NVIDIA (NVDA) — Q2 FY2027 Catalysts and Risks
Source: transcripts/NVDA/Q2_2027.txt — NVIDIA Q2 FY2027 earnings call, August 26, 2026; cross-checked against /Users/Jun/Documents/Personal/Investments/Semiconductors/NVIDIA.xlsx (Overview tab).
Positive Catalysts
| # | Catalyst | Evidence | Horizon | Monitorable | Source |
|---|---|---|---|---|---|
| P1 | Vera Rubin turns production into revenue — the platform cycle becomes the fiscal 2028 engine | ~20% of Q3 Data Center revenue guided (workbook C3); purchase orders from every major hyperscaler, AI cloud and system OEM — “we expect Vera Rubin to mark the fastest product ramp in NVIDIA’s history” | Next quarter | Vera Rubin’s revenue contribution and ramp commentary in the Q3 report versus the guided mix | Company-specific |
| P2 | Multi-year commitments underwrite growth beyond the guide — OpenAI and AWS lock in capacity | OpenAI’s existing and planned commitments: ~12 GW of NVIDIA compute through 2030; AWS deploying an additional 2 million GPUs through Q2 FY2029 | Twelve months | Deployment milestones and purchase commitments in filings; new multi-year supply agreements | Company-specific |
| P3 | The non-hyperscaler half scales — ACIE and sovereign demand diversify the buyer base | ACIE revenue +138% year over year; sovereign AI more than tripled year over year — “that part of the world is invisible to everybody” | Next quarter | ACIE growth in the Q3 print versus guidance that sequential growth is driven primarily by ACIE | Company-specific |
| P4 | AI infrastructure demand keeps compounding — capex and backlogs expand the addressable market | Cloud industry backlog >$2 trillion; top-5 hyperscaler capex expected to reach $1.3 trillion in 2027 (workbook C4) — “more compute drives more revenue as new GPU capacity comes online” | Twelve months | Hyperscaler capex updates and cloud backlog disclosures; AI-native funding data | Sector/macro |
| P5 | Networking and CPU attach lift the dollar content per deployment | Spectrum-X Ethernet grew 2.6x year over year; CPU revenue expected to more than double in fiscal 2028 | Twelve months | Networking growth and CPU revenue progression against the more-than-doubling fiscal 2028 target | Company-specific |
| P6 | The platform moat keeps lifting monetization — every model and generation monetizes better | Revenue opportunity per gigawatt: $40 billion with Vera Rubin versus $25 billion with Blackwell (workbook C6); the only platform that runs every frontier model, closed or open (Q5) — “the vast majority of the world’s data centers will just be Vera Rubin and NVLink 72” | Twelve months | Revenue per gigawatt disclosed in future workbooks and platform-generation commentary | Company-specific |
| P7 | Groq 3 LPX opens a second inference architecture — a new product cycle starts shipping | Nearly 4x tokens per second versus the next-best alternative on the Artificial Analysis benchmark; volume shipments expected later this quarter — “We achieved record token interactivity rate” | Next quarter | Volume shipments to early adopters and follow-on design wins in the Q3 report | Company-specific |
| P8 | Verticals and software add a second demand leg beyond data center | Automotive trailing-twelve-month on-prem revenue $8 billion; financial services, manufacturing and healthcare combined $7 billion (workbook C8) | Twelve months | Vertical trailing-twelve-month revenue and AI Enterprise software commentary in future updates | Company-specific |
| P9 | Capital returns accelerate — holders get paid while the buildout continues | Record $26 billion returned in Q2; 60% of free cash flow returned year-to-date against a plan of 50% or more — “we intend to increase and return excess free cash flow net of strategic uses” | Twelve months | Quarterly return of capital versus the stated free-cash-flow payout commitment | Company-specific |
Risks
| # | Risk | Evidence | Horizon | Monitorable | Source |
|---|---|---|---|---|---|
| R1 | Memory-cost inflation resets the margin profile — the trough is guided, the edge is not | Q4 gross margin guided to bottom at 71–72% (workbook D6) — “we are experiencing extreme pricing conditions in memory… headed even higher into next year” | Next quarter | Memory contract pricing and the Q3/Q4 gross margin prints versus the guided floor | Sector/macro |
| R2 | Supply stays the binding constraint — growth is capped by capacity, not demand | Fiscal 2028 growth guided to ~70% with supply a bottleneck at least through end-FY2028; over 60% of TSMC’s 2026 CoWoS capacity secured (workbook D7) — “our entire supply chain is challenged” | Twelve months | Whether guidance inflects higher as capacity comes online; foundry and packaging capacity announcements | Company-specific |
