framework 03 · platform economics

Information Flywheels & Advantages

Information is the one asset that can appreciate while you use it, and almost every enterprise spends it like a consumable.

By · V-Squared AI

Information is the one asset that can appreciate while you use it, and almost every enterprise spends it like a consumable. Every customer interaction, model output, and agent decision is a chance to sharpen the shared models, or a chance thrown away.

Build the flywheel

An information flywheel is the mechanism that keeps your shared models current as the world moves. Customer needs change and the flywheel catches it; the marketplace shifts and the flywheel picks it up. Instead of a model that decays until someone notices, you get one that refreshes itself from the flow of work it already produces. That's what converts a one-time transformation into a compounding one, the engine underneath Transformation Dominance.

A structural advantage

A competitor can copy a model you publish, given enough time. They can't copy the loop that keeps yours improving, because the loop is fed by your customers, operations, and agents doing their work. Information, AI, and agents stop being cost centers and start being the assets that compound your other advantages.

They can copy a model you publish. They can't copy the loop that keeps yours improving.
the evidence · earnings calls, july 2026

Three CEOs made the flywheel argument in three vocabularies.

The framework claims information is the one asset that appreciates while you use it, and that the loop keeping your models current is the part competitors cannot copy. In one earnings cycle that claim went from contrarian to consensus.

MicrosoftFY26 Q4 · Jul 29, 2026

Do not outsource your core IP

Nadella framed Microsoft's goals for the year as amplifying individual agency and helping every organization build its own continuous learning loop without outsourcing its core IP. He described the firm as a learning machine that needs its own learning machine, and talked about companies accumulating human capital and token capital. Models, in his framing, are an input rather than an extraction of the enterprise's knowledge.

what it confirms

This is the flywheel argument stated as vendor strategy by the largest enterprise software company on earth. The significant word is outsource. The risk he names is not buying the wrong model, it is letting the loop that improves your models belong to somebody else.

MetaQ2 2026 · Jul 29, 2026

Scaling laws showed up on proprietary behavioral data

Susan Li reported hitting a first research milestone by continuously pre-training a large-scale model on recommendations data and observing healthy scaling laws in the process. Every public Reels and feed post on Instagram now passes through an LLM. Asked what durable advantage the lab builds, Zuckerberg pointed to the data and the flywheel from watching a community use the product rather than to intelligence itself.

what it confirms

The flywheel does not just keep a model accurate, it produces capability. Scaling laws on proprietary behavioral data mean the loop generates a training asset no competitor can assemble, because it is made of your customers doing their work.

AlphabetQ2 2026 · Jul 2026

The loop is metered, and its confidentiality is the sales pitch

Google reported model APIs processing roughly 22 billion tokens per minute, up from about 16 billion a quarter earlier, with nearly 500 cloud customers each past a trillion tokens over the past year and more than 2,000 enterprises past a hundred billion. Pichai's answer on competitive advantage was that the model is just an ingredient, and that customer data and trajectories stay confidential with nothing flowing back.

what it confirms

Trajectories are the exhaust of the loop: what a system tried, what worked, what got corrected. A hyperscaler now sells the promise that yours stay yours, which is only worth selling because that exhaust is the asset.

Sources: Alphabet Q2 2026, Microsoft FY26 Q4, and Meta Q2 2026 earnings calls and releases, plus Amazon Q2 2026 results reported July 30, 2026. Figures are as stated by company executives on those calls. Amazon reported after market close on July 30, so Amazon figures here come from the release and initial call remarks rather than the full transcript.

the wider evidence · loops audited from outside

A flywheel nobody outside can check is a story. This one got checked.

The framework says the loop is the asset, and competitors can't copy it because your own operations feed it. The strongest test of that claim is a case where an outside party with no stake in the answer verified the output.

peer-reviewedsurvey researchcompany filingsmarket dataindependent audit
Waymocompany filingsMarch 2026 · 220.6M miles

Miles you can't buy, only accumulate

Waymo reported 220.6 million rider only miles driven without a human driver through March 2026, at more than four million miles per week. Against human benchmarks in its operating cities, the company reported roughly 90% fewer crashes involving serious injury or worse, and published its methodology and per city results.

Source: Waymo Safety Impact data hub, figures through March 2026.
what it confirms

Every mile improves the driver and the improved driver earns more miles. The input comes from the service running, not from a purchase order. A competitor can hire the same engineers and buy the same sensors and still be 220 million miles behind.

Insurance Institute for Highway Safetyindependent auditJuly 2026

An outside party checked the output and it held

IIHS researchers independently compared crash rates in San Francisco, Phoenix, Los Angeles, and Austin. They found Waymo vehicles involved in 68% fewer police reportable crashes per mile than human drivers, 85% fewer single vehicle crashes, and 81% fewer injury crashes. The measured advantage ranged from 76% fewer crashes in Phoenix to 35% in San Francisco.

Source: Insurance Institute for Highway Safety study, reported July 2026.
what it confirms

Independent verification is what separates a compounding advantage from a marketing claim, and most enterprise flywheels never get audited. The city spread is the more useful number. The loop is strongest where it has run longest, so the asset is local, cumulative, and tied directly to operating time.

Bloomberg L.P.market datafour decades

Pricing power built on accumulated data, not features

The Bloomberg Terminal has reportedly held around $30,000 per seat per year, with increases the company describes as linked to weighted global inflation, while principal competitors are reported between $14,000 and $22,000. The durable pieces are the accumulated data methodology, the benchmark indices built on it, and the closed messaging network.

Sources: reported subscription pricing and industry analysis, 2025 to 2026. Bloomberg L.P. is privately held and does not publish per seat pricing, so these figures are third party reported rather than company confirmed.
what it confirms

Four decades of accumulated methodology became infrastructure. The counterpoint belongs here too. AI era entrants are attacking that position at a fraction of the price, which is a reminder that the loop protects your data and your network. It never protects the interface you wrapped around them.

Waymo figures come from published safety reporting and the IIHS figures from an independent study. Bloomberg does not publish per seat pricing, so those figures are third party reported and should be treated as directional.