The SI 100 and Company Files · Method 1.0
How The SI 100 is made
The SI 100 scores each company on seven public measures, weights them to a total out of 100, and ranks the companies that have data on at least 3 of the seven. This page gives every measure with its weight, source and date, the formula, the tie rule, what is left out on purpose, how a release is frozen, and the company name table that joins the desks.
The measures
Which measures count, and how much
The seven measures of the Q4 2026 release. The weights add up to 100. The date is the day the desk read its source.
| Measure | Weight | Desk and source | Data of | Largest value |
|---|---|---|---|---|
Compute in the ground IT power at the SI data center sites the company owns or uses, in megawatts, as Epoch AI estimates it | 20 | The Power Map and Heat Watch SI data center sites the company owns or uses, with Epoch AI's power estimates. | Nov 10, 2026 | 3.18 GW |
Compute announced Megawatts announced in the Gigawatt Ledger projects that name the company as an operator | 15 | The Gigawatt Ledger Compute projects announced with the company as an operator, and what is confirmed. | Nov 11, 2026 | 7.32 GW |
App reach App Store storefronts where at least one of the company's SI apps is in the Top Free Apps chart | 15 | The SI App Chart The company's SI apps in Apple's Top Free Apps charts. | Nov 19, 2026 | 169 |
Hiring Open jobs on the company's own public job board | 15 | SI Jobs Open jobs on the company's own public job board. | Nov 12, 2026 | 913 |
Lobbying on SI Dollars reported on federal lobbying filings that name SI issues, last four quarters in the snapshot | 15 | The SI Money Map Federal lobbying reports that name SI issues, from LDA.gov. | Nov 17, 2026 | $25.4M |
Federal use SI use cases in the 2025 federal inventory whose vendor field names the company | 10 | The Government SI Register Federal SI use cases whose vendor field names the company, from the 2025 inventory. | Nov 17, 2026 | 131 |
SEC filings 10-K, 10-Q and 8-K filings in the last four full quarters that contain "artificial intelligence" | 10 | SI Exposure SEC filings that contain the searched SI terms, from EDGAR full text search. | Nov 11, 2026 | 12 |
The formula
From seven counts to one score
- Measure score. For each measure, score = 100 x ln(1 + value) / ln(1 + the largest value among all companies with a file). A company with no record on a measure scores 0 on it. The log scale keeps one very large figure from deciding the whole ranking.
- Total. The total is the sum of weight x score over the seven measures, divided by 100, rounded to one decimal. It runs from 0 to 100.
- Who is ranked. A company is ranked when it has a value above zero on at least 3 of the seven measures. Fewer than that is too little data for a rank, so the company is listed as not ranked.
- Ties. Equal totals go to the company with more measures, then the larger compute in the ground, then the name in alphabetical order. Each company gets its own rank.
- Length. If fewer than 100 companies qualify, the list is published at its real length under the same name, with the count stated. It is never padded.
Not counted, on purpose
- Market value, revenue or funding
- Stock or token prices
- Any opinion, survey or vote
- Incidents, recalls, promises and off switches (they are on each Company File, and they are records, not scores)
- Anything a desk does not publish with a source and a date
Overlaps are kept as the desks record them: a site with an owner and a user counts for both, and a project with several operators counts for each.
Releases
Dated, frozen and kept
- Quarterly. One release per calendar quarter, named by the quarter, such as 2026-Q4. The first is 2026-Q4, released Nov 19, 2026.
- Frozen. A release file is written once and never changed. It holds each company's seven values, seven scores, total and rank, the measures with their data dates, and the names of the companies not ranked.
- Checksum. The SHA-256 on the release is taken over the release body as canonical JSON: keys sorted, no spaces, the sha256 field left out.
- Movers. From the second release on, each row shows its move against the release before, and new entries and companies that left are named.
- Between releases. Company Files keep moving with their desks. The ranking does not move until the next release.
