Competition from Hyperscaler Custom Silicon and In-House ASICs

Bullish investor: NVIDIA's fungible, full-stack architecture remains structurally advantaged over workload-specific ASICs.

• NVIDIA's platform runs training, post-training, and inference on a single architecture, while ASICs are optimized for narrow workloads and risk obsolescence as model architectures evolve rapidly toward mixture-of-experts and reasoning designs.
• Even equity partners exploring custom silicon keep committing to NVIDIA at scale: OpenAI's existing and planned commitments represent approximately 12 gigawatts of NVIDIA compute as of the FY27Q2 call, despite public discussion of its own "Jalapeno" chip.
• NVLink Fusion lets NVIDIA capture value even when customers integrate third-party CPUs or ASICs into their racks, turning a competitive threat into an incremental revenue opportunity.
• CUDA's 7.5 million-developer ecosystem and 20+ years of software investment remain a switching-cost moat that ASIC programs have yet to replicate for broad workload coverage.

Bearish investor: NVIDIA's biggest customers are becoming its biggest long-term competitors.

• Google, Amazon, and Microsoft — hyperscalers that together represent roughly half of Data Center revenue — are all funding internal TPU, Trainium, and Maia programs, with Broadcom and Marvell professionalizing ASIC design capability behind them.
• Even NVIDIA's closest equity partners are hedging: OpenAI has publicly claimed performance advantages for its own Jalapeno chip versus Blackwell, despite NVIDIA's multibillion-dollar investment and infrastructure partnership.
• Inference workloads, where model architectures are more stable, are the segment most vulnerable to ASIC encroachment over time — and inference is becoming a larger share of NVIDIA's overall mix as reasoning and agentic AI scale.

Frontier AI Lab Credit Exposure and "Circular Financing"

Bullish investor: NVIDIA's compute is fungible collateral, not a risky loan book.

• Management argues risk is "limited" because NVIDIA's compute can be redeployed to other customers if any single AI lab underdelivers, unlike a purpose-built ASIC that has no alternative use.
• NVIDIA has structured partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize over $500B of third-party capital, shifting most of the AI lab financing burden away from NVIDIA's own balance sheet.
• The nearly $50B invested in Frontier AI labs as of the FY27Q2 call represents a small fraction of NVIDIA's expected free cash flow over the same period.
• These are framed as ecosystem and equity investments in "once-in-a-generation" companies, several of which management expects to go public and become some of the most valuable technology companies in history.

Bearish investor: Customer concentration and credit exposure have both grown sharply and remain untested.

• Customer concentration jumped materially in FY26: one direct customer represented 22% of total revenue and another 14%, up from a top-three group of just 12%, 11%, and 11% in FY25.
• NVIDIA has committed up to $105B in cumulative payment obligations tied to the SB Energy/OpenAI Portsmouth campus alone, with maximum gross guarantee exposure of $108.5B as of Q2 FY27.
• Management now discloses it has been asked to offer direct financing for customer datacenter buildouts, and expects AI labs where NVIDIA leverages its balance sheet to represent roughly a quarter of business in FY28 — a large and growing revenue base tied to companies not yet cash-flow self-sufficient.
• The "fungibility" defense against circular financing concerns has never been tested through an actual AI demand downturn.

Supply-Constrained Growth vs. the Underlying Demand Gap

Bullish investor: Being supply-constrained in a 100%-demand-growth environment is the best problem a company can have.

• Management guided to approximately 70% revenue growth for FY28 while stating unconstrained demand growth is closer to 100%, meaning every incremental unit of supply converts directly into upside rather than needing to be sold.
• Vera Rubin commenced production shipments with purchase orders already received from every major hyperscaler, AI cloud, and system OEM, suggesting the ramp will again be supply-limited rather than demand-limited.
• Cloud industry backlog now exceeds $2T, and NVIDIA has visibility into $500B of Blackwell and Rubin revenue through the end of calendar 2026, underpinning confidence that any new capacity gets absorbed instantly.
• NVIDIA has locked in $366B of total future purchase commitments as of Q2 FY27, including $279B for supply and capacity, showing it is aggressively working to close the gap between the 70% guided and 100% unconstrained demand growth.

Bearish investor: A perpetually supply-constrained narrative is also a balance sheet and execution risk.

• Locking in $279B of supply and capacity commitments, predominantly memory procurement, creates enormous forward financial exposure regardless of whether end demand fully materializes — echoing the dynamics that preceded prior inventory write-downs.
• Inventory nearly doubled sequentially to $32B in Q2 FY27 ahead of the Vera Rubin launch, meaning NVIDIA is taking on significant balance sheet risk to chase a demand curve that has not been tested through any downturn.
• New risk disclosures acknowledge that customers' own inability to secure sufficient data center capacity or power could independently constrain NVIDIA's growth — meaning execution risk extends beyond NVIDIA's own supply chain.
• Repeatedly disappointing customer orders quarter after quarter could eventually push large buyers to accelerate dual-sourcing or internal chip development out of frustration with allocation constraints.