| R3 | Customer financing concentrates credit exposure — fiscal 2028 revenue leans on NVIDIA-supported buyers | ~25% of projected FY2028 revenue tied to labs and neoclouds receiving NVIDIA balance-sheet or credit support (workbook D8); top three direct customers were 44% of H1 FY2027 revenue (workbook D4) — “some will call this circular financing. We see it differently” | Twelve months | Credit-support and guarantee disclosures; lab fundraising closings; customer-concentration figures in the next 10-Q | Company-specific |
| R4 | Custom silicon and AMD contest the fastest-growing workload — inference alternatives industrialize | Custom AI ASICs already hold 66.3% of the inference segment of that market (workbook D3); analysts relayed OpenAI’s “Jalapeno” chip claimed to outperform Blackwell (analyst-provided; management did not address the claim directly) — “we’re building something very different” | Twelve months | Custom-ASIC deployment news and benchmarks; AMD accelerator ramp; NVIDIA inference-share commentary | Sector/macro |
| R5 | Inventory and purchase obligations raise the cost of a demand stumble | Inventory increased to $32 billion; non-cancelable purchase and supply obligations reached $145 billion (workbook D5) — extended “payment terms for large purchases by certain investment-grade customers” | Twelve months | Inventory, days sales outstanding, and commitment balances in the next filing | Company-specific |
| R6 | China stays closed — policy keeps a large market unmonetized | Less than 1% of Q2 data center revenue shipped to China under licenses — “given ongoing geopolitical uncertainty, there is no China data center compute revenue in our forward outlook” | Twelve months | Export-license policy changes and whether China revenue returns to guidance | Sector/macro |
Footnote — evidence-backed items not given rows: Edge Computing’s growth (+27% year over year to $7.2 billion, workbook’s Q2 FY2027 financial statements) is excluded as immaterial at less than a tenth of Data Center’s scale, with no call evidence that it steers the investment case; the open-model debate (Q5) resolved in NVIDIA’s favor with no figure or metric to monitor and is absorbed by the P6 platform-moat row; the workbook’s $250 billion OpenAI financial-guarantee talks (Important Events) fold into the R3 credit-exposure row; and pre-quarter events (the Meta multiyear partnership, the Humain GB300 deal) are superseded by the commitment rows above.
Spreadsheet Cross-Check
Coverage Audit
| Spreadsheet row | Mapped report row(s) | Verdict |
|---|---|---|
| C3 — Vera Rubin production ramp | P1 | covered |
| C4 — Hyperscaler capex and cloud backlogs | P4 | covered |
| C5 — NeoCloud and sovereign AI expansion | P3 | covered |
| C6 — Revenue-per-gigawatt ladder with attached CPUs and networking | P5, P6 | covered |
| C7 — OpenAI and AWS commitments | P2 | covered |
| C8 — Automotive, vertical, and software revenue | P8 | covered |
| C9 — CUDA moat | P6 | covered |
| C10 — Co-design advantage | P6 | covered |
| D3 — AMD and custom ASIC competition | R4 | covered |
| D4 — Customer concentration | R3 | covered |
| D5 — Inventory, DSO, purchase commitments | R5 | covered |
| D6 — Memory-driven margin reset | R1 | covered |
| D7 — China restrictions and TSMC dependency | R2, R6 | covered |
| D8 — Balance-sheet-supported demand exposure | R3 | covered |
| — | P7 | new |
| — | P9 | new |
New Positive Catalysts (Missing from Spreadsheet)
- Groq 3 LPX begins shipping into a new inference niche — the P7 row’s evidence pairs a first-of-its-kind rack-scale LPU system with volume shipments starting this quarter, opening a latency-sensitive, high-ASP revenue stream the workbook’s platform ladder does not yet carry. Horizon: Next quarter · Monitorable: Volume shipments to early adopters and follow-on design wins in the Q3 report · Source: Company-specific.
- Capital returns are running ahead of plan — the P9 row’s evidence shows excess cash returned well above the stated payout commitment, with management intent to increase it; a rising distribution cadence gives holders a direct claim on the buildout’s economics. Horizon: Twelve months · Monitorable: Quarterly return of capital versus the stated free-cash-flow payout commitment · Source: Company-specific.
New Future Risks (Missing from Spreadsheet)
Nothing new — every risk row maps to a workbook Future Risks row, as the Coverage Audit shows.