Check a release yourself
Download the release, drop the sha256 field, write the rest as canonical JSON and hash it. The result must match the value printed on the ranking page.
python3 - <<EOF
import json, hashlib, urllib.request
r = json.load(urllib.request.urlopen("https://tsio.si/si-100/releases/2026-Q4.json"))
sha = r.pop("sha256")
body = json.dumps(r, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
print(hashlib.sha256(body.encode()).hexdigest() == sha)
EOFThe company name table
127 companies, the names each desk uses
One company files, hires and builds under many names. Each row gives the pattern matched in the owner, operator, user, vendor, developer and maker fields, and the key of the company on each desk that has its own. Lobbying client names are mapped by the Money Map's company map. 3 names in the table have no record on any desk today and no file.
| Company | Names matched | Money Map | Jobs | Ticker | Promise Tracker |
|---|---|---|---|---|---|
OpenAI SI labs | \bOpen ?AI\b | openai | openai | openai | |
Anthropic SI labs | \bAnthropic\b | anthropic | anthropic | anthropic | |
xAI and SpaceXAI SI labs | (?<![A-Za-z])xAI\b|\bSpaceX ?AI\b|\bSpaceXSI\b | xai | spacexsi | xai | |
Mistral AI SI labs | \bMistral(?: AI)?\b | mistral-ai | mistral | ||
DeepSeek SI labs | \bDeepSeek\b | deepseek | |||
Z.ai (Zhipu) SI labs | \bZ\.ai\b|\bZhipu\b | zai | |||
Alibaba SI labs | \bAlibaba\b | alibaba | |||
Scale AI SI labs | \bScale\ AI\b | scale-ai | scale-ai | ||
Perplexity SI labs | \bPerplexity\b | perplexity | |||
Cohere SI labs | \bCohere\b | cohere | cohere | ||
Hugging Face SI labs | \bHugging\ Face\b | hugging-face | |||
Safe Superintelligence SI labs | \bSafe\ Superintelligence\b | safe-superintelligence | |||
Character.AI SI labs | \bCharacter\.?AI\b|\bCharacter Technologies\b | character-ai | |||
Inflection AI SI labs | \bInflection\ AI\b | inflection-ai | inflection-ai | ||
Databricks SI labs | \bDatabricks\b | databricks | databricks | ||
Thinking Machines SI labs | \bThinking\ Machines\b | thinking-machines | |||
Reflection AI SI labs | \bReflection\ AI\b | reflection-ai | |||
Cognition SI labs | \bCognition(?: AI| Labs)?\b | cognition | |||
ElevenLabs SI labs | \bElevenLabs\b | elevenlabs | |||
Luma AI SI labs | \bLuma\ AI\b | luma-ai | |||
Cartesia SI labs | \bCartesia\b | cartesia | |||
Nous Research SI labs | \bNous\ Research\b | nous-research | |||
World Labs SI labs | \bWorld\ Labs\b | world-labs | |||
Poolside SI labs | \bPoolside\b | poolside | |||
Stability AI SI labs | \bStability\ AI\b | stability-ai | |||
Imbue SI labs | \bImbue\b | imbue | |||
Sakana AI SI labs | \bSakana\ AI\b | ||||
Cursor SI labs | \bCursor\b|\bAnysphere\b | ||||
Replit SI labs | \bReplit\b | ||||
NVIDIA Chip makers | \bNVIDIA\b | nvidia | NVDA | ||
AMD Chip makers | \bAMD\b|\bAdvanced Micro Devices\b | amd | AMD | ||
Intel Chip makers | \bIntel\b(?! ?(?:ligen|lect)) | intel | INTC | ||
Qualcomm Chip makers | \bQualcomm\b | qualcomm | QCOM | ||
Micron Technology Chip makers | \bMicron\b | micron | MU | ||
Broadcom Chip makers | \bBroadcom\b | broadcom | AVGO | ||
TSMC Chip makers | \bTSMC\b|\bTaiwan Semiconductor\b | tsmc | |||
Samsung Chip makers | \bSamsung\b | samsung | |||
