Check the denominator first

The video title says: “70% of all AI tokens in Russia go to one task.” The conversation itself supports a different statement. Dmitry Yudin estimates that about 70% of tasks on Cloud.ru’s hosted foundation-model service are related to software-development automation.

Two things changed on the way to the title. Tasks became tokens, which changes the metric. The workload of one service expanded to all of Russia, which changes the denominator. The original estimate may be useful; the headline makes a larger claim than the conversation supports.

I do not treat that as a minor editing slip. A percentage without a precise answer to “out of what?” becomes a ready-made argument. It then travels into slides, notes and decisions after the source has fallen away.

The same transcript contains another unresolved check. Yudin says the service has generated about 450 billion tokens since launching in “November last year,” while current usage is about 10–12 billion per month. If “November” means November 2025, reaching 450 billion by July 2026 implies a historical average of roughly 50–56 billion per month — well above the stated current rate.

There may be a reasonable explanation: a different recording date, a sharp change in traffic, a transcription error, or different token definitions. Until the dates and definitions are clear, 450 billion and 10–12 billion per month remain two claims with an open reconciliation gap.

There are stronger, externally checked numbers nearby. None needs a larger denominator:

→ Yakov & Partners and Yandex estimate that AI may add ₽7.9–12.8 trillion per year by 2030, up to 5.5% of forecast GDP; generative AI accounts for ₽1.6–2.7 trillion. That is an economy-wide effect estimate, not AI vendor revenue or market size.

→ In the same study, 71% of surveyed companies use generative AI in at least one function, 86% of generative-AI users fine-tune external open-source models, and 46% deploy or test agents. The study covered 150 CTOs across 16 sectors, 150 vendors and more than 3,500 users. The 86% denominator is companies already using generative AI, not every large company.

→ Cloud.ru’s audited FY2025 summary reports ₽76.5 billion in revenue, up 50%, with AI services and infrastructure contributing ₽41 billion, up 73%. That shows the scale of one vendor, not the size of Russia’s AI market.

Before I copy a percentage into a working note, I check four things: what was measured, across which population, over what period, and whether the wording belongs to the source or its headline.

Checking the denominator is working discipline. It is what keeps an observation about one service from turning into a claim about an entire country.