iTero, GIANTX and the Overlooked Question: Who Should Own the AI Tool?

At Gamescom 2026 in Cologne, Natus Vincere lifted the Aegis of Champions and took home USD 1 million from The International 2026's USD 1.6 million prize pool. The roster that year featured Dendi, Puppey, XBOCT, LighTofHeaveN and ArtStyle. Fourteen years later, a writer named Ollie says he still dreams of reliving that moment.

If you opened that interview looking for more about the Aegis, you will be disappointed. The body of the piece is about something else: an artificial-intelligence coaching tool, a European esports organisation, and an exclusivity arrangement. In the entire document I could access, only two section headings are named. One covers the exclusive partnership with GIANTX and the likelihood of the product being copied. The other covers AI-assisted cheating.

Those two headings, plus the author's biography, are close to everything I can verify. No performance data. No sample size. No evaluation methodology. No game version, no schedule, no roster. An interview about a measurement tool that supplies no measurement.

That gap points to a larger structural problem: who gets to own an AI coaching tool inside a closed league, and what happens when that ownership creates an advantage that cannot be relegated away?

One piece of software, one organisation, one league with no trapdoor

Jack Williams is the interview's subject. His name is attached to iTero, an AI-powered coaching platform. Nothing I have describes his exact title, the size of the team, the funding raised, or the customer list. That is a hard stop: any claim about iTero's capability cannot be verified from this source.

GIANTX is the second entity. Based on industry reporting I track, GIANTX was formed through the merger of Excel Esports and Giants Gaming and competes in the League of Legends EMEA Championship, known as the LEC. If that holds, the governing framework for the iTero arrangement is Riot Games' third-party software and competitive-integrity rules. I flag this as an assumption requiring verification, not a fact.

The most important detail about the LEC is not its name. It is its structure. Since 2026, the LEC has operated as a closed franchise with no relegation. In such a system, structural advantages do not get competed away each season. A weak team stays. A strong team keeps its edge. That is the foundation for understanding why a tooling exclusivity deal carries far more weight here than the same deal would in an open circuit.

Patch cadence is the third variable. I track two opposing cycles. Valve updates Dota 2 infrequently and heavily, with large systemic patches followed by long stretches of stability. Riot updates League of Legends on a dense cadence, typically every two weeks. Those two rhythms create two different commercial environments for any analytics vendor.

That is the full extent of the groundwork I can build. What follows is annotated inference, not assertion.

Three windows of intervention, and only one worth arguing about

An AI tool can touch three moments in competitive esports. Pre-match, it processes opponent data, suggests draft bans, and estimates probabilities. In-game, it issues real-time recommendations. Between games of a BO3 or BO5, it synthesises what just happened and proposes adjustments.

The second window is closed in every major title. Real-time assistance is explicitly prohibited, leaving no grey area to debate. If the interview only concerned that window, there would be nothing to write about.

The first window is less controversial than people assume. Pre-match analysis has existed for two decades in the form of spreadsheets, video, and scouting reports. A machine-learning model does faster what an analyst already did by hand. Speed does not create a new category of ethics.

The genuinely contested window is the between-game break. A BO5 contains up to four breaks of a few minutes each. Inside that window, an AI tool can compress hundreds of situations from the previous game into a short adjustment list. The question is not how powerful the tool is, but whether any quantitative threshold exists to separate "coaching support" from "a sixth coach who is not allowed in the room".

I have re-read the rulebooks of several leagues in recent years. None publishes such a threshold. Current rules describe prohibited conduct in qualitative language, for example "unauthorised assistance during competition". A language model can read that sentence several ways. This is a governance hole, not an engineering fault.

What makes the hole dangerous is that it does not stay still. The line between "assisted" and "automated" shifts every time a model's inference capability steps up. A rule written for a five-minute break becomes obsolete when inference time drops to seconds. League operators are trying to fix a boundary on a plane that is sliding.

Patch cadence: the commercial variable nobody puts in the deck

If iTero is title-agnostic, its value proposition inverts between ecosystems.

In Dota 2, a sparse patch rhythm keeps patterns learned from historical data valid for longer. There, the edge belongs to depth of historical modelling. A tool that can answer "what does this team usually do in situation X" with three years of data beats one with three months.

In League of Legends, a biweekly patch cycle shortens the half-life of every learned rule. There, value no longer lies in solving the meta. It lies in detecting the meta's movement faster than opponents. That is a tempo advantage, not a knowledge advantage.

These two environments demand two different product architectures. A product marketed identically to both is a red flag. The silence on patch cadence in the interview is the single largest analytical gap, because patch cadence directly determines the lifespan of the advantage iTero sells to its clients.

To answer that question I need three data points: the patch cycle of the target title, the tournament-server version lock rules, and the data window a vendor is permitted to access. All three are missing.

Exclusivity inside a closed league

In an open circuit, tooling exclusivity is a temporary edge. Weak teams relegate, strong teams change, new organisations arrive with new tools. In a closed league, the same edge becomes a long-term asset.

The iTero-GIANTX arrangement raises three questions the interview does not answer. How long is the exclusivity term? Is exclusivity scoped by region or by title? And what does the termination clause say about data usage rights once the contract ends?

The third question is the heaviest. If GIANTX supplies proprietary data, including scrim data, internal communications, and practice logs, then the real value sits there, not in the model. The vendor can leave the contract with a better-trained model. The organisation loses control of an asset it created.

This is the point I want to underline: the biggest risk in an exclusivity arrangement is not on the vendor's side, it is on the team's side. The seller keeps the knowledge. The buyer keeps only the results for the duration of the contract.

