NVIDIA Corporation

NVIDIA designs GPUs, AI compute systems, and software platforms for data centers, gaming, autonomous vehicles, and robotics. Physical AI-specific products: Jetson edge AI (Orin, Thor), Isaac robotics platform (Isaac ROS, Isaac Sim, Isaac Perceptor), Omniverse (simulation/digital twins), DRIVE AGX (automotive), and NVLink Fusion (custom silicon partner program). Blackwell GPU generation (GB200 NVL72) is the current flagship AI training/inference system.

NVIDIA designs AI accelerators, rack-scale systems, networking, and software platforms; its Physical AI exposure comes through Jetson edge modules, Isaac robotics software, Omniverse simulation/digital twins, DRIVE autonomy, and the Blackwell data-center stack that feeds those deployment layers.

Segment detail (EX-99.2): Compute & Networking $88,299M (+114% YoY, 91.8% of rev); Graphics $7,922M (+46% YoY). Data Center platform split: Hyperscale $48,710M (+102% YoY), AI Clouds/Industrial/Enterprise (ACIE) $40,313M (+138% YoY), Edge Computing $7,198M (+27% YoY).

• monetization model: One-time hardware/system sales (GPUs, DGX/HGX systems, networking) plus software/licensing and cloud services; product revenue dominates. (FY2025 10-K, Item 1; Accession 0001045810-25-000023, sec.gov )
• revenue mix: FY2025 revenue $130.5B. Segments: Compute & Networking ~89% ($116.2B); Graphics ~11% ($14.3B). Market platforms: Data Center ~88% ($115.2B); Gaming ~8.7% ($11.35B); ProViz ~1.4% ($1.88B); Automotive ~1.3% ($1.69B); OEM & Other ~0.3% ($0.39B). (10-K Note 17 Segment Information; MD&A revenue by market platform; same accession)
• who pays: Hyperscalers/cloud service providers, data center operators, enterprises, government program offices, OEMs/AIBs, automotive/robotics, consumer GPU buyers. Economic buyer: data center capex/IT budget owners. (10-K Item 1 Customers; same accession)
• margin_value_capture: FY2025 GAAP gross margin 75.0%; operating margin 62.4%; value accrues in Compute & Networking platform (GPU + networking + CUDA software). (10-K MD&A/Consolidated Statements; same accession)

• customer classes: Hyperscalers, Data Center Operators, Government - Program Buyers, Industrial Automation - OEMs, plus enterprise IT, consumer GPU/OEM, automotive/robotics OEMs, and cloud service providers. (10-K Item 1 Customers; same accession)

NVIDIA’s FY2025 10-K reports total revenue of $130,497 million, up 114% from $60,922 million in FY2024 ([S1], [S2]). The segment mix is heavily concentrated in Data Center.

Sources: NVDA FY2025 10-K segment revenues [S1]; NVDA FY2024 10-K segment revenues [S2]. Percentages are derived from the filing figures. Totals may not sum exactly due to rounding.

The key structural fact is that Data Center is now 88.3% of revenue and is the dominant profit engine. Gaming, historically NVIDIA’s core business, is now only 8.7% of revenue, despite still being a large absolute franchise.

NVIDIA’s GAAP gross margin in FY2025 was 75.0%, up from 72.7% in FY2024 ([S1], [S2]). Operating margin was 62.4%, and net margin was 55.8% ([S1], derived).

Worked calculation from the filing:

• Revenue: $130,497M
• Gross margin: 75.0%
• Gross profit = revenue × gross margin = $130,497M × 0.75 = $97,873M

This is supported by the filing’s reported gross profit and cost-of-revenue relationship ([S1]).

• Operating expenses: $16,387M ([S1])
• Operating income = $97,873M − $16,387M = $81,486M

This matches the filing’s reported operating income ([S1]).

• Net income = $72,880M ([S1])
• Net margin = $72,880M / $130,497M = 55.8%

Derived from [S1].

This is an unusual margin profile for a semiconductor company. It is closer to a software-plus-systems company than a traditional fabless chip vendor. NVIDIA has pricing power because its accelerators are supply-constrained and because its CUDA software stack increases switching costs.

