Research summary
Research summary
Topic summary
Capital gains from generative AI have concentrated primarily in chipmakers and cloud vendors rather than application software.
Key research questions
- Investigate "Who Actually Benefits From AI?" from this distinct point of divergence: Hyperscaler and hardware capital capture.
Important claims and supporting citations
A September 2025 deal between OpenAI and Nvidia involved a commitment of up to $100 billion conditioned on OpenAI deploying at least 10 gigawatts of Nvidia hardware.
OpenAI's October 2025 agreement with AMD included warrants allowing acquisition of up to 160 million AMD shares (~10% equity) tied to chip procurement milestones.
Internal OpenAI documents reported by The New York Times showed the company lost approximately $5 billion on roughly $3.7 billion in revenue in 2024, driven primarily by inference compute expenses.
Nvidia booked zero Data Center Hopper shipments to China in Q1 FY27 compared to $4.6 billion in the previous year, yet overall Data Center revenue grew 92% YoY.
CoreWeave's March 2025 S-1 registration statement reported 2024 revenue of $1.92 billion, a net loss of $863 million, and roughly 77% customer concentration across two clients, including Microsoft at approximately 62%.
For Q1 FY27, Nvidia reported net income of $58.3 billion on revenue of $81.6 billion, with GAAP gross margin rising to 74.9% from 60.5% YoY.
Nvidia's Q2 FY27 revenue reached $96.2 billion (up 106% YoY), driven by Data Center revenue of $89.0 billion (up 117% YoY).
Hyperscalers accounted for only about 50% of Nvidia's Data Center revenue in Q1 FY27, with the remainder coming from AI clouds, industrial, enterprise, and sovereign customers.
Alle Aussagen anzeigen (Text)
Aussagen
Claim text is for reading and citation. Bulk copying and AI training need a licensed maps data API key. Markdown
SK Hynix reported its High Bandwidth Memory (HBM) production was sold out through 2025 as early as May 2024,…
SK Hynix reported its High Bandwidth Memory (HBM) production was sold out through 2025 as early as May 2024, commanding pricing approximately five times that of standard DRAM.
Labels: Inferred, Low, known
Microsoft, Alphabet, Amazon, and Meta collectively guided to roughly $300 billion or more in 2025 capital exp…
Microsoft, Alphabet, Amazon, and Meta collectively guided to roughly $300 billion or more in 2025 capital expenditures, primarily driven by AI data center buildouts.
Labels: Source-linked, Low, known
Rental pricing on spot markets for Nvidia H100 GPUs dropped from roughly $8 per hour in 2023 to under $2–$3 p…
Rental pricing on spot markets for Nvidia H100 GPUs dropped from roughly $8 per hour in 2023 to under $2–$3 per hour by late 2024 or early 2025.
Labels: Source-linked, Low, known
Nvidia's board authorized an additional $80.0 billion share repurchase and increased its quarterly dividend f…
Nvidia's board authorized an additional $80.0 billion share repurchase and increased its quarterly dividend from $0.01 to $0.25 per share on May 18, 2026.
Labels: Source-linked, Low, known
Nvidia reported fiscal 2026 revenue of $215.9 billion, up 65% year-over-year, with GAAP net income of $120.07…
Nvidia reported fiscal 2026 revenue of $215.9 billion, up 65% year-over-year, with GAAP net income of $120.07 billion and a GAAP operating margin above 60%.
