GitHub - sammysltd/euromesh: A sourced model and short report: can Europe train a sovereign frontier AI model on the public compute it already owns, while gigawatt datacenters wait years for grid power? · GitHub
/" data-turbo-transient="true" />
Skip to content
Search or jump to...
Search code, repositories, users, issues, pull requests...
-->
Search
Clear
Search syntax tips
Provide feedback
--><br>We read every piece of feedback, and take your input very seriously.
Include my email address so I can be contacted
Cancel
Submit feedback
Saved searches
Use saved searches to filter your results more quickly
-->
Name
Query
To see all available qualifiers, see our documentation.
Cancel
Create saved search
Sign in
/;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up
Appearance settings
Resetting focus
You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.
Dismiss alert
{{ message }}
sammysltd
euromesh
Public
Notifications<br>You must be signed in to change notification settings
Fork
Star
master
BranchesTags
Go to file
CodeOpen more actions menu
Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit
History<br>1 Commit<br>1 Commit
model
model
paper
paper
.gitignore
.gitignore
README.md
README.md
requirements.txt
requirements.txt
View all files
Repository files navigation
EuroMesh
A sourced model and short report on a single question:
Can Europe stand up a sovereign frontier-class AI model now, by federating<br>the public compute it already owns, while the gigawatt datacenters it is<br>planning take years to connect to the grid?
The answer the model gives is yes, as a stopgap. Europe already operates tens of<br>exaflops of public AI compute across the EuroHPC supercomputers and the national<br>AI Factories. A 1 GW campus, by contrast, waits a mean of 7.6 years for grid<br>power. Federated with low-communication (DiLoCo-style) training, the compute<br>Europe already has can deliver a frontier-class model around 2028, against around<br>2033 for a new gigawatt campus.
Read this first
The report is paper/compute-at-home.pdf (built<br>from paper/compute-at-home.md). It is a short,<br>sourced read aimed at a general audience. Title: "Do We Need OpenAI or Anthropic?<br>Europe Has Tens of Exaflops at Home."
What is in the repo
euromesh/<br>├── README.md<br>├── requirements.txt<br>├── paper/<br>│ ├── compute-at-home.md / .pdf the report<br>│ ├── grid_queue_dataset.md sourced 1 GW vs 40 MW grid-connection lead times<br>│ ├── eurohpc_substrate.md sourced EU public-compute inventory + "is it enough" math<br>│ ├── build_pdf.sh, _report.typ PDF build (pandoc + typst)<br>│ └── figures/ generated charts (PNG + SVG)<br>└── model/<br>├── MODEL_SPEC.md the model specification (equations, params, invariants)<br>├── RESULTS.md full results, scenarios, sensitivity, caveats<br>├── run.py regenerates every CSV and figure<br>├── src/ the three-layer model (efficiency, ramp, regions)<br>├── params/ hardware.yaml, training.yaml, regions.csv + SOURCES<br>├── results/ generated CSVs (do not hand-edit)<br>└── tests/ pytest suite (52 tests) + invariant self-checks
The model in one paragraph
Three layers. Layer 1 is the per-FLOP efficiency of low-communication training<br>(how much the DiLoCo penalty costs). Layer 2 is time-to-availability (when sites<br>energize and how fast cumulative compute accrues). Layer 3 is a per-region<br>scorecard on time, cost, carbon, and feasibility. The headline result is set<br>almost entirely by Layer 2: it reduces to one inequality, the federation wins if<br>its sites are online before a gigawatt campus is. The training efficiency penalty<br>is second-order, confirmed by the sensitivity tornado.
Run it
python3 -m venv .venv<br>.venv/bin/pip install -r requirements.txt<br>.venv/bin/python -m model.run # regenerates all CSVs in model/results and figures in paper/figures<br>.venv/bin/python -m pytest model/tests/ # 52 passed<br>bash paper/build_pdf.sh # rebuilds paper/compute-at-home.pdf (needs pandoc + typst)
The run is reproducible from a clean tree: deleting every output and re-running<br>exits 0 and regenerates everything.
Data and sources
Grid-connection lead times: paper/grid_queue_dataset.md, seven regions,<br>per-region primary sources, anchored by the AWS "up to seven years" statement<br>and the IEA 2-to-10-year range, with limitations stated.
EU public compute: paper/eurohpc_substrate.md, the EuroHPC flagships and<br>the 19 AI Factories, accelerator counts and the training-time math.
Model parameters: model/params/SOURCES.md and<br>model/params/SOURCES_hardware_training.md, with confidence tags.
Honest caveats
The point of this repo is clarity, not novelty. The thesis rests on grid-queue<br>lead times, which are sourced central estimates rather than observed figures (no<br>European operator has yet energized a 1 GW point load). The compute is owned but<br>not yet usable for one...