NASHVILLE — Four years into the generative AI (GenAI) era, capital has made itself at home, and it is not being subtle. The exhibit hall of ILTACON 2026 featured a custom-wrapped Lamborghini, a robot mixing coffees, and booths staged to look more like the bridge of the Starship Enterprise than a trade show. The conference also featured three separate country music concerts, Little Big Town, Shania Twain, and Lady A.

Clearly, private equity and technology money have discovered legal services, and they are spending into a market that’s flush with cash from consecutive years of double-digit revenue growth.

Of course, it’s not surprising that the money is loud. What was new at ILTACON this year was what was happening in the practitioner sessions two floors away, where firms have started doing arithmetic.

The receipt is not the cost

The per-seat subscription for AI models is the number firms budgeted for; yet, this is increasingly not the number they will pay.

"We're all now thinking of tokenomics," said James Grice, Head of Innovation and AI at Lawfront Group Limited, a private-equity-backed UK consolidator that has acquired seven regional firms in four years and, with them, seven separate technology stacks. "Do you have multiple MCPs?” he asked, referencing Anthropic’s standard that lets AI models connect to external tools, data sources, and systems through a unified interface. “Are AI models talking in the background to each other, racking up bills you can't see?"

If so, many law firms are realizing they have to multiply those costs by the tool count. Tim Fox, Chief Data and AI Officer at Ogletree Deakins, said he has spent years in legal technology and has never seen the market move like this, with the tool count, as tracked by Legaltech Hub, growing better than 30% year over year. In the conference’s Managing AI Tool Overload session, Christina Natale, Industry Solutions Director at HIKE2, put up a Legaltech Hub slide that listed the AI tool market by category, a wall of logos she warned should come with a trigger warning. The tracked count now runs past 1,100 companies.

When the panel asked how many organizations had 20 or more tools in use or under evaluation, hands went up from about 1 of every 6 attendees. At 10 or more tools, roughly half the room raised their hands. At five or more, nearly everyone did.

Stacks of programs per firm, each with its own consumption meter, increasingly running agents that spend tokens on work no human is directly watching. Needless to say, that adds up.

The $2 PowerPoint

Littler Knowledge Management Counsel Johnny Wong put the arithmetic in the plainest possible terms. "Firms are investing millions in this technology," he said. "But are we getting that money back?" He described checking a consumption bill and finding that a single PowerPoint had cost him $2. For that PowerPoint, he was glad to pay.

However, the $2 PowerPoint is the example you volunteer, it is not the median. The examples that nobody volunteers are the memo regenerated nine times when the second version was already fine, the frontier model summarizing a two-paragraph email, the agent that ran all night against a task nobody opened in the morning. Then, extend that run across an entire organization and Wong landed the stinging punchline. At that rate, he said, it would be cheaper to hire people to do this.

HIKE2's Natale has already seen major companies switch AI services off entirely, because usage was not producing revenue.

Indeed, Fox of Ogletree Deakins, offered the number the industry seems to be converging upon. Most people, Fox said, are settling on AI costing roughly 1% of revenue. His own firm, he added, is not there yet.

Ferraris for emails

Which brings us to the behavior nobody planned for.

Mark Babcock, Manager of Knowledge & AI Solutions at Barnes & Thornburg, described lawyers using “Ferraris to send emails” by reaching for the largest and most expensive model available for the smallest possible task. They enable the premium tier because they want the AI to be right, and they want the client to get the best. "That's cute," Babcock said. “But it isn't right."

The reason is that lawyers cannot see what they are spending or what the model is doing while it drafts. They do not watch a reasoning model double back, re-verify, and cover the same ground before answering — they just see an answer. "For most lawyers, they don't see how tokens are spent," Babcock explained. "They don't have any vision on it."

Unfortunately, Babcock noted, this voracious appetite is real. "Lawyers are hooked on this," he said. "No level of legal technology other than email has this adoption." Firms spent four years persuading a famously change-averse profession to use these tools. Now, it’s critical to teach this newly enthusiastic profession restraint, which may be the much more difficult task. As Babcock observed, you cannot tell a client you ran out of tokens.

The solution that Babcock proposed was the closest thing the conference had to a group therapy suggestion. Everyone in the room, he said, has an opportunity here — just stop. Stop and collectively take a month off.

Stacks of programs per firm, each with its own consumption meter, increasingly running agents that spend tokens on work no human is directly watching. Needless to say, that adds up.

The anxiety under the surface of that suggestion was evident in attendees’ questions. One said tokens terrify her and asked how firms are supposed to cope with agents roaming in the background and talking to each other all day. Taft Innovation Manager Erik Miller's answer was to go back to the use case and ask, are you building an agent because you want to have built an agent that connects to 72 systems because that looks impressive? Or because it eliminates specific line items and saves specific money?

Yet, with the average law firm’s profit and rate growth in exceptionally strong territory,  it is precisely these conditions under which people get comfortable ignoring the meters.

However, Ballard Spahr's Lisa Mayo Haynes gave the optimist's case that the evolution towards tokenization holds promise unto itself. Metering makes AI cost attributable to a specific client in a way that a flat subscription never can. Right now, she explained, there’s no way to do that on a subscription. However, a line item you can trace is a line item you can bill.

Which leaves one central question: Does token spend go the way of e-discovery, becoming a passed-through line item clients simply expect to see?

Whatever the answer, the subscription era may be closing. Law firms bought into AI technology like software and are being billed for it like electricity. The meter has been running for a while, and the recent ILTACON2026 is where the industry may have begun noticing it.

You can find more coverage of ILTACON from this year and past years here‍