lliberopoulou commited on
Commit
cd3216c
·
verified ·
1 Parent(s): 1ea1ad0

Created the Compute Watch README

Browse files
Files changed (1) hide show
  1. README.md +105 -0
README.md CHANGED
@@ -1,3 +1,108 @@
1
  ---
2
  license: cc-by-sa-4.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc-by-sa-4.0
3
+ language:
4
+ - en
5
+ tags:
6
+ - AI industry analysis
7
+ - structural analysis
8
+ - industrial economics
9
+ - AI infrastructure
10
+ - GPU rental prices
11
+ - compute financing
12
+ - chip-backed lending
13
+ - depreciation
14
+ - time series
15
+ task_categories:
16
+ - time-series-forecasting
17
+ size_categories:
18
+ - n<1K
19
+ configs:
20
+ - config_name: observations
21
+ data_files: observations.csv
22
+ default: true
23
+ - config_name: ladder
24
+ data_files: ladder.csv
25
+ - config_name: events
26
+ data_files: events.csv
27
+ - config_name: lifespans
28
+ data_files: lifespans.csv
29
+ - config_name: loans
30
+ data_files: loans.csv
31
+ pretty_name: The Compute Watch
32
  ---
33
+ # The Compute Watch
34
+
35
+ A monthly public record of what AI chips (GPUs) rent for, what lenders charge against them, and how long their owners say they last. Kept by Lida Liberopoulou at [Threadbaire](https://threadbaire.com), independent research on AI economics and infrastructure.
36
+
37
+ Much of the AI buildout is financed with debt, and chip rent is what repays it. Owners' accounts assume a server stays in use for five or six years, while a better chip arrives every year or two. This dataset records, once a month, whether the rent supports that assumption, together with the dated events and reasoning behind each reading.
38
+
39
+ The notes themselves are published at [threadbaire.com/blog/posts/compute-watch.html](https://threadbaire.com/blog/posts/compute-watch.html). This repository holds the data and the structured context, not copies of the notes.
40
+
41
+ ## Current status
42
+
43
+ - **Latest note:** No. 0, The Baseline (6 October 2026). [Read it](https://threadbaire.com/blog/posts/compute-watch-00.html).
44
+ - **Test:** inactive. The shortage is not yet easing (memory index at 100, the baseline).
45
+ - **Premium (B200 ÷ H100):** 2.12× on Silicon Data, 2.99× on Ornn.
46
+ - **Next note:** No. 1, 2 November 2026.
47
+
48
+ *This block is replaced with each note. Earlier states are in `observations.csv` and in this repository's commit history.*
49
+
50
+ ## The prediction, pre-registered
51
+
52
+ Once the chip shortage eases, the newest chip tracked by the planned futures (the B200) loses its rent premium over the older H100, well before the five or six years the owners' accounts assume.
53
+
54
+ The rules, as published in No. 0 on 6 October 2026:
55
+
56
+ - **Shortage easing:** the GDDR6 memory index is at least 15% below its 2 October 2026 baseline for two notes in a row (`gddr6_rebased` ≤ 85).
57
+ - **Once easing, right:** the premium is below 1.6× on Silicon Data for two notes in a row.
58
+ - **Once easing, wrong:** the premium stays above 1.9× on Silicon Data for two notes in a row.
59
+ - **In between:** too early to say.
60
+ - **Until easing:** B200 price moves prove nothing either way.
61
+ - **If not eased by No. 6:** the prediction stays open, and that is stated publicly.
62
+
63
+ The rules do not change. Known confounds (new workloads running on older chips; scarcity of powered data-centre space rather than memory) are recorded in the notes, not used to adjust the test.
64
+
65
+ ## Files
66
+
67
+ | File | One row per | What it holds |
68
+ | ------------------ | ------------------ | ------------------------------------------------------------ |
69
+ | `observations.csv` | note | The headline readings, test status, and what changed in the indexes themselves |
70
+ | `ladder.csv` | chip per note | Each chip generation's rent relative to the H100 |
71
+ | `events.csv` | dated event | What happened in the market around each note, in plain words, with a source |
72
+ | `lifespans.csv` | company per filing | Server lifespans stated in owners' annual reports |
73
+ | `loans.csv` | disclosed deal | Terms of disclosed chip-backed borrowing |
74
+ | `notes/no-XX.md` | note | Structured reasoning: what changed, what it might mean, where sources disagree, test status |
75
+
76
+ ### Column notes
77
+
78
+ - **Premiums and ladder ratios** are computed from index prices recorded on a fixed day (the first working day of the month from No. 1; 2 October 2026 for No. 0). Silicon Data is the headline index; Ornn is the cross-check.
79
+ - **`gddr6_rebased`** is Silicon Data's GDDR6 index set to 100 at the 2 October 2026 baseline. GDDR6 is graphics memory, not the HBM inside data-centre chips; it is used as a public reading of the wider memory shortage.
80
+ - **`index_content_changes`** records changes in what an index contains (a new chip, new providers, a method change), so they are not mistaken for market moves.
81
+ - **`causal_claim`** in `events.csv` is always `none`. Events are recorded because they might later show up in the numbers, not as explanations.
82
+
83
+ ## Method
84
+
85
+ - Both index pages are recorded on the same day each month, and a screenshot is kept for every reading.
86
+ - Specialist-provider prices are used unless marked. The same H100 rents for about 2.6× more at the largest cloud companies.
87
+ - Lifespans and loan terms change only when a new filing or deal is disclosed; rows are added, never edited.
88
+ - Corrections are made in a new commit with a note in the relevant `notes/` file.
89
+
90
+ ## What is not included, and why
91
+
92
+ Raw index prices are not republished here. They are quoted on the published notes under Silicon Data's [Terms of Service](https://portal.silicondata.com/terms-of-service). This dataset contains derived figures (ratios, rebased values), figures from public filings, and the author's own analysis.
93
+
94
+ Reader contributions about what people pay for compute are never included in raw form. Pooled ranges, where they exist, appear only in the published notes.
95
+
96
+ ## Sources
97
+
98
+ The indexes are [Silicon Data GPU Rental Price Indices](https://www.silicondata.com/products/silicon-index) and the [Ornn Compute Price Index](https://data.ornn.com/markets). Every row in `events.csv`, `lifespans.csv` and `loans.csv` carries its own source link.
99
+
100
+ ## Citation
101
+
102
+ Liberopoulou, L. (2026). *The Compute Watch* [Dataset]. Threadbaire. https://huggingface.co/datasets/Threadbaire/compute-watch
103
+
104
+ Licensed CC BY-SA 4.0. Research, not investment advice.
105
+
106
+ Corrections, counter-arguments and questions: [threadbaire.com/contact.html#compute-watch](https://threadbaire.com/contact.html#compute-watch).
107
+
108
+ Related: the full articles, as raw markdown, are in [Threadbaire/research-articles](https://huggingface.co/datasets/Threadbaire/research-articles).