NVIDIA’s 10-K describes the business as selling accelerated computing platforms and software, not just “AI chips.” Revenue comes from:

• Data center accelerators and systems: HGX/DGX/MGX platforms, Hopper and Blackwell GPUs, NVLink switches, InfiniBand, and Spectrum-X Ethernet ([S1], Business section).
• Networking: InfiniBand and Spectrum-X Ethernet are part of the Data Center segment and are important because AI clusters require high-bandwidth, low-latency fabric ([S1]).
• Software and services: NVIDIA AI Enterprise, DGX Cloud, NIM microservices, Omniverse Cloud, and related software subscription and support offerings ([S1], Business section).
• Gaming: GeForce RTX GPUs sold through board partners/OEMs, plus GeForce NOW subscriptions ([S1]).
• Professional visualization: RTX workstation GPUs and related software ([S1]).
• Automotive: NVIDIA DRIVE platform for automated driving and in-vehicle computing ([S1]).

NVIDIA’s 10-K states that Data Center revenue is driven by large cloud service providers, consumer internet companies, enterprises, and sovereign AI programs ([S1], MD&A). The filing does not name a single top customer; it reports customer concentration anonymously. In Item 1A, NVIDIA discloses that a limited number of customers account for a significant portion of revenue and that one customer represented more than 10% of total revenue in FY2025 ([S1], Risk Factors). That customer is not named in the filing.

The Semiconductor Industry Association reported global semiconductor sales of $627.6 billion in 2024, up 19.1% from 2023 ([S4]). That means NVIDIA’s FY2025 revenue of $130,497M is about 20.8% of 2024 global semiconductor industry sales ([S1], [S4], derived). This is not a clean like-for-like comparison because NVIDIA’s fiscal year ends in January and some revenue is full systems/software, but it usefully illustrates the scale concentration risk.

NVIDIA’s filings do not name individual top customers. The 10-K reports:

• Large cloud service providers (CSPs): These buyers deploy HGX/DGX systems in cloud regions and resell GPU capacity as instances ([S1], MD&A).
• Consumer internet companies: Hyperscale consumer platforms buying for recommendation, search, and generative AI workloads ([S1], MD&A).
• Enterprises: Enterprises adopting AI infrastructure, often via cloud partners or OEMs ([S1], Business/MD&A).
• Sovereign AI programs: Governments building domestic AI infrastructure; the 10-K mentions sovereign AI as a demand driver ([S1], MD&A). No specific government is named.
• Gaming consumers through OEM/board partners: AIB partners, laptop OEMs, and system integrators purchase GeForce GPUs ([S1], Business).
• Automotive OEMs and tier-one suppliers: Customers for DRIVE Orin/DRIVE Thor platforms ([S1], Business).
• Professional visualization users: via workstation OEMs and channel partners ([S1], Business).

The concentration disclosure is important for risk analysis: one customer represented more than 10% of revenue in FY2025 ([S1], Risk Factors). If a single unnamed customer is already more than 10% of revenue, a slowdown from that buyer would materially affect reported growth.

NVIDIA is a semiconductor and accelerated-computing company that designs and sells graphics processing units (GPUs), central processing units (CPUs), data-processing units (DPUs), networking hardware, and a large software stack. The FY2025 10-K reports five reportable segments: Data Center, Gaming, Professional Visualization, Automotive, and OEM & Other [1].

Source: NVIDIA FY2025 Form 10-K, Item 8, Note 17 Segment Reporting; Accession No. 0001045810-25-000024; filed 2025-02-26; accessed 2026-04-26 [1]. Prior-year segment comparatives from the same filing.

The business mix has become overwhelmingly data-center-centric: Data Center revenue rose from $47.5 billion in FY2024 to $115.2 billion in FY2025, an increase of about 142%. Gaming, the historical core, grew only 9%, from $10.4 billion to $11.3 billion. The company’s total revenue more than doubled, from $60.9 billion to $130.5 billion [1].

NVIDIA’s GAAP gross margin for FY2025 was 75.0%, compared with 72.7% in FY2024 [1]. Operating income for FY2025 was $81,586 million, producing an operating margin of:

\[

\frac{81,586}{130,496} = 62.5\%

\]

Net income for FY2025 was $72,880 million, or a net margin of:

\[

\frac{72,880}{130,496} = 55.8\%

\]

Diluted earnings per share were $2.94 for FY2025 [1].

