluted EPS was $4.77 on the old basis; the SBC-inclusive historical recast gives $4.64.
---
## What Doesn’t Fit Yet
- **The goodwill jump of $15.6B is not fully explained by what’s been disclosed.** Groq was recorded separately in the investing cash flows at $13B, but goodwill typically reflects acquisition premiums over net assets. The magnitude of the goodwill increase suggests deal accounting for multiple transactions beyond the known disclosures. This warrants a close read of the Q4 10-K notes on acquisitions — what was acquired, at what implied multiple, and what is it contributing to revenue?
- **China is a zero with unknown tails in both directions.** H200 imports were approved but no revenue was generated. NVIDIA has assumed zero China data center compute revenue in Q1 FY27 guidance — and has done so for two consecutive quarters. Jensen named IPO-funded Chinese competitors as a potential long-term structural risk in the same breath as advocating for U.S. policy to re-engage China. The regulatory posture is actively fluid and the revenue impact is entirely unquantified.
- **Capital returns slowed dramatically in Q4 despite record free cash flow.** The company returned $4.1B to shareholders in Q4, down from $8.1B in Q4 FY25. Operating cash flow was $36.2B. The gap between cash generated and cash returned is the widest it has ever been, explained entirely by the $13B Groq payment and $12.8B in purchases of non-marketable equity securities. The UBS question on the call (why not a large buyback?) got a diplomatic non-answer from Colette. This is a deliberate strategic choice but investors who bought NVDA partly on capital return expectations should note the shift.
- **Gaming is broken for at least two quarters, with no clear recovery timeline.** Q4 gaming fell 13% sequentially to $3.7B. Management would not commit to year-over-year growth for FY27. The supply constraint is real and distinct from demand, but “beyond Q1” without further specificity means gaming could be a drag through H1 FY27.
- **Opex acceleration is larger than the non-GAAP numbers suggest.** On a like-for-like SBC-inclusive basis, Q1 FY27 non-GAAP opex of $7.5B compares to Q4 FY26’s restated non-GAAP of ~$6.7B (adding back $1.6B SBC that was previously excluded). That is a 12% single-quarter jump in opex. The “low 40s% full-year growth” guidance for opex is a real cost commitment at a company that will generate ~$100B in FCF— manageable but not trivial.
---
## Why It Matters
At $177.39, NVDA trades at 36.2x trailing GAAP EPS ($4.90) and approximately 38.2x the SBC-inclusive non-GAAP trailing EPS ($4.64). On FY27 consensus EPS of ~$8.29 (still on the old non-GAAP basis for most models), the stock looks like ~21x forward. On the SBC-inclusive basis, adjust that to ~$7.80–$8.00, implying ~22–23x. That multiple is not cheap for hardware, but NVIDIA is generating $96B in annual FCF with 75% gross margins, $54B in net cash, sequential revenue growth guided through all of calendar 2026, and a Rubin transition in H2 that every cloud model builder has committed to deploy. The math is defensible at current prices if the Rubin ramp executes.
The more interesting investor question is what the balance sheet transformation signals. NVIDIA is becoming a hybrid compute-infrastructure-ecosystem company, deploying capital across a strategic investment portfolio (now $22B+) and making $10–13B single-quarter bets on inference technology and frontier model companies. This is not how a traditional semiconductor company allocates capital. It is how a platform company builds lock-in. The Mellanox parallel is apt: the market didn’t immediately price Mellanox as a $31B/year business either. If the Groq and Anthropic investments follow similar trajectories, the FY27 consensus is likely too conservative.
The concentrated risk: everything in the bear case requires an external shock — China policy escalation, a hyperscaler CapEx reversal, or a Rubin execution miss. None of these are internal operational failures. They are policy and cycle risks. That is an unusual risk profile for a company at this scale, and it is why the stock is frustrating to short — the endogenous risks are low, and the exogenous ones are hard to time.
---
## What To Chase Next
1. **Read the Q4 FY26 10-K acquisitions note.** Goodwill jumped $15.6B in a year. The Groq payment was $13B in cash. These numbers need reconciliation with what was formally acquired at what price and what it’s contributing. Disclosed at the 10-K level, not in the press release or CFO commentary.
2. **Rebuild FY27 forward EPS on the SBC-inclusive non-GAAP basis** before running any multiple analysis. The $1.9B/quarter SBC load (~$7.6B annualized) reduces the FY27 non-GAAP EPS baseline by approximately $0.25–$0.30/share. Verify whether sell-side consensus has made this adjustment.
3. **Track China policy.** Any change to BIS export control posture on H200 or next-gen products is an immediate revenue event. NVIDIA is assuming zero — any positive development is unpriced upside; any escalation is additional downside risk to a $62B/quarter business already operating with a visible gap.
4. **GTC (March 16) — Groq integration details.** Jensen explicitly said to come to GTC for the Groq architecture roadmap. This is the most important near-term product disclosure. If NVIDIA can demonstrate a credible low-latency inference accelerator built on the Groq stack and CUDA-compatible, it directly addresses the one area where custom silicon (Groq, Cerebras, Tenstorrent) has competed on spec.
5. **Monitor the non-marketable equity portfolio.** $22.3B sitting in strategic investments (up from $3.4B in a year) with $8.9B in unrealized gains flowing through FY26 GAAP income. This portfolio will be marked to market quarterly and can create significant GAAP earnings noise in either direction. Worth disaggregating from operating performance in any GAAP-based analysis.
6. **Gaming supply resolution timeline.** If supply constraints ease by end of FY27 H1, gaming could contribute ~$1B+ incremental quarterly revenue versus current levels. Not material to the thesis but worth tracking as a free option on the revenue line.
---
## Sources Used
- NVIDIAAn.pdf` — NVIDIA official Q4 FY26 earnings press release including full GAAP P&L, balance sheet, cash flow statement, segment highlig, and GAAP Non-GAAP reconciliation.
- NVDA-F4Q26-Quarterly-Presentation.pdf` — Q4 FY26 investor presentation deck with visual P&L, gross margin, EPS, FCF, and outlook slides
- Rev_by_Mkt_Qtrly_Trend_Q426.pdf` — 8-quarter revenue-by-market table (Q1 FY25 through Q4 FY26)
- NVDA-Q4-2026-Earnings-Call-25-February-2026-5_00-PM-ET.pdf` — FactSet CallStreet corrected transcript, 19 pages; Q&A with BofA, Morgan Staley, JPMorgan, Bernstein, Citi, Melius, Goldman Sachs, andothers
- Q4FY26-CFO-Commentary.pdf` — Colette Kress official CFO commentary with revenue-by-platform detail, balance sheet, and Q1 FY27 outlook
- Live NVDA market data — price $177.39, trailing P/E 36.2x, analyst consensus target $268.22, next earnings May 20, 2026
- CEORater — Jensen Huang: Score 99, Alpha Score 100, TSR tenure 431,650%.