## Executive takeaway: “Beat-and-raise” is no longer the whole game NVIDIA’s latest disclosures read less like a near-term demand debate and more like a *positioning* debate: investors are shifting from celebrating the current print to stress-testing (i) durability of hyperscaler/AI-model-maker spend, (ii) customer concentration risk, and (iii) whether NVIDIA’s “full-stack” edge persists as workloads tilt toward inference and custom silicon. ["revenue could be adversely affected"](0001045810-26-000021|0) ["revenue is concentrated"](0001045810-26-000021|0) ["addressing both training and inferencing"](0001045810-26-000021|0) --- ## 1) The tell in the 10-K: OpenAI is framed as upside—explicitly not “in the bag” Buried in the investments discussion, NVIDIA states it is “finalizing an investment and partnership agreement with OpenAI,” but then immediately caveats that there is “no assurance” it will enter into an agreement or that any transaction will be completed. That language matters because it converts what many investors mentally model as *strategic inevitability* into *contingent optionality*—i.e., potential upside with real execution/timing risk. ["finalizing an investment"](0001045810-26-000021|0) ["no assurance"](0001045810-26-000021|0) **Alpha between the lines:** the market is increasingly pricing AI demand as “known,” so *incremental* valuation sensitivity migrates to the *uncertain nodes* (partnerships, customer pacing, and mix)—exactly where NVIDIA is choosing to insert explicit caution. ["no assurance"](0001045810-26-000021|0) --- ## 2) Concentration is the real macro: two direct customers = 36% of revenue NVIDIA discloses that in fiscal year 2026, one direct customer represented 22% of total revenue and another represented 14%—primarily attributable to the Compute & Networking segment. In practical terms, the company can deliver extraordinary aggregate results and still see the stock trade poorly if investors fear even a modest “air pocket” from a small set of buyers (purchase orders can be “cancel[ed], change[d], or delay[ed]… with little notice”). ["22% of total revenue"](0001045810-26-000021|0) ["represented 14%"](0001045810-26-000021|0) ["cancel, change, or delay"](0001045810-26-000021|0) **Alpha between the lines:** this is why the market’s attention snaps to signals about hyperscaler free cash flow discipline, internal chips, and pacing—because concentration makes forward demand *lumpier* than the headline growth rate implies. ["revenue is concentrated"](0001045810-26-000021|0) --- ## 3) Guidance is strong—but it quietly de-risks (and caps) one pillar: China compute In its latest earnings materials, NVIDIA guides Q1 FY2027 revenue to $78.0 billion (±2%) and explicitly says it is “not assuming any Data Center compute revenue from China” in that outlook. That is simultaneously (i) a credibility/visibility move (don’t rely on what may be constrained) and (ii) an admission that a prior source of volume is being treated as non-base-case. ["Revenue is expected to be $78.0 billion"](0001045810-26-000019|1) ["not assuming any Data Center compute revenue from China"](0001045810-26-000019|1) **Alpha between the lines:** the “beat” can be real, but the *composition* of the forward curve is getting more conservative—so the equity reaction becomes more about *what’s excluded* than what’s included. ["not assuming"](0001045810-26-000019|1) --- ## 4) The durability question: NVIDIA is investing hard, and it’s signaling confidence with commitments NVIDIA reports it spent $6.1 billion on capital expenditures in FY2026 (vs. $3.4 billion in FY2025) and expects to increase capex again in FY2027. Separately, management commentary notes inventory at $21.4 billion and “total supply-related commitments” of $95.2 billion, emphasizing it has “strategically secured inventory and capacity to meet demand beyond the next several quarters.” Those are not the disclosures of a management team bracing for an immediate cliff; they read like a company planning for continued scale—while accepting that investors may worry about the cash flow implications of running that hard. ["spent $6.1 billion"](0001045810-26-000021|0) ["increase capital expenditures"](0001045810-26-000021|0) ["Inventory was $21.4 billion"](0001045810-26-000019|2) ["total supply-related commitments were $95.2 billion"](0001045810-26-000019|2) ["secured inventory and capacity"](0001045810-26-000019|2) **Alpha between the lines:** this is the pivot point for the next leg—if demand persists, these commitments can widen the moat (availability, speed-to-deploy). If demand normalizes, they become the raw material for gross margin and working capital anxiety. ["secured inventory and capacity"](0001045810-26-000019|2) --- ## 5) Why the stock can drop on “blowout” numbers: the market is now underwriting *moat maintenance* NVIDIA is explicitly positioning around “both training and inferencing,” and highlights software (e.g., “Dynamo inference software”) plus architectural cadence as part of its edge. The subtext: investors are not just buying GPUs—they’re buying an ecosystem that must stay sticky as the industry experiments with alternatives (custom silicon, integration partners, shifting workload mix). So a selloff post-beat is less paradox than it looks: the market is demanding evidence that NVIDIA’s advantage persists *as inference becomes the dominant unit of compute economics*. ["training and inferencing"](0001045810-26-000021|0) ["Dynamo inference software"](0001045810-26-000021|0) --- If you want, I can turn this into a one-page “hidden alpha” tear sheet (Key debate / What’s priced in / What could surprise) using only the same 10‑K + 8‑K exhibits. ["10-K"](0001045810-26-000021|0) ["EX-99.1"](0001045810-26-000019|1) ["EX-99.2"](0001045810-26-000019|2)