Arm Chip makers | \bArm (?:Holdings|Limited|Ltd)\b | arm | |||
Marvell Technology Chip makers | \bMarvell\b | marvell | MRVL | ||
Applied Materials Chip makers | \bApplied\ Materials\b | applied-materials | AMAT | ||
Cerebras Chip makers | \bCerebras\b | cerebras | cerebras | ||
Groq Chip makers | \bGroq\b | groq | |||
SK hynix Chip makers | \bSK\ hynix\b | sk-hynix | |||
GlobalFoundries Chip makers | \bGlobalFoundries\b | globalfoundries | |||
Texas Instruments Chip makers | \bTexas\ Instruments\b | texas-instruments | TXN | ||
Etched Chip makers | \bEtched\b | etched | |||
Tenstorrent Chip makers | \bTenstorrent\b | tenstorrent | |||
Graphcore Chip makers | \bGraphcore\b | graphcore | |||
Astera Labs Chip makers | \bAstera\ Labs\b | astera-labs | |||
SambaNova Chip makers | \bSambaNova\b | sambanova | |||
Lightmatter Chip makers | \bLightmatter\b | lightmatter | |||
MatX Chip makers | \bMatX\b | matx | |||
d-Matrix Chip makers | \bd\-Matrix\b | d-matrix | |||
Axelera AI Chip makers | \bAxelera\ AI\b | axelera-ai | |||
Huawei Chip makers | \bHuawei\b | ||||
Microsoft Cloud and big tech | \bMicrosoft\b|\bAzure\b | microsoft | MSFT | microsoft | |
Alphabet and Google Cloud and big tech | \bGoogle\b|\bAlphabet\b|\bDeepMind\b | alphabet-google | GOOGL | google-deepmind | |
Amazon Cloud and big tech | \bAmazon\b|\bAWS\b | amazon | AMZN | amazon | |
Meta Cloud and big tech | \bMeta\b(?! ?data) | meta | META | meta | |
Apple Cloud and big tech | \bApple\b | apple | AAPL | ||
Oracle Cloud and big tech | \bOracle\b | oracle | ORCL | ||
IBM Cloud and big tech | \bIBM\b | ibm | IBM | ||
Salesforce Cloud and big tech | \bSalesforce\b | salesforce | CRM | ||
Palantir Cloud and big tech | \bPalantir\b | palantir | PLTR | ||
CoreWeave Cloud and big tech | \bCoreWeave\b | coreweave | coreweave | ||
Adobe Cloud and big tech | \bAdobe\b | adobe | |||
Cisco Cloud and big tech | \bCisco\b | cisco | CSCO | ||
Dell Technologies Cloud and big tech | \bDell\b | dell | DELL | ||
HP and HPE Cloud and big tech | \bHP Inc\b|\bHewlett Packard\b|\bHPE\b | hp | |||
Super Micro Computer Cloud and big tech | \bSuper ?micro\b | super-micro | |||
Workday Cloud and big tech | \bWorkday\b | workday | |||
ServiceNow Cloud and big tech | \bServiceNow\b | servicenow | |||
Snowflake Cloud and big tech | \bSnowflake\b | snowflake | |||
X Corp. Cloud and big tech | \bX Corp\b|^X$|\bTwitter\b | x-corp | |||
Tesla Cloud and big tech | \bTesla\b | tesla | TSLA | ||
Uber Cloud and big tech | \bUber\b | uber | UBER | ||
ByteDance and TikTok Cloud and big tech | \bByteDance\b|\bTikTok\b | bytedance-tiktok | |||
Intuit Cloud and big tech | \bIntuit\b | intuit | |||
Equinix Cloud and big tech | \bEquinix\b | equinix | |||
Tencent Cloud and big tech | \bTencent\b | ||||
SoftBank Cloud and big tech | \bSoft[Bb]ank\b | ||||
Nebius Compute and data centers | \bNebius\b | nebius | |||
Crusoe Compute and data centers | \bCrusoe\b | crusoe | |||
Fluidstack Compute and data centers | \bFluidstack\b | fluidstack | |||
Lambda Compute and data centers | \bLambda(?: Labs)?\b | lambda | |||
Together AI Compute and data centers | \bTogether\ AI\b | together-ai | |||
Vultr Compute and data centers | \bVultr\b | vultr | |||
TensorWave Compute and data centers | \bTensorWave\b | tensorwave | |||
