One Screen, Three Limits: What Two Weeks Revealed About Concentration
What do a chipmaker’s leadership change, an email outage, a grounded national airspace and an AI assistant that reads messages from people who never signed up have in common?
On the surface, nothing. They’re different industries, different companies, different failures. But seen through a living-systems lens, they start to look like one story, and over the past two weeks the news has been telling it in real time.
Limit one: the person at the center
On November 12, Jensen Huang resigned as CEO of NVIDIA, the supplier at the center of the AI economy; its CFO became interim CEO and the board named its first independent chair (NVIDIA filing). When one person’s departure can shake a company like that, it tells you something about how the system was built. Culture, openness and the way scarce chips were allocated had all depended, to a large degree, on one founder’s style.
That isn’t a judgment of the person. Founders who build something remarkable often build it around themselves. The lesson is structural:
If it can be undone by one resignation, it was never protected.
Protection that lasts is built into structure, through charters, licenses, distributed ownership and mirrors, not into a person.
Limit two: the dependency
When a large cloud provider has a bad week, many organizations discover they can’t email, meet, authenticate or even coordinate a response to the outage itself, because every one of those runs on the same platform.
David Linthicum put it plainly in a piece I’d recommend to any board: the outage wasn’t the story; the dependency was. Many enterprises, he writes, “have allowed convenience to masquerade as architecture.”
The same pattern showed up across very different systems in the same few weeks:
- Cloud and email. Microsoft’s Exchange Online had repeated incidents this year. In early November some users couldn’t send email to Outlook.com and Hotmail addresses for about three days (status record). For many organizations, the same provider also runs their identity, security, documents and, increasingly, their AI.
- National airspace. On November 13 the FAA ordered a nationwide ground stop after a communications outage affected all 22 regional air-traffic-control centers, delaying or cancelling more than 11,000 flights (ABC News). Two months earlier, a regional outage revealed that the backup line had already been cut by a construction crew: the backup shared the same physical path (Travel Noire).
- Satellites and launch. SpaceX’s own prospectus says it “may prioritize our own launch payloads” over government and third-party customers, and its rockets are reported booked years ahead, crowding out companies that depend on them (Reuters, via Bonik Barta).
- AI agents. Meta’s new assistant, Muse, can act across a user’s email, messages and accounts. Independent journalists and researchers have reported privacy problems, including that people who simply correspond with a user are exposed without ever signing up (Cybernews; Tom’s Hardware).
- The open-model commons. NVIDIA agreed to acquire Hugging Face, the largest library of open AI models, for about $12.9 billion (Semafor). Open licenses keep models portable, but whoever owns the library sets the defaults.
Different sectors, same shape: one control plane, one large blast radius, a narrow exit.
His remedy is one I’d call the barbell in boardroom language: selective independence. Keep the large provider where it adds real value. Build independent paths for the things that keep the business alive.
Limit three: physics
The third limit can’t be restated, renamed or recast.
Chips can be bought, buildings finished and racks installed in a year or so. Firm power takes much longer. Gas turbines have multi-year lead times, the handful of manufacturers can only build so many, and the fuel has to come from somewhere. An analysis I found useful, “Does Not Compute”, lays out the numbers: announced data-center capacity is running well ahead of the equipment and fuel that could power it.
Accounting can change the map. Power, water, land and community consent can’t. When reported sales and physical capacity drift apart, sooner or later the physical world has the last word.
Labels vs. structure
For years, frameworks like ESG and “impact” tried to steer capital by scoring disclosures and labels. In my experience they added cost and paperwork without changing the structures that actually decide outcomes.
A structural lens works differently. It doesn’t ask what a company says about itself; it looks at how the company is built. That’s why it works just as well on cloud platforms, chip supply and national infrastructure, where the labels never reached.
Labels measure what you say. Structure decides what happens.
The residual: what the edge can grow
Here’s the hopeful part. When a concentrated center strains, it releases things: engineers who are free to move, tools that are cheap enough for small firms, and ideas that never fit inside the center.
Where they go matters. They can be absorbed by another center, captured by the next wave of the same kind of funding, or become the builders of something more coherent:
- small, portable tools that keep a business running when the giant has a bad day;
- open, interchangeable components in place of one company’s stack;
- many independently owned nodes that coordinate without being captured.
Funding is shifting too. Hardware-heavy AI is moving from venture capital toward infrastructure finance, because physical systems need patient, asset-backed money. The open question is whether the small-node version of that patient capital appears: regional, cooperative and engineer-led.
A practical rule I keep coming back to:
Pay for convenience. Never pay for captivity.
Questions to bring to your own advisers
- If our main platform provider has a bad week, what still works?
- Which of our holdings depend on the same few providers, and on the same few people?
Where the lens comes from
This isn’t a new compliance checklist. It comes from how living systems stay healthy. Forests, immune systems and coral reefs survive shocks because they’re built from many semi-independent parts, connected by membranes rather than walls, with no single point whose failure takes down the whole. They make and repair themselves from the inside.
At Pythia Capital I call this lens Living Systems Intelligence (LSI). It looks at a company, a technology or an investment the way ecology looks at a living system, and it sorts what is coherent from what only looks it, often years before markets notice.
The coherent design is rarely “small instead of big.” It’s a barbell: a protective center where scale truly helps, and a generative edge of many independent nodes that coordinate without being captured.
Where this fits
This note continues threads from The Barbell and the Map (April 2026) and New Investment Models Needed, Revisited (November 2026). It is summarized from detailed, dated reports available to members of The Pythia Scrolls. Builders working on the coherent edge can find a home in the Edge of Chaos Collective, and the free Coherence Lab Letter carries new garden notes as they’re published.
Enjoy this? Get the Coherence Lab Letter delivered to your inbox → subscribe
Educational only; not investment advice. Company examples illustrate structure, not recommendations. Drafted with Claude (an Anthropic model) as research assistant and editor.