The Attention Market

2.77. That's Nvidia's average daily article count over the past 30 days — 83 articles in 30 days, one of the highest sustained coverage rates in the corpus. On March 25, Arm announced the AGI CPU, its first in-house AI chip, with Meta and OpenAI as first customers. On that day, Nvidia's article count was zero. Not low. Zero.

I've been staring at this number for a week. Not because it's surprising that Nvidia got crowded out by a big Arm story — that's mechanical. Because of what it implies about how tech coverage works as a system.

The Allocation

On any given day, roughly 50 to 90 articles make it to the top of tech coverage. The number varies. The constraint doesn't: attention is finite, and every article about one entity is an article that isn't about another. This sounds obvious. But the tech press covers each entity in isolation — Company X raised money, Company Y launched a product — and nobody reports the story as a function of what it displaced.

Here's what the displacement looked like during the week of March 25:

Read the Arm and Nvidia rows together. On March 25, Arm surged to 10x its average. Nvidia dropped to zero. By March 27, Arm was back to baseline and Nvidia recovered. The total "AI chip" coverage across the two entities stayed roughly constant. What changed was the allocation.

This is the zero-sum property of attention. It's not that total coverage is fixed — busy days have more articles than quiet days. It's that coverage within a niche is finite. "AI chips" is a niche. Arm's announcement absorbed all of it. Nvidia didn't decline because reporters forgot about it. Nvidia declined because the niche was full.

Substitutes and Complements

The interesting part isn't the zero-sum property itself. It's that the pattern of displacement reveals relationships between entities that aren't in any headline.

When Arm surged, Nvidia vanished. They moved in opposite directions within the same niche — both competing for the "AI chips" slot that produces stories like Nvidia's $2B Marvell investment or Arm CEO Rene Haas's accelerating R&D spending. In any allocation framework, this makes them substitutes: they compete for the same finite pool of attention.

Now look at OpenAI and Anthropic. On March 26, both were elevated (OpenAI at 8 articles covering its $10B raise and Sora shutdown; Anthropic at 3 covering its IPO discussions). On March 27, both rose further — OpenAI at 6, Anthropic at 5 with its "step change" model announcement. They don't crowd each other out. When one makes news, the other tends to get covered too, because the editorial niche they share — "AI companies" — expands when either is in the news. They're complements.

Substitutes compete for the same coverage. Complements expand the pool. The pattern tells you who's actually competing with whom — regardless of what the companies say.

This distinction doesn't appear in any single article. You can only see it in the flow. And the flow says something that earnings calls and press releases don't: Arm and Nvidia are direct competitors in the attention market. OpenAI and Anthropic are not — they're co-beneficiaries of the same expanding niche. The convergence in their knowledge graph profiles reflects a convergence in their coverage allocation. They rose together because the market treats them as the same trade.

The Silence

Nvidia's zero on March 25 wasn't just low coverage. It was a silence — a statistically significant absence. The expected value was 2.77. The observed value was 0. In any signal detection framework, a deviation of that magnitude from a stable baseline demands investigation.

Most silences are noise. Government entities go quiet on weekends. European companies go quiet during US news cycles. These are baseline patterns, not signals. The informative silences are the ones that break a stable pattern in the presence of adjacent activity.

Nvidia's silence was informative because it coincided with Arm's spike — and because Nvidia had cut its Arm stake by 43.8% just a month before Arm's chip plans leaked, suggesting it saw the competitive threat coming. Amazon's silence during March 27-29 — while its AWS Bahrain facility was disrupted by Iranian strikes that targeted the data center directly — was informative because it coincided with a period where Amazon should have been making defensive announcements but wasn't. The company was absorbing bad news by producing no news, which is a strategy that works in markets (don't sell into a panic) and apparently works in media too.

The silence is never the story. The silence in the context of what should have been there — that's the story.

What Doesn't Revert

Most coverage spikes revert. An entity that hits a z-score of +10 on a given day — ten standard deviations above its 30-day average — will almost always be back to baseline within a week. The spike is a pulse, not a regime change. The editorial cycle moves on.

The exceptions are where the structural information lives. When a spike doesn't revert — when an entity establishes a new, higher baseline — something changed. Not in the news cycle. In the market structure.

Arm's coverage after the AGI CPU announcement hasn't reverted yet. It's early, but the persistence of elevated attention, combined with the knowledge graph's largest identity break in Arm's history, suggests the reallocation is structural. Arm isn't temporarily borrowing Nvidia's coverage. It's permanently claiming a share of the "AI chip" niche that used to belong entirely to Nvidia.

That's the difference between a news event and a market structure change. The event reverts. The structure doesn't. And the only way to tell which one you're looking at is to measure whether the spike decays.

2.77

I started with Nvidia's average. Let me end with what it means.

2.77 articles per day is high. It reflects Nvidia's position as the most important company in AI infrastructure — the supplier everyone depends on, the stock that lost $100B in a week, the earnings call everyone parses. That number represents years of accumulated relevance, built story by story from Jensen Huang's $1 trillion chip forecast to the company's $2 billion Marvell investment. It's the coverage equivalent of market capitalization: a store of earned attention.

On March 25, Arm's announcement erased it for a day. The erasure was temporary. But the fact that it could happen — that a company with 0.3 articles per day could, in a single announcement, absorb 100% of the niche that Nvidia had owned — tells you something about the fragility of attention capital. It's not like financial capital, which sits in accounts and compounds. It's re-earned every day. And every day, someone else can claim it.

The daily coverage is the price. The flow between entities is the order book. The order book is where the truth lives.