## Executive takeaway: “Beat-and-raise” is no longer the whole game NVIDIA’s latest disclosures read less like a near-term demand debate and more like a *positioning* debate: investors are shifting from celebrating the current print to stress-testing (i) durability of hyperscaler/AI-model-maker spend, (ii) customer concentration risk, and (iii) whether NVIDIA’s “full-stack” edge persists as workloads tilt toward inference and custom silicon. ["revenue could be adversely affected"](0001045810-26-000021|0) ["revenue is concentrated"](0001045810-26-000021|0) ["addressing both training and inferencing"](0001045810-26-000021|0) --- ## 1) The tell in the 10-K: OpenAI is framed as upside—explicitly not “in the bag” Buried in the investments discussion, NVIDIA states it is “finalizing an investment and partnership agreement with OpenAI,” but then immediately caveats that there is “no assurance” it will enter into an agreement or that any transaction will be completed. That language matters because it converts what many investors mentally model as *strategic inevitability* into *contingent optionality*—i.e., potential upside with real execution/timing risk. ["finalizing an investment"](0001045810-26-000021|0) ["no assurance"](0001045810-26-000021|0) **Alpha between the lines:** the market is increasingly pricing AI demand as “known,” so *incremental* valuation sensitivity migrates to the *uncertain nodes* (partnerships, customer pacing, and mix)—exactly where NVIDIA is choosing to insert explicit caution. ["no assurance"](0001045810-26-000021|0) --- ## 2) Concentration is the real macro: two direct customers = 36% of revenue NVIDIA discloses that in fiscal year 2026, one direct customer represented 22% of total revenue and another represented 14%—primarily attributable to the Compute & Networking segment. In practical terms, the company can deliver extraordinary aggregate results and still see the stock trade poorly if investors fear even a modest “air pocket” from a small set of buyers (purchase orders can be “cancel[ed], change[d], or delay[ed]… with little notice”). ["22% of total revenue"](0001045810-26-000021|0) ["represented 14%"](0001045810-26-000021|0) ["cancel, change, or delay"](0001045810-26-000021|0) **Alpha between the lines:** this is why the market’s attention snaps to signals about hyperscaler free cash flow discipline, internal chips, and pacing—because concentration makes forward demand *lumpier* than the headline growth rate implies. ["revenue is concentrated"](0001045810-26-000021|0) --- ## 3) Guidance is strong—but it quietly de-risks (and caps) one pillar: China compute In its latest earnings materials, NVIDIA guides Q1 FY2027 revenue to $78.0 billion (±2%) and explicitly says it is “not assuming any Data Center compute revenue from China” in that outlook. That is simultaneously (i) a credibility/visibility move (don’t rely on what may be constrained) and (ii) an admission that a prior source of volume is being treated as non-base-case. ["Revenue is expected to be $78.0 billion"](0001045810-26-000019|1) ["not assuming any Data Center compute revenue from China"](0001045810-26-000019|1) **Alpha between the lines:** the “beat” can be real, but the *composition* of the forward curve is getting more conservative—so the equity reaction becomes more about *what’s excluded* than what’s included. ["not assuming"](0001045810-26-000019|1) --- ## 4) The durability question: NVIDIA is investing hard, and it’s signaling confidence with commitments NVIDIA reports it spent $6.1 billion on capital expenditures in FY2026 (vs. $3.4 billion in FY2025) and expects to increase capex again in FY2027. Separately, management commentary notes inventory at $21.4 billion and “total supply-related commitments” of $95.2 billion, emphasizing it has “strategically secured inventory and capacity to meet demand beyond the next several quarters.” Those are not the disclosures of a management team bracing for an immediate cliff; they read like a company planning for continued scale—while accepting that investors may worry about the cash flow implications of running that hard. ["spent $6.1 billion"](0001045810-26-000021|0) ["increase capital expenditures"](0001045810-26-000021|0) ["Inventory was $21.4 billion"](0001045810-26-000019|2) ["total supply-related commitments were $95.2 billion"](0001045810-26-000019|2) ["secured inventory and capacity"](0001045810-26-000019|2) **Alpha between the lines:** this is the pivot point for the next leg—if demand persists, these commitments can widen the moat (availability, speed-to-deploy). If demand normalizes, they become the raw material for gross margin and working capital anxiety. ["secured inventory and capacity"](0001045810-26-000019|2) --- ## 5) Why the stock can drop on “blowout” numbers: the market is now underwriting *moat maintenance* NVIDIA is explicitly positioning around “both training and inferencing,” and highlights software (e.g., “Dynamo inference software”) plus architectural cadence as part of its edge. The subtext: investors are not just buying GPUs—they’re buying an ecosystem that must stay sticky as the industry experiments with alternatives (custom silicon, integration partners, shifting workload mix). So a selloff post-beat is less paradox than it looks: the market is demanding evidence that NVIDIA’s advantage persists *as inference becomes the dominant unit of compute economics*. ["training and inferencing"](0001045810-26-000021|0) ["Dynamo inference software"](0001045810-26-000021|0) --- If you want, I can turn this into a one-page “hidden alpha” tear sheet (Key debate / What’s priced in / What could surprise) using only the same 10‑K + 8‑K exhibits. ["10-K"](0001045810-26-000021|0) ["EX-99.1"](0001045810-26-000019|1) ["EX-99.2"](0001045810-26-000019|2)