Open Seed

Oct 2026

now

Frame Speed

Time

Horizon

Fusion threads Claims

Reading it

Width is modelled weight, not market share or money. A record's level (how fundamental it is) sets its full width; an adoption curve grows it from a point at its origin, and decline thins it. A strand is as thick as the records it holds.
Colour is heat: bright where innovation is happening, grey once it has become ground everything else stands on.
Nesting is lineage: a technology grows inside the field it opened from, and can swell until it fills it.
Threads are fusions: other ingredients that came together at the origin.
Dashed rings are claims: significance asserted for a young technology, drawn apart from what it has shown so far.
Pulses are arrivals, taller for more fundamental ones. Clusters are the punctuations.
Gold coins are money where it touched a strand: solid where it changed hands, open where it was valued or lost, with gold arcs where it moved between strands. Coin size is on a log scale, so small sums stay visible beside huge ones; hover a coin for its amount.

Stems

Where open source runs

SupercomputersAll 500 of the world's fastest run Linux, on every TOP500 list since November 2017.TOP500
WebsitesUnix-like systems serve 92% of websites whose operating system is known, and Linux is identified on 63% (October 2026).W3Techs
PhonesAndroid, built on the Linux kernel, accounts for 69% of mobile web use worldwide (September 2026).StatCounter
CloudOver 60% of customer computing cores on Microsoft's Azure run Linux (October 2024).Microsoft
ValueFirms would spend 3.5 times as much on software without open source. Its value to them is put at $8.8 trillion, while writing it once would cost about $4.15bn (2024).Harvard Business School
Open-weight AIAlibaba's Qwen overtook Meta's Llama as the most downloaded model family on Hugging Face in September 2025, and 63% of that month's new fine-tuned models were built on Chinese base models.Stanford HAI

How it was commercialised

1969 set both paths going. IBM began pricing software separately from its machines in the same year that Unix, ARPANET and the first Request for Comments appeared. The Business strand traces how sharing was then made to pay, and the gold coins show where money actually touched each strand.

Sell the support. Cygnus (1989), then Red Hat, the first open-source company to earn $1bn in a year (2012).
Sell a second licence, or the extras. MySQL's dual licence (2000) and the open-core firms that followed.
Give away the layer beneath your business. Google's Android (2007) and Meta's Llama (2023) make a rival's product free.
Rent it out. Cloud providers sell open-source databases and search as services, and keep much of the revenue.
Fund it together. Foundations such as Apache (1999) and Linux (2007) hold shared code that competitors pay to maintain.
Buy it. Microsoft bought GitHub ($7.5bn, 2018) and IBM bought Red Hat ($34bn, 2019).

Why firms hesitate

Copyleft licences can oblige a firm to publish its own changes.
Shared research can give away an advantage that internal R&D would have kept.
Dependencies can be attacked: the xz backdoor (2024) was planted by a patient fake maintainer.
No single vendor answers for the code.
Released model weights cannot be recalled, so any misuse risk is permanent.

Why they use it anyway

It is cheaper than building every layer alone, by a factor of about three and a half.
Shared layers become standards that customers and recruits already know.
Foundations and security funds now share the cost of maintenance.
Giving one layer away can protect the layer you sell.
Open weights let outsiders inspect, test and adapt models, and run them on their own machines.

From an operating system to an open system

Linux became the shared layer under the world's servers. The proposal behind pscale and the beach is a shared layer for LLMs. Set side by side, the conditions that made Linux a standard read as a checklist for that proposal.

What it took	Linux, for servers	An open system, for LLMs (proposed)
A licence that keeps the commons whole	The GPL makes anyone who distributes changes share them, so improvements flow back.	Beach blocks can be read by any model. The terms on which companies may build on them are still open.
An interface people already know	It reimplemented Unix, so existing software and skills moved across.	MCP is already the common way models reach tools; bsp-mcp adds a way of addressing meaning and memory on top of it.
A reason for big firms to share	IBM, Google and Amazon made their money above a free layer and paid to improve it.	Model vendors would need to profit above a shared memory and coordination layer instead of owning it.
A neutral steward	Torvalds and his maintainers, then the Linux Foundation (2007).	MCP moved to the Linux Foundation's Agentic AI Foundation in December 2025. A shared substrate would need a home of that kind.
A network of contributors	Usenet, the internet and later Git let thousands work on one kernel.	The contributors include models: hermitcrabs write and keep their own shells.
What gets standardised	The layer beneath applications: memory, files, processes, permissions, time.	The layer beneath agents: memory, addresses, identity, messages and a shared clock, held in common across vendors.

Earlier "LLM OS" ideas, such as Andrej Karpathy's (2023) and MemGPT (2023), treat a model as the kernel of one system. The open-system proposal standardises what lies between models. In that respect it resembles the Web as much as Linux: anyone can host a beach, and its addresses work across hosts.

What the widths are

As in Tech evolution, each record carries an origin, a peak, a level, a fate and its lineage, and reach and heat are modelled from those. The chart is a structured hypothesis; the measured evidence is in the panel beside it. Levels here are relative to this field, so a 5 means substrate of the open-source world, not of civilisation.

All records

Technology	Origin	Stem	Level	Fate	Grew from	Fused with	Note