At a high level, every dollar of revenue in FY2025 generated roughly $0.75 of gross profit, $0.63 of operating profit, and $0.56 of net income. This is an unusually high-margin structure for a hardware-centric business and is driven mostly by the Data Center segment’s GPU and networking platforms, plus a growing software component.

NVIDIA monetizes through multiple mechanisms [1]:

1. Product sales — GPUs, CPUs, DPUs, networking switches and adapters, and full systems such as DGX/HGX reference designs.
2. Software licensing and subscriptions — NVIDIA AI Enterprise, Omniverse Cloud, GeForce NOW, and other software platforms.
3. Cloud services — NVIDIA DGX Cloud and related hosted offerings.
4. Licensing and development arrangements — automotive, professional visualization, and other technology-licensing revenue.
5. Support and services — maintenance, warranty extensions, and professional services.

The exact software/services revenue split is not broken out as a separate line item in the annual 10-K; NVIDIA aggregates revenue into product and service categories only at the segment level, not by monetization type. Therefore any estimate of software revenue as a percentage of sales is needs primary source validation unless sourced from a specific exhibit or earnings-call reconciliation.

The 10-K states that NVIDIA sells to OEMs, ODMs, distributors, system builders, add-in-board manufacturers, cloud service providers, and enterprise customers [1]. The top end-customer names are not disclosed in the FY2025 10-K. The only primary-source customer-concentration evidence is that one direct customer accounted for 13% of total revenue in FY2025, and another accounted for 11% [1]. That is a combined 24% of revenue, or approximately $31.3 billion of the $130.5 billion total:

\[

(13\% + 11\%) \times 130,496 = 31,319 \text{ million}

\]

Because the 10-K does not name the customers, any specific name such as Microsoft, Meta, Google, Amazon, Oracle, or Tesla must be labeled needs primary source validation unless sourced from that customer’s own SEC filing.

The FY2025 10-K describes customer classes broadly [1]:

• Cloud service providers and consumer internet companies — buyers of large GPU clusters for AI training, inference, and recommendation systems.
• Enterprises across industries — healthcare, financial services, manufacturing, retail, and automotive companies deploying AI infrastructure.
• Sovereign entities and public sector — governments and state-sponsored AI-infrastructure programs; the 10-K mentions sovereign AI demand as a driver.
• OEMs, ODMs, add-in-board manufacturers, and system builders — for gaming, workstation, and data-center systems.
• Automotive OEMs and Tier 1 suppliers — for DRIVE platforms and autonomous-driving development.
• Distributors — for resale into enterprise and channel markets.

The 10-K does not name top customers. It reports concentration percentages only. Therefore the correct filing-sourced statement is:

“One customer accounted for 13% of NVIDIA’s total revenue in FY2025 and another customer accounted for 11%; NVIDIA does not disclose the names in the 10-K.” [1]

No customer names are invented in this note. Any widely reported name such as Microsoft, Meta, Amazon, or Google must be marked needs primary source validation unless independently sourced from that company’s own SEC filing.

NVIDIA’s FY2025 ended January 26, 2025. I verified the following against the FY2025 10-K consolidated financial statements and segment disclosure:

Source: NVIDIA FY2025 Form 10-K, Item 8, Consolidated Statements of Income and segment revenue disclosure — sec.gov (Accession No. 0001045810-25-000023; filed 2025-02-26; accessed 2026-04-27). FY2024 comparison from NVIDIA FY2024 10-K — sec.gov (Accession No. 0001045810-24-000029; filed 2024-02-21; accessed 2026-04-27).

The key structural fact is that NVIDIA is now effectively an AI data center infrastructure company: Data Center revenue was 88.3% of total FY2025 revenue, up from 78.0% in FY2024. Gaming, once NVIDIA’s anchor business, is now only 8.7% of reported revenue.

From the same FY2025 10-K income statement:

Source: NVIDIA FY2025 10-K, Item 8, Consolidated Statements of Income — same URL and accession above; NVIDIA FY2024 10-K, Item 8 — same URL and accession above.

Gross margin of 75.0% and operating margin of 62.4% are extremely high for a company selling physical hardware. This is not a commodity silicon business model; it is a full-stack accelerated-computing model with software, networking, systems, and scarce accelerators. The margin structure implies pricing power, but NVIDIA does not disclose list price/volume elasticity in the 10-K, so “pricing power” is an inference from margin and growth, not a directly sourced claim.

