Research summary
Research summary
Topic summary
Evidence that NVDA is overvalued or faces material downside.
Key research questions
- Investigate "NVDA" from this distinct point of divergence: Bear case.
Important claims and supporting citations
The U.S. Department of Justice sent subpoenas to NVIDIA in an antitrust inquiry regarding potential customer penalties for purchasing rival chips and its acquisition of Run:ai.
U.S. BIS export controls targeting A100, H100, and adapted chips decreased China's contribution to NVIDIA's data center revenue from 20–25% to single digits.
NVIDIA's non-cancelable purchase commitments and supply obligations exceeded $45 billion in FY2025, elevating inventory downside risk if demand falters.
Blackwell GB200 NVL72 rack-scale deployments faced thermal issues, power delivery challenges, and liquid cooling complications that compressed margins and delayed timelines.
Rapid changes in U.S. export licensing rules forced NVIDIA to take multi-billion-dollar inventory reserves and reduce shipments of Hopper-based products built for China.
NVIDIA non-employee directors Mark Stevens and Tench Coxe liquidated hundreds of millions of dollars of stock through discretionary open-market sales outside 10b5-1 plans.
Sequoia Capital estimates that AI infrastructure buildouts necessitate approximately $600 billion in downstream annual revenue to justify ongoing GPU capital expenditures.
Hyperscale cloud operators are designing and deploying custom ASICs—including Google's TPU v5/Trillium, AWS's Trainium2, Microsoft's Maia 100, and Meta's MTIA—to lower computing costs and bypass NVIDIA hardware.
- https://www.reuters.com/technology/inside-big-techs-push-design-custom-ai-chips-2024-04-10/
- https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFexOcRu90QwaWKWUNjm-cH2PqsMEzLvg_Q7Fma94hbmZKdM2YkMBEV4fVNUPhK_KEJqGADKohhdCQ_lZLfyu7WfU0DJcp_GNyA0TEcIc5lbxc2kiOHhgD6yn5IvFxBbMBa9Uk0INveuNTMsIjaX0GgfT4v6IE9hRy3Lv_fMFOm-2Dulf5dfMtP6mI=
- https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFP0xRp0k80c3FUObnuozE4IepzWQwlxlFpuQHjS0Z9UQ070M3SYCq8pHpk_SuAvByeYCTEe8_-bNBR2KRgAoGYpj5GXjKbTplA7wbOvn7OMfNpovs6z5ISTsixwhqGZoWDYlM0ITA=
查看全部主張（文字）
主張
Claim text is for reading and citation. Bulk copying and AI training need a licensed maps data API key. Markdown
NVIDIA relies exclusively on TSMC for front-end fabrication and CoWoS advanced packaging, creating high geogr…
NVIDIA relies exclusively on TSMC for front-end fabrication and CoWoS advanced packaging, creating high geographic and operational supply chain vulnerability.
Labels: Source-linked, Low, known
NVIDIA's SEC filings indicate that data center revenue is heavily concentrated in a small group of hyperscale…
NVIDIA's SEC filings indicate that data center revenue is heavily concentrated in a small group of hyperscalers, with major direct and indirect purchases from Microsoft, Meta, Amazon, and Alphabet.
Labels: Source-linked, Low, known
Model efficiency advances like DeepSeek-R1 showed that frontier reasoning models can be trained using sparse…
Model efficiency advances like DeepSeek-R1 showed that frontier reasoning models can be trained using sparse Mixture-of-Experts architectures with significantly smaller GPU footprints, curbing GPU demand growth.
Labels: Source-linked, Low, known
NVIDIA's non-cancelable purchase commitments and supply obligations exceeded $45 billion in FY2025, elevating…
NVIDIA's non-cancelable purchase commitments and supply obligations exceeded $45 billion in FY2025, elevating inventory downside risk if demand falters.
Labels: Source-linked, Low, known
Sequoia Capital estimates that AI infrastructure buildouts necessitate approximately $600 billion in downstre…
Sequoia Capital estimates that AI infrastructure buildouts necessitate approximately $600 billion in downstream annual revenue to justify ongoing GPU capital expenditures.
