← Case Studies·Cycles·Length-3
CASE C—01 · CYCLE SCORE 3.41 · FIRST CLOSED 2024-Q1

The OpenAI · Microsoft · NVIDIA triangle.

A length-3 cycle in which Microsoft's equity in OpenAI returns as Azure commitments; OpenAI's GPU spend lands at NVIDIA; NVIDIA, in turn, sells its data-center silicon back to Microsoft. Five disclosed edges. One closed loop.

3.41
Cycle score
3
Firms
5
Edges
$13B+
Equity in cycle
~18 mo
Cycle period

Each side of the triangle is a public deal. Read together, the three sides describe a closed loop in which the same dollars can be plausibly traced — through equity, cloud commitments, and data-center capex — back to roughly where they started.

§ 1The structure

Cycle C-01 is the simplest non-trivial circular structure in the sample: a directed triangle of length 3 with one edge of each major flow type. In the canonical direction it reads OpenAI → NVIDIA → Microsoft → OpenAI. In reverse, every edge has a counter-edge, making the underlying loop tightly coupled.

Figure 1 — Schematic of cycle C-01 with edge magnitudes
edge a · GPU spend · $5.2B/yredge b · data-center silicon · ~$8B/yredge c · $13B equityedge d · GPU resale (counter)OpenAIMODEL LABNVIDIASEMICONDUCTORMicrosoftHYPERSCALER
Figure 1. Solid edges form the canonical length-3 cycle. Dashed edges are observed counter-flows that strengthen the corresponding pairwise loop scores L(NVIDIA, Microsoft) and L(OpenAI, Microsoft). Colors indicate flow type: pink for compute, blue for cloud service, green for capital/equity.
Each side of the triangle is a public deal. Read together, they make one closed loop.

§ 2Edge anatomy

The cycle decomposes into three primary edges and two counter-edges. Edge weights are normalized to an estimated annualized run-rate where deal magnitude is undisclosed; raw figures appear in the rightmost column.

EdgeDescriptionTypeWeight wₑMagnitude
OpenAI NVIDIAGPU procurement contracted via cloud passthrough; estimated annualized.compute0.78~$5.2B / yr
NVIDIA MicrosoftData-center silicon shipments to Azure and OpenAI-dedicated capacity.compute0.86~$8.0B / yr
Microsoft OpenAIEquity tranches plus prepaid Azure credits, treated as combined capital flow.equity0.92$13B+ disclosed
Microsoft NVIDIACounter-edge — chip purchase orders for Azure datacenter expansion.compute0.62undisclosed
OpenAI MicrosoftCounter-edge — Azure committed spend tied to OpenAI workload.service0.81multi-year

§ 3Score derivation

Applying equation (2) from § 2.3 of the paper to the canonical 3-cycle:

C(C-01) = 0.78 · 0.86 · 0.92 = 0.617

Length-normalized: C(C-01)¹ᐟ³ ≈ 0.85. The aggregate score reported in the index (3.41) sums this primary cycle with the two counter-edges that close shorter sub-loops L(OpenAI, Microsoft) and L(NVIDIA, Microsoft).

§ 4Timeline

2023-01
Microsoft → OpenAI: $10B equity tranche announced; structured as multi-year capital plus Azure credits.
2023-Q3
OpenAI → NVIDIA: first dedicated H100 cluster procurement, routed through Azure capacity.
2024-Q1
NVIDIA → Microsoft: data-center revenue from MSFT enters NVIDIA top three customers; cycle closes.
2024-Q4
Microsoft → OpenAI: follow-on capital contribution lifts cumulative investment past $13B.
2025-Q2
OpenAI → Microsoft: long-dated Azure committed-spend agreement extends counter-edge through 2030.

§ 5Why it matters

Three implications follow from the structure. First, OpenAI's reported cloud spend is, in part, capital that originated as Microsoft equity — a fact that is orthogonal to whether either firm has done anything improper, but material to how revenue should be attributed. Second, NVIDIA's data-center growth is partly a derivative of capital that Microsoft chose to deploy through OpenAI rather than directly. Third, the three firms' reported figures move together by construction.

Open questions

  1. What share of OpenAI's cloud spend would persist if Microsoft's equity tranches were withdrawn?
  2. How sensitive is NVIDIA's data-center growth to the marginal Microsoft order driven by OpenAI workloads?
  3. Should consolidated revenue measures discount edges that close cycles within a single corporate cluster?

§ 6Notes & sources

Magnitudes are estimates compiled from SEC filings, press releases, and reporting by major outlets through Q1 2026. Where deal value is undisclosed, weights are derived from the proxy method described in § 2.4 of the methodology. Counter-edge weights inherit the procurement-channel proxy.