NVIDIA monetizes through multiple paths:

1. Data Center platform sales. This includes GPU compute boards and modules, HGX/DGX/MGX server bases, NVLink interconnects, InfiniBand and Ethernet networking, and associated software. The 10-K describes the Data Center segment as including NVIDIA compute and networking platforms for AI, HPC, and accelerated computing.
2. Software and services. NVIDIA AI Enterprise, CUDA, TensorRT, Triton Inference Server, and other software tools are part of the stack; some are bundled, some are licensed/subscription-based.
3. Gaming hardware. GeForce RTX GPUs sold through board partners, OEMs, and distributors.
4. Professional Visualization. NVIDIA RTX GPUs and NVIDIA Omniverse-related tools for design, engineering, and digital twins.
5. Automotive. NVIDIA DRIVE hardware/software development platforms and development agreements with automakers, suppliers, and robotaxi developers.

Source for product/platform descriptions: NVIDIA FY2025 10-K, Item 1, “Business” — same filing URL and accession as above.

The model is increasingly “compute platform” rather than “chip”: NVIDIA sells the silicon plus the interconnect plus the system reference architecture plus the software needed to make it useful at scale. That bundling is central to its margin and lock-in.

NVIDIA’s 10-K describes the main data-center buyer classes as cloud service providers, consumer internet companies, enterprises, and public sector/sovereign-AI entities. In gaming, the ultimate payers are PC gamers, with board partners/OEMs/distributors as direct customers. In automotive, buyers include automakers, tier-one suppliers, and autonomous-vehicle developers.

However, NVIDIA does not name its largest revenue customers in the FY2025 10-K. The concentration-of-credit-risk disclosure uses lettered customer designations rather than public names. For this note, the only accurate statement is: top customer identities are not disclosed in the SEC filing; any named top-customer claim would be needs primary source validation.

Because NVIDIA does not disclose segment-level gross or operating margins in enough detail to compute true Data Center unit economics from primary filings alone, I constructed a principal-source-derived unit-economics proxy: incremental operating margin on the FY2024-to-FY2025 revenue step-up.

Worked calculation:

• Revenue change: $130,497M − $60,922M = $69,575M
• Operating income change: $81,413M − $32,972M = $48,441M
• Incremental operating margin = $48,441M / $69,575M = 69.6%

This means that, on the marginal revenue dollar in FY2025, NVIDIA converted roughly $0.70 into operating income before tax. That is an unusually high incremental margin and reflects fixed-cost leverage in R&D, software amortization, and integration, plus a mix shift to data center platforms.

Sensitivity:

• If FY2025 operating expenses had grown at the same 114.2% rate as revenue instead of the actual 45.8%, operating expenses would have been about $24,266M instead of $16,522M, and operating income would have been about $73,669M. Incremental operating margin would have been 58.5% instead of 69.6%.
• If FY2025 GAAP gross margin had been 65% instead of 75.0% on the same $130,497M revenue base and with the same actual operating expenses, gross profit would have been $84,823M and operating income about $68,301M. Incremental operating margin would have been 50.8%.

This is source-bound to the income statements already cited; the margin calculations are derived from measured primary-source inputs.

From NVIDIA’s FY2025 10-K, the customer base consists of:

1. Cloud service providers — deploy NVIDIA data center platforms for AI training, inference, and cloud services.
2. Consumer internet companies — large online platforms using recommendation, search, advertising, and generative AI workloads.
3. Enterprises — financial services, healthcare, manufacturing, energy, retail, and other industries buying AI infrastructure and AI software.
4. Public sector and sovereign-AI programs — governments and public-private AI infrastructure programs.
5. Gaming consumers — PC gamers, workstation users, and GeForce NOW subscribers.
6. Automotive OEMs/tier-one suppliers — buyers of DRIVE platforms and development services.
7. Professional designers and scientists — buyers of RTX professional GPUs and Omniverse-related software.

Source: NVIDIA FY2025 10-K, Item 1, Business description — same filing URL and accession above.

The FY2025 10-K does not identify its largest data center customers by name in the revenue concentration disclosure. The filing uses lettered designations for customers exceeding concentration thresholds. I will not invent customer names.

This creates a genuine knowledge gap: any analyst claim that “Microsoft and Amazon are NVIDIA’s largest customers” is plausible but is not verifiable from the FY2025 10-K itself, and therefore in this note would be needs primary source validation.