Labels: Source-linked, Low, known
TSMC 2024 Revenue and Packaging Capacity
TSMC 2024 Revenue and Packaging Capacity
Labels: Inferred, Low
CoreWeave Financials and Debt Structure
CoreWeave Financials and Debt Structure
Labels: Inferred, Low
Nvidia Q1 FY27 Financial Results
Nvidia Q1 FY27 Financial Results
Labels: Inferred, Low
TSMC 2026 Operational and Financial Performance
TSMC 2026 Operational and Financial Performance
Labels: Inferred, Low
AI Software Margins and Infrastructure Costs
AI Software Margins and Infrastructure Costs
Labels: Inferred, Low
OpenAI Hardware Partnerships and Procurement Agreements
OpenAI Hardware Partnerships and Procurement Agreements
Labels: Inferred, Low
AI Model Developer Serving Costs and Financial Losses
AI Model Developer Serving Costs and Financial Losses
Labels: Inferred, Low
David Cahn AI Capex and Revenue Gap Analysis
David Cahn AI Capex and Revenue Gap Analysis
Labels: Inferred, Low
Microsoft AI Business Performance and Margins
Microsoft AI Business Performance and Margins
Labels: Inferred, Low
Nvidia Q2 FY27 Financial Results
Nvidia Q2 FY27 Financial Results
Labels: Inferred, Low
For Q1 FY27, Nvidia reported net income of $58.3 billion on revenue of $81.6 billion, with GAAP gross margin…
For Q1 FY27, Nvidia reported net income of $58.3 billion on revenue of $81.6 billion, with GAAP gross margin rising to 74.9% from 60.5% YoY.
Labels: Source-linked, Low, known
TSMC reported Q2 2026 revenue of NT$1,270.38B ($40.20B), a gross margin of 67.7%, operating margin of 60.3%,…
TSMC reported Q2 2026 revenue of NT$1,270.38B ($40.20B), a gross margin of 67.7%, operating margin of 60.3%, and net profit margin of 55.6%, with net income up 77.4% YoY.
Labels: Inferred, Low, known
Nvidia's Q2 FY27 revenue reached $96.2 billion (up 106% YoY), driven by Data Center revenue of $89.0 billion…
Nvidia's Q2 FY27 revenue reached $96.2 billion (up 106% YoY), driven by Data Center revenue of $89.0 billion (up 117% YoY).
Labels: Source-linked, Low, known
Microsoft reported that its AI business exceeded a $13 billion annual revenue run-rate in January 2025, repre…
Microsoft reported that its AI business exceeded a $13 billion annual revenue run-rate in January 2025, representing a 175% year-over-year increase.
Labels: Source-linked, Low, known
Microsoft Vice Chair Brad Smith stated in a January 2025 blog post that Microsoft alone committed approximate…
Microsoft Vice Chair Brad Smith stated in a January 2025 blog post that Microsoft alone committed approximately $80 billion in FY2025 capex for AI data centers.
Labels: Source-linked, Low, known
ICONIQ reported in January 2026 that 84% of scaling-stage B2B AI companies saw gross-margin erosion of six pe…
ICONIQ reported in January 2026 that 84% of scaling-stage B2B AI companies saw gross-margin erosion of six percentage points or more due to AI infrastructure expenses.
Labels: Inferred, Low, known
TSMC's CoWoS advanced packaging capacity was fully booked through 2025 while roughly doubling each year from…
TSMC's CoWoS advanced packaging capacity was fully booked through 2025 while roughly doubling each year from 2023 to 2025.
Labels: Source-linked, Low, known
OpenAI's October 2025 agreement with AMD included warrants allowing acquisition of up to 160 million AMD shar…
OpenAI's October 2025 agreement with AMD included warrants allowing acquisition of up to 160 million AMD shares (~10% equity) tied to chip procurement milestones.
Labels: Source-linked, Low, known
TSMC guided Q4 2026 revenue to $52.3B–$54.3B and raised its full-year 2026 capex guidance to $60–$64 billion.
TSMC guided Q4 2026 revenue to $52.3B–$54.3B and raised its full-year 2026 capex guidance to $60–$64 billion.
Labels: Inferred, Low, known
High-Performance Computing accounted for 68% of TSMC's Q3 2026 revenue, while smartphone revenue fell to 21%,…
High-Performance Computing accounted for 68% of TSMC's Q3 2026 revenue, while smartphone revenue fell to 21%, with quarterly net income hitting a record NT$763.06B.
Labels: Inferred, Low, known
TSMC controls approximately 73% of the global advanced foundry market, with CoWoS packaging capacity fully so…
TSMC controls approximately 73% of the global advanced foundry market, with CoWoS packaging capacity fully sold out on lead times exceeding one year.
Labels: Inferred, Low, known
A September 2025 deal between OpenAI and Nvidia involved a commitment of up to $100 billion conditioned on Op…
A September 2025 deal between OpenAI and Nvidia involved a commitment of up to $100 billion conditioned on OpenAI deploying at least 10 gigawatts of Nvidia hardware.
