Concentration among covered chip filers (HHI)
Herfindahl index of data-centre revenue between NVIDIA (Data Center segment) and AMD (Data Center segment), per calendar quarter, from filed segment disclosures. Derived metric hhi_by_layer, semis slice.
Reason through thisThe reading
Two filers only, so the level is near the top of the scale by construction; the direction is what carries information.
12 readings drawn from the derived metric hhi_by_layer; 24 observations in the evidence.
Derived (hhi_by_layer) from: SEC EDGAR 10-Q/10-K XBRL instances (segment data). SEC EDGAR, inline XBRL filings (segment disclosures).
The numbers
| as of | reading | value | record |
|---|---|---|---|
| 2026-06-30 | compute semis | 0.87 index | measured |
| 2026-03-31 | compute semis | 0.87 index | measured |
| 2025-12-31 | compute semis | 0.85 index | measured |
| 2025-09-30 | compute semis | 0.86 index | measured |
| 2025-06-30 | compute semis | 0.86 index | measured |
| 2025-03-31 | compute semis | 0.84 index | measured |
| 2024-12-31 | compute semis | 0.82 index | measured |
| 2024-09-30 | compute semis | 0.81 index | measured |
| 2024-06-30 | compute semis | 0.82 index | measured |
| 2024-03-31 | compute semis | 0.83 index | measured |
| 2023-09-30 | compute semis | 0.83 index | measured |
| 2023-06-30 | compute semis | 0.77 index | measured |
Why it matters
Rents stay where concentration stays. AMD's data-centre ramp and custom silicon are the routes by which the chip layer's rent would erode; this is the first of them in filed numbers.
What cuts against it
Broadcom, TSMC, memory and custom silicon are not covered, so the index overstates concentration; segment definitions differ between filers.
How it is measured
- derived metric
hhi_by_layer(formula in the semantic layer) - source SEC EDGAR 10-Q/10-K XBRL instances (segment data) · default tier 4 · Public domain (U.S. government); fair-access policy requires a descriptive User-Agent
- proxy market_structure · cadence quarterly
Four calendar quarters; a move under 200 HHI points is noise. Rising = the covered chip layer concentrating further.
Applied to metric:hhi_by_layer.
Derived rows (12)
| as of | dims | value | inputs |
|---|---|---|---|
| 2026-06-30 | compute_semis | 0.87 index | sources12 |
| 2026-03-31 | compute_semis | 0.87 index | sources12 |
| 2025-12-31 | compute_semis | 0.85 index | sources123+5 more |
| 2025-09-30 | compute_semis | 0.86 index | sources12 |
| 2025-06-30 | compute_semis | 0.86 index | sources12 |
| 2025-03-31 | compute_semis | 0.84 index | sources12 |
| 2024-12-31 | compute_semis | 0.82 index | sources123+5 more |
| 2024-09-30 | compute_semis | 0.81 index | sources12 |
| 2024-06-30 | compute_semis | 0.82 index | sources12 |
| 2024-03-31 | compute_semis | 0.83 index | sources12 |
| 2023-09-30 | compute_semis | 0.83 index | sources12 |
| 2023-06-30 | compute_semis | 0.77 index | sources12 |
Latest readings
- 2026-06-30 · 0.87 indexas of 2026-06-30(2 obs)
- 2026-03-31 · 0.87 indexas of 2026-03-31(2 obs)
- 2025-12-31 · 0.85 indexas of 2025-12-31(8 obs)
- 2025-09-30 · 0.86 indexas of 2025-09-30(2 obs)
- 2025-06-30 · 0.86 indexas of 2025-06-30(2 obs)
- 2025-03-31 · 0.84 indexas of 2025-03-31(2 obs)
Why this status, and its history
AMD's fiscal fourth quarters are now filled from its 10-Ks (they file full-year segment figures under a differently spelled member), so Q4 2025 exists. The four-quarter window now runs from 0.865 (Q2 2025) to 0.870 (Q2 2026), a move of 0.005 inside the 0.02 dead band. The concentrating reading set on 27 Sep compared across the missing quarter and no longer holds.
- 2026-09-27concentrating to stableconf 60 → 60 · Claude (model)
AMD's fiscal fourth quarters are now filled from its 10-Ks (they file full-year segment figures under a differently spelled member), so Q4 2025 exists. The four-quarter window now runs from 0.865 (Q2 2025) to 0.870 (Q2 2026), a move of 0.005 inside the 0.02 dead band. The concentrating reading set on 27 Sep compared across the missing quarter and no longer holds.
Weighed against it: Over two years the index has risen from 0.82 (Q4 2024) to 0.87, so a longer window would read concentrating; the rule's four quarters say stable. Two filers only; Broadcom, TSMC and the clouds' own chips are not in it.
- 2026-09-27stable to concentratingconf 60 → 60 · Claude (model)
Re-read on quarters paired by their midpoint (formula v3), so NVIDIA's May-to-July quarter now sits with AMD's April-to-June one instead of the next quarter. The NVIDIA-vs-AMD data-centre revenue HHI rises from 0.843 (Q1 2025) to 0.870 (Q2 2026), a move of 0.027 against a 0.02 dead band; on the old pairing it was 0.019, just inside. Three of four steps rise; Q4 2025 is missing because AMD files no full-year segment figures.
Weighed against it: The index covers two filers only, so it says nothing about Broadcom, TSMC or the clouds' own chips; Epoch's shipment data shows designers other than NVIDIA gaining share over 2025, so this is concentration against AMD, not across all AI chips.
- 2026-09-27unclear to stableconf 60 → 60 · Claude (model)
With the fourth quarters filled where the filers publish full-year figures, NVIDIA's and AMD's data-centre revenue shares give an index of 0.830 (Q1 2025) and 0.849 (Q2 2026): a move of under two points, inside the dead band, so it reads stable. The earlier unclear reading compared a gappy series; AMD's fourth quarters stay missing because AMD files no full-year segment figures.
Weighed against it: Only two chip makers are covered; Broadcom, TSMC and the custom-silicon designers are absent, so the level overstates concentration.
- 2026-09-10unmeasured to unclearconf — → 60 · the evaluator (automatic)
HHI between NVIDIA Data Center and AMD Data Center revenue by calendar quarter: 0.79 (Sep 2024), 0.86 (Jun 2025), 0.83 (Sep 2025), 0.85 (Jun 2026). Steps exceed the 200-point dead band in both directions, so the evaluator reads unclear: NVIDIA's share oscillates with its product cycle rather than trending. Two filers only; the level is high by construction. Initial seed.
Why this confidence
Rests on 24 observations from SEC EDGAR 10-Q/10-K XBRL instances (segment data); best evidence tier 4, grade A; current.
Confidence is independent of status: 90–95 multiple strong independent sources; 70–89 good evidence, some ambiguity; 50–69 mixed or hard to operationalise; 0–49 limited or vague evidence.
Predictions that lean on this
- Not happening