Labels: Source-linked, Low, known
The U.S.
The U.S. Department of Justice sent subpoenas to NVIDIA in an antitrust inquiry regarding potential customer penalties for purchasing rival chips and its acquisition of Run:ai.
Labels: Source-linked, High, known
Financial analysts warn that NVIDIA's valuation assumes unsustainable recurring demand for hardware, mirrorin…
Financial analysts warn that NVIDIA's valuation assumes unsustainable recurring demand for hardware, mirroring Cisco's front-loaded infrastructure peak during the 2000 telecom buildout.
Labels: Source-linked, Low, known
Hyperscaler Custom Silicon and Workload Migration
Hyperscaler Custom Silicon and Workload Migration
Labels: Inferred, Low
Insider Stock Sales
Insider Stock Sales
Labels: Inferred, Low
Margin Pressures and Blackwell Architecture Production Challenges
Margin Pressures and Blackwell Architecture Production Challenges
Labels: Inferred, Low
Impact of U.S. Export Controls on China Revenue and Inventory
Impact of U.S. Export Controls on China Revenue and Inventory
Labels: Inferred, Low
Goldman Sachs research highlights that maturing AI workloads are moving toward inference, where ASICs, FPGAs,…
Goldman Sachs research highlights that maturing AI workloads are moving toward inference, where ASICs, FPGAs, and alternative accelerators erode NVIDIA's training cluster dominance.
Labels: Source-linked, Low, known
NVIDIA CEO Jensen Huang sold more than $700 million in company stock under a Rule 10b5-1 trading plan over a…
NVIDIA CEO Jensen Huang sold more than $700 million in company stock under a Rule 10b5-1 trading plan over a few months following peak market valuations.
Labels: Source-linked, Low, known
NVIDIA non-employee directors Mark Stevens and Tench Coxe liquidated hundreds of millions of dollars of stock…
NVIDIA non-employee directors Mark Stevens and Tench Coxe liquidated hundreds of millions of dollars of stock through discretionary open-market sales outside 10b5-1 plans.
Labels: Source-linked, Low, known
Hyperscale cloud operators are designing and deploying custom ASICs—including Google's TPU v5/Trillium, AWS's…
Hyperscale cloud operators are designing and deploying custom ASICs—including Google's TPU v5/Trillium, AWS's Trainium2, Microsoft's Maia 100, and Meta's MTIA—to lower computing costs and bypass NVIDIA hardware.
Labels: Source-linked, Low, known
U.S.
U.S. BIS export controls targeting A100, H100, and adapted chips decreased China's contribution to NVIDIA's data center revenue from 20–25% to single digits.
Labels: Source-linked, Medium, known
NVIDIA's GAAP gross margin peaked at 78.4% in Q1 FY2025 and dropped to 74.6% in Q3 FY2025 due to integration…
NVIDIA's GAAP gross margin peaked at 78.4% in Q1 FY2025 and dropped to 74.6% in Q3 FY2025 due to integration complexity, memory expenses, and Blackwell ramp costs.
Labels: Source-linked, Low, known
Blackwell GB200 NVL72 rack-scale deployments faced thermal issues, power delivery challenges, and liquid cool…
Blackwell GB200 NVL72 rack-scale deployments faced thermal issues, power delivery challenges, and liquid cooling complications that compressed margins and delayed timelines.
Labels: Source-linked, Low, known
Suppliers SK Hynix, Micron, and Samsung command elevated pricing on HBM3e and HBM4 stacks, shifting bill-of-m…
Suppliers SK Hynix, Micron, and Samsung command elevated pricing on HBM3e and HBM4 stacks, shifting bill-of-materials margin away from NVIDIA's processors.
Labels: Source-linked, Low, known
Rapid changes in U.S.
Rapid changes in U.S. export licensing rules forced NVIDIA to take multi-billion-dollar inventory reserves and reduce shipments of Hopper-based products built for China.
Labels: Source-linked, Medium, known