Transfers are a fertile gamble, but I count the cards before I place a bet. In the traditional transfer market, ownership of a player is written into a contract. In the analytics-tool market, ownership of knowledge usually is not. That is the kind of asset that walks out without leaving an invoice.

The moat is not the model

The common assumption is that an AI model is a hard-to-copy asset. That assumption is weakening. Foundation models ship with open weights, fine-tuning techniques are mainstream, and inference costs fall every quarter. A team of three engineers can rebuild the "model" layer of most sports analytics tools in weeks.

If so, the real moat sits in two other layers.

The first is the proprietary data pipeline. Non-public scrim data, in-game communication data, the history of draft decisions made in the war room, none of it appears on any public stats site. Models can be copied. Data relationships cannot.

The second is workflow penetration. A tool that sits in a coach's feed for 18 months carries far higher switching costs than one opened twice a week. Switching cost, not accuracy, is what retains clients.

The fact that the interview devotes a section to copy risk shows the founder knows which layer is easy to breach. Copying a model takes weeks. Copying a data relationship takes years. But if that data relationship rests on a time-limited exclusivity contract, the moat has an expiry date too.

The fragile line between assistance and cheating

The interview's other section covers AI-assisted cheating. This is the topic I believe is being framed incorrectly.

The esports integrity cases adjudicated publicly in recent years mostly fall into match-fixing and in-game cheat software. Both leave clear technical traces: server logs, transaction history, malware signatures. AI-assisted cheating leaves no such trace, because the tool itself breaks no rule. It only reads data and produces text.

In other words, the new cheating layer is not in the software. It is in the timing of use.

I propose a test operators could apply. A tool stays within permitted limits if its conclusions are something a human coach could produce in an equivalent timeframe with the same volume of data. If the tool produces conclusions a human could not reach in that timeframe, it is substituting for a prohibited role.

The test is imperfect. It depends on defining "equivalent timeframe" and "same volume of data". But it moves the debate from sentiment to measurement. And it accepts something many current rulebooks refuse to accept: the boundary is not fixed, so the threshold cannot be fixed either.

Data context

Every analysis above rests on a narrowly limited source, and I am stating those limits so readers can judge for themselves.

First, the timeline. The piece self-anchors to roughly 2026 by simple arithmetic. Natus Vincere first won The International at Gamescom 2026, and the text says "14 years ago". No absolute publication date exists in the material I could access.

Second, content distribution. Of 13 information points at the primary-source level, 10 describe the article's author rather than its subject. That is why I cannot assess anything about patch, tournament, roster, or region.

Third, no performance data. No claim about iTero's accuracy comes with a sample size or methodology. Every number about product effectiveness is unverifiable.

Fourth, no environmental factors. It is unclear whether the interview was conducted remotely or in person, or whether commercial pressure from the partner shaped it.

This is the ceiling of the analysis. Anything beyond it is speculation.

The contrarian angle: AI tools may widen the gap, not close it

The popular story about AI in sport is a democratisation story. Cheap tools, wide access, small teams reaching what only big teams could reach before. I do not buy that version, at least not in the early phase of a tooling cycle.

A tool's value depends on the quality of the data fed into it. The best data in esports is not on public servers. It sits in scrim rooms, in internal voice channels, in the decision history of the coaching staff. Large organisations have more data, more people recording it, and more accumulated seasons.

If so, the same tool installed at two organisations produces two different levels of value. The gap does not narrow. It widens, compounding with input data quality.

That is why I think the public debate is asking the wrong question. "Is AI cheating" is a question about individual conduct. The right question is about resource allocation inside a closed league: when one member holds exclusive access to a competitively material tool, what obligation does the operator have to guarantee equivalent access?

Historically, leagues have had to answer similar questions. Riot Games once tightened restrictions on coach communication during live matches, converting an existing advantage into prohibited conduct. The mechanism is identical here: a preparation advantage gets re-evaluated once it grows large enough to change outcomes.

Operators have two exits. One is mandating equivalent access for all members. The other is restricting the tool within official competition. Both reduce the value of an exclusivity deal. Both are plausible within the next two seasons.

Where my assumptions could be wrong

Let me list where my reasoning could collapse.

One: GIANTX's identity. I assume it is an LEC organisation formed from the Excel Esports and Giants Gaming merger. If this is a different entity under a similar name, my entire governance frame is wrong.

Two: the contested window. I assume the interview's debate centres on the between-game break. If it actually concerns post-match analysis, the ethical problem largely disappears and much of this piece loses its focus.

Three: copyability. I assume the model layer is commodity. If iTero owns a training method or a proprietary dataset nobody can replicate, its moat is far stronger than my assessment suggests.

Four: the link between patch cadence and product value. That is my inference, not a claim from the source. If the tool is designed to be patch-rhythm agnostic, the Dota 2 versus League of Legends distinction I built dissolves.

Five: the timeline. The "14 years" arithmetic may be rounded in the original, shifting the piece by a year in either direction.

Signals for the next cycle

Three signals I will track over the next two seasons.

First, regulatory language. If Riot Games or Valve adds a quantitative threshold for analytical assistance, it signals pressure has grown sufficient to force specific rulemaking.

Second, contract structure. If tooling exclusivity deals begin including clauses on post-expiry data usage rights, the market has matured. If not, organisations are still selling assets without knowing the price.

Third, inference latency. Every time a model's response time drops a tier, the line between permitted and prohibited must be redrawn.

From the Bundesliga to Worlds, I look for the same thing: a truth that repeats. The truth that repeats here may be simply this: every new tool in sport passes through the same sequence. It arrives as a technical edge, gets used as a commercial edge, then gets governed as an integrity problem. The interesting question is not how far iTero will go. The interesting question is how long it takes an organisation to realise that what it is giving away is worth more than what it is getting back.