Labels: Source-linked, Medium, known
Internal OpenAI documents reported by The New York Times showed the company lost approximately $5 billion on…
Internal OpenAI documents reported by The New York Times showed the company lost approximately $5 billion on roughly $3.7 billion in revenue in 2024, driven primarily by inference compute expenses.
Labels: Source-linked, Low, known
David Cahn of Sequoia Capital calculated that AI companies required roughly $600 billion in annual revenue to…
David Cahn of Sequoia Capital calculated that AI companies required roughly $600 billion in annual revenue to justify the capex implied by Nvidia's data-center run-rate, leaving an estimated gap of approximately $500 billion.
Labels: Source-linked, Low, known
Anthropic expended the majority of its 2024 revenue on model serving, resulting in gross margins estimated ne…
Anthropic expended the majority of its 2024 revenue on model serving, resulting in gross margins estimated near or below 50%.
Labels: Inferred, Low, known
CoreWeave's March 2025 S-1 registration statement reported 2024 revenue of $1.92 billion, a net loss of $863…
CoreWeave's March 2025 S-1 registration statement reported 2024 revenue of $1.92 billion, a net loss of $863 million, and roughly 77% customer concentration across two clients, including Microsoft at approximately 62%.
Labels: Source-linked, Low, known
TSMC's monthly revenue in August 2026 hit NT$514.8 billion, a 53.3% YoY increase, bringing Taiwan's August ex…
TSMC's monthly revenue in August 2026 hit NT$514.8 billion, a 53.3% YoY increase, bringing Taiwan's August exports to a record US$82.4 billion.
Labels: Inferred, Low, known
Nvidia's GAAP and non-GAAP gross margins were both reported at 75.0% for Q2 FY27.
Nvidia's GAAP and non-GAAP gross margins were both reported at 75.0% for Q2 FY27.
Labels: Source-linked, Low, known
Average AI product gross margins dropped to 52% (compared to 70–80% for traditional SaaS), with inference cos…
Average AI product gross margins dropped to 52% (compared to 70–80% for traditional SaaS), with inference costs accounting for roughly 23% of total revenue at scaling companies.
Labels: Inferred, Low, known
Hyperscalers accounted for only about 50% of Nvidia's Data Center revenue in Q1 FY27, with the remainder comi…
Hyperscalers accounted for only about 50% of Nvidia's Data Center revenue in Q1 FY27, with the remainder coming from AI clouds, industrial, enterprise, and sovereign customers.
Labels: Source-linked, Low, known
CoreWeave held approximately $8 billion in debt secured directly by its inventory of GPUs, following a debt f…
CoreWeave held approximately $8 billion in debt secured directly by its inventory of GPUs, following a debt financing model initiated with Magnetar Capital in 2023.
Labels: Inferred, Low, known
Nvidia booked zero Data Center Hopper shipments to China in Q1 FY27 compared to $4.6 billion in the previous…
Nvidia booked zero Data Center Hopper shipments to China in Q1 FY27 compared to $4.6 billion in the previous year, yet overall Data Center revenue grew 92% YoY.
Labels: Source-linked, Low, known
Nvidia's Data Center networking revenue reached $14.8 billion in Q1 FY27, representing a 199% year-over-year…
Nvidia's Data Center networking revenue reached $14.8 billion in Q1 FY27, representing a 199% year-over-year increase.
Labels: Source-linked, Low, known
TSMC CFO Wendell Huang attributed a 1.3 percentage point dip in Q3 2026 gross margin (to 66.4%) to the ramp d…
TSMC CFO Wendell Huang attributed a 1.3 percentage point dip in Q3 2026 gross margin (to 66.4%) to the ramp dilution of the 2nm N2 process rather than pricing pressure.
Labels: Inferred, Low, known
Inference costs increased from 20% to 23% of total spend as AI products matured, while talent expenses decrea…
Inference costs increased from 20% to 23% of total spend as AI products matured, while talent expenses decreased from 32% to 26%.
Labels: Inferred, Low, known