Runpod Compute and data centers | \bRunpod\b | runpod | |||
San Francisco Compute Compute and data centers | \bSan\ Francisco\ Compute\b | san-francisco-compute | |||
Nscale Compute and data centers | \bNscale\b | ||||
Vantage Data Centers Compute and data centers | \bVantage\ Data\ Centers\b | ||||
Core Scientific Compute and data centers | \bCore\ Scientific\b | ||||
G42 and Core42 Compute and data centers | \bG42\b|\bCore42\b | ||||
TeraWulf Compute and data centers | \bTeraWulf\b | ||||
Applied Digital Compute and data centers | \bApplied\ Digital\b | ||||
Cipher Mining Compute and data centers | \bCipher\ Mining\b | ||||
Related Digital Compute and data centers | \bRelated\ Digital\b | ||||
SB Energy Compute and data centers | \bSB\ Energy\b | ||||
Aker Compute and data centers | \bAker\b | ||||
Chirisa Technology Parks Compute and data centers | \bChirisa\ Technology\ Parks\b | ||||
Machine Investment Group Compute and data centers | \bMachine\ Investment\ Group\b | ||||
EcoDataCenter Compute and data centers | \bEcoDataCenter\b | ||||
BorderPlex Digital Assets Compute and data centers | \bBorderPlex\ Digital\ Assets\b | ||||
STACK Infrastructure Compute and data centers | \bSTACK\ Infrastructure\b | ||||
Firmus Compute and data centers | \bFirmus\b | ||||
Baseten Platforms and tools | \bBaseten\b | baseten | |||
Fireworks AI Platforms and tools | \bFireworks\ AI\b | fireworks-ai | |||
Modal Platforms and tools | \bModal Labs\b | modal | |||
Anyscale Platforms and tools | \bAnyscale\b | anyscale | |||
Pinecone Platforms and tools | \bPinecone\b | pinecone | |||
Physical Intelligence Robotics | \bPhysical\ Intelligence\b | physical-intelligence | |||
Figure Robotics | \bFigure AI\b | figure | |||
1X Robotics | \b1X Technologies\b | 1x | |||
Apptronik Robotics | \bApptronik\b | apptronik | |||
METR Safety and research labs | \bMETR\b | metr | |||
Apollo Research Safety and research labs | \bApollo\ Research\b | apollo-research | |||
Snap Apps and services | \bSnap\b(?! ?shot) | ||||
Slack Apps and services | \bSlack\b | ||||
Zoom Apps and services | \bZoom\b | ||||
Grammarly Apps and services | \bGrammarly\b | ||||
Mozilla Apps and services | \bMozilla\b | ||||
Opera Apps and services | \bOpera\b | ||||
DuckDuckGo Apps and services | \bDuckDuckGo\b | ||||
Pinterest Apps and services | \bPinterest\b | ||||
WordPress.com Apps and services | \bWordPress\.com\b|\bAutomattic\b | ||||
Suno Apps and services | \bSuno\b |
Company Files
How a file is put together
- Read only. A file joins what a desk already publishes. It adds no new claim.
- One block per desk, with the desk's source, the day it was read and a link back.
- No empty blocks. A desk with no record for the company is left out. A company with no record anywhere has no file.
- Promises appear only with the dated promise and the status the Promise Tracker records, each with its own source.
- Facts only. No adjectives about a company, no motives, no sides.
Corrections
A wrong match, a missing name or a record on the wrong file: tell the team on Telegram or X @entersio with the source. A fix to the name table applies to every desk at once and is noted with the next release. See the corrections policy.