Gross Margin Trajectory Amid Memory Cost Inflation

Bullish investor: NVIDIA has pricing power to outrun rising input costs.

• Management's stated playbook is to deliver generational performance-per-watt leaps that significantly exceed the rate of component cost inflation, preserving customer value and NVIDIA's own margin capture.
• Non-GAAP gross margin recovered from 61.0% in Q1 FY26 (depressed by the $4.5B H20 charge) to 75.2% in Q4 FY26, demonstrating the company's ability to manage through architecture transitions via cost structure and mix improvements.
• Guidance issued on the FY27Q2 call has margins recovering to 72-73% in FY28 as executed price increases take effect in Q1, showing NVIDIA can pass through memory inflation rather than fully absorb it.
• NVIDIA's long-standing relationships with all three major memory suppliers should help it secure the incremental capacity its roadmap requires as the industry-wide memory crunch eases.

Bearish investor: Margin guidance keeps getting reset downward, and the pattern is recurring.

• On the FY27Q2 call, management cut gross margin guidance, citing memory pricing increases that "exceeded prior expectations," with margins now guided to bottom at 71-72% in Q4 FY27 — below the "mid-70s" target repeatedly promised to investors.
• NVIDIA's own move to a one-year product cadence for Data Center solutions is acknowledged by the company to "magnify the challenges associated with managing supply and demand," raising the odds of recurring transition-related charges.
• History supports this concern: the $4.5B H20 charge in Q1 FY26 and the Q2 FY25 "low-yielding Blackwell material" provisions both show that every major architecture transition has brought at least one quarter of material margin compression.
• Margin outcomes now depend partly on pricing behavior from a concentrated group of just three memory suppliers, a factor largely outside NVIDIA's direct control.

NeoCloud and ACIE Segment: A New, Unproven Financing Model

Bullish investor: The revenue-sharing structure creates a new, high-margin, recurring stream on top of hardware sales.

• ACIE (AI clouds, industrial, enterprise) revenue grew 138% year-over-year and 25% sequentially to $40B in Q2 FY27, evidence that diversification beyond hyperscalers is working at scale.
• NVIDIA's revenue-sharing structure lets it get paid twice — once on the hardware sale and again through a share of rental revenue above a guaranteed floor — layering a usage-linked stream on top of one-time equipment purchases.
• NeoCloud capacity is expected to reach 8 gigawatts by year-end 2026, up from approximately 3 gigawatts at the end of 2025, showing the segment can scale quickly using NVIDIA's own DSX reference designs.
• Management stresses that independent capital still underwrites every NeoCloud deal on its own merits — NVIDIA provides a take-or-pay minimum revenue guarantee, not a loan, which limits direct credit exposure.

Bearish investor: This growth is coming from a harder-to-underwrite, capital-constrained customer base.

• Management itself concedes the ACIE segment is growing off a smaller, harder-to-underwrite base of smaller, capital-constrained operators rather than investment-grade hyperscalers.
• NVIDIA's take-or-pay revenue guarantees to lenders represent a new, unproven exposure to NeoCloud utilization and pricing risk that has never been tested through a downturn in GPU rental rates.
• If NeoCloud token economics deteriorate — for example, from oversupply of rental capacity — NVIDIA could be forced to make good on its revenue guarantees, converting what looked like a hardware sale into a cash outflow.
• The rapid ramp from roughly 3 to a targeted 8 gigawatts of NeoCloud capacity in about a year assumes continued access to external financing for these operators, financing that could dry up quickly if AI infrastructure sentiment turns.

China Market Foreclosure and the Rise of a Rival Ecosystem

Bullish investor: NVIDIA is thriving despite total exclusion from China, with any future access as pure upside.

• NVIDIA still delivered 68% year-over-year Data Center revenue growth in FY26 to $193.7B while assuming zero China compute revenue, showing the rest-of-world AI buildout more than compensates for the lost opportunity.
• Because China revenue is assumed at zero in every forward outlook, any policy change that reopens the market would represent incremental upside rather than something already priced into guidance.
• NVIDIA's global CUDA developer base and ecosystem remain far larger than Huawei's, preserving a long-term platform advantage even while a parallel Chinese ecosystem develops in the interim.

Bearish investor: Every quarter of exclusion further entrenches a capable rival ecosystem.

• NVIDIA estimates the foreclosed China market at approximately $50B annually, and revenue from customers headquartered outside the U.S. fell from 41% of total revenue in FY25 to 31% in FY26 as restrictions tightened.
• NVIDIA itself states that the foreclosure "helped our competitors build larger developer and customer ecosystems to challenge us worldwide," a direct acknowledgment that Huawei's Ascend platform is gaining real scale domestically as a result.
• China's antitrust authority has issued a preliminary finding, tied to Mellanox-related conditions, that NVIDIA's export-control compliance discriminates against Chinese customers — opening a new avenue for fines or further restrictions.
• The longer NVIDIA remains locked out, the more durable and exportable Huawei's ecosystem could become, potentially challenging NVIDIA even in markets outside China over time.