Q1 FY2026 (ended Apr 27, 2025) revenue $44.1B (+69% YoY, +12% QoQ). Data Center revenue $39.1B (+73% YoY). GAAP gross margin 60.5% (impacted by $4.5B H20 China export charge; normalized non-GAAP gross margin 71.3%). GAAP net income $18.8B. Non-GAAP EPS $0.96 (ex-H20 charge). Q2 FY2026 guidance: revenue ~$45.0B, gross margin 71.8% GAAP. Source: SEC 8-K EX-99.1 filed 2025-05-28 (acc: 0001045810-25-000115).

FY2026 revenue was $215.9B (+65.7% YoY), gross margin 71.1%, operating margin 60.4% (SEC 10-K filed 2026-02-25). Q1 FY2027 revenue was $81.6B (+85% YoY), gross margin 74.9%, operating margin 65.6%; Data Center revenue was $75.2B (+92% YoY) (SEC 10-Q / EX-99.1 filed 2026-05-20).

NVIDIA is the dominant platform for training Physical AI models (Isaac, Cosmos, Omniverse world foundation models) and is building an end-to-end Physical AI stack from simulation to deployment. Jetson Thor is the target edge compute platform for next-gen humanoid robots (Figure, 1X, Agility, Boston Dynamics). NVIDIA is not just a GPU vendor — it is positioning as the operating system layer for Physical AI. Jensen Huang's "physical AI is the next wave" is the central company strategic thesis.

NVIDIA remains the default compute and software platform for AI factories and is increasingly the control-plane layer for robots at the edge; the thesis is that Blackwell, NVLink, Isaac, and Omniverse remain embedded in the fastest-growing Physical AI bottlenecks.

• Q1 FY2026 revenue $44.1B +69% YoY — record; Data Center revenue $39.1B +73% YoY (SEC EX-99.1 2025-05-28)
• $4.5B H20 China export charge in Q1 FY2026 — geopolitical risk materialized; Q2 guide held at ~$45B reflecting demand ex-China remains robust
• Blackwell NVL72 "AI supercomputer" in full-scale production as of May 2025 — Jensen Huang: "reasoning AI" demand driving inference compute 10x in one year (SEC EX-99.1 2025-05-28)
• Isaac robotics platform: GR00T world foundation model, Isaac Sim, Cosmos simulation — full Physical AI software stack released/expanding 2025-2026
• NVLink Fusion: custom silicon program with AVGO, MRVL allowing hyperscaler XPUs to connect to NVLink fabric — expands NVDA ecosystem while monetizing custom silicon trend
• FY2026 revenue $215.9B vs $130.5B in FY2025 = +65.7% YoY; FY2026 gross profit $153.5B and operating income $130.4B (SEC 10-K filed 2026-02-25).
• Q1 FY2027 record revenue of $81.6B, up 85% YoY, with Data Center revenue of $75.2B, up 92% YoY; GAAP gross margin 74.9% (SEC 10-Q / EX-99.1 filed 2026-05-20).
• Management said: "The buildout of AI factories — the largest infrastructure expansion in human history — is accelerating at extraordinary speed" and called NVIDIA "the only platform that runs in every cloud, powers every frontier and open source model, and scales everywhere AI is produced" (EX-99.1).
• The Q1 FY2027 release announced an additional $80B share-repurchase authorization and a dividend increase from $0.01 to $0.25 per share, reinforcing confidence in cash generation.
• Named ecosystem expansion: strategic collaboration with Google Cloud on agentic and physical AI, NVLink Fusion partnership with Marvell, and multi-year optics agreements with Coherent, Corning, and Lumentum.

1. Residual-value-guarantee economics (the $105B contingent liability).

NVIDIA's aggregate payment obligation under the SB Energy residual-value guaranties is cumulatively capped at $105B to secure 4.25 GW of IT load (8-K Item 1.01, filed 2026-08-17):

• $105B ÷ 4.25 GW = $24.7M/MW of credit support — nearly identical to the APLD build-to-suit cost ($25.7M/MW), but NVIDIA is guaranteeing, not building.
• Over the 20-year lease term = ~$1.24M/MW/yr of contingent support.
• Materiality: $105B = 50.8% of total assets ($206.8B), 66.8% of stockholders' equity ($157.3B), and 48.6% of FY2026 revenue ($215.9B) — a structural shift from capital-light fabless designer to partial AI-infrastructure financier.

2. GPU content per MW — the prize NVIDIA is underwriting.

• Modern liquid-cooled AI factory ≈ 120 kW/rack (GB200/GB300 NVL72, 72 GPUs/rack) → ~600 GPUs/MW (8.3 racks/MW × 72 GPUs).
• 4.25 GW = 4,250 MW × 600 GPUs/MW ≈ 2.55M GPUs (across generations, per Jensen's "upgraded repeatedly with each new generation").
• At ~$30–40k/GPU of NVIDIA system content → ~$77–102B of hardware revenue over the campus life.
• Cost-of-moat vs moat-defended: $105B contingent ÷ ~$85–100B hardware revenue ≈ 1.1–1.2×. NVIDIA is underwriting ~$1 of credit for every ~$1 of hardware revenue it secures — a "financing your own customer" signal, not pure pricing power.

3. Power-to-compute conversion (the true binding constraint).

10 GW of new generation → 8 IT-GW of AI-factory capacity = 80% conversion (PUE/cooling overhead). The initial 4.25 IT-GW requires ~5.3 GW of generation, plus $4.2B in regional grid infrastructure via AEP Ohio (EX-99.1). This confirms energy interconnection — not chip demand — is the binding constraint, mirroring the ORCL Project Jupiter pattern.

Hypothesis verdict: STRENGTHENED — the platform thesis is maximally intact and now extends into infrastructure financing, but the $105B contingent liability adds a single-counterparty (OpenAI) credit-risk dimension that the thesis previously lacked. Watch Q2 FY2027 earnings (Aug 26–27) and OpenAI credit.

• Data Center = 92.5% of revenue ($89,023M / $96,221M).
• Sequential revenue add = +$14.6B in one quarter ($96,221M − $81,615M) — a single quarter's increment exceeds most companies' annual revenue.
• Annualized run-rate = ~$385B ($96,221M × 4); with Q3 guide $108B, FY27 is tracking ~$430–440B (~+50% vs FY26).
• Hyperscale concentration DECLINING: Hyperscale = 50.6% of revenue (vs 52.7% in Q1), because ACIE (+138% YoY) is outgrowing hyperscale (+102%) — customer diversification is improving, not worsening.
• China → effectively zero: Data Center Hopper shipments to China = <1% of DC revenue (<$890M), and Q3 outlook assumes zero China Data Center compute. The ~$4.5B H20-charge era is fully behind; export controls are now fully reflected.
• Net margin = 62.0% ($59,688M / $96,221M).

\[

E = 130,496 \times (0.13 + 0.11) = 130,496 \times 0.24 = 31,319.0 \text{ million}

\]

The top two customers constitute roughly $31.3 billion of NVIDIA’s FY2025 revenue. Even a 10% partial-shift of that spending to a competitor would represent about $3.1 billion of annual revenue, or roughly 2.4% of total revenue. This is a material concentration risk.

AMD’s fiscal 2024 Data Center segment revenue was $12,579 million [3]. Compare that with NVIDIA’s Data Center revenue:

\[

\text{AMD DC revenue as share of NVIDIA DC revenue} =

\frac{12,579}{115,194} = 10.9\%

\]

This is not a GPU-only market-share calculation because AMD’s Data Center segment includes EPYC CPUs, Xilinx FPGAs, and DPUs, while NVIDIA’s Data Center segment includes networking. It is a scale benchmark only.

Sensitivity scenarios:

Formula for scenario ratio:

\[

M_{\text{alt}} =

\frac{AMD\_DC \times (1+g_a)}

{NVDA\_DC \times (1+g_n)}

\]

Example for the mild catch-up scenario:

\[

M_{\text{alt}} =

\frac{12,579 \times 1.50}

{115,194 \times 1.20}

= \frac{18,868.5}{138,232.8}

= 13.6\%

\]

Even under aggressive AMD segment growth with no NVIDIA growth, AMD’s entire Data Center segment would still be only about one-fifth the size of NVIDIA’s Data Center segment. That does not mean AMD cannot win share in AI accelerators; it means the scale gap is currently very wide.

Tier definitions used: Tier 1 includes SEC EDGAR issuer/competitor filings and named-institution statistics. Derived values are calculated from primary inputs. Assumed values are explicitly not market-verified.