AI Models Are Becoming Commodities. The Intelligence Feeding Them Won't Be. - Polaris I/O

The next advantage in enterprise AI belongs to whoever knows what changed this morning.

About 250 customers now pay Anthropic more than $10 million a year.

That figure comes from Anthropic’s IPO filing. Reuters reported this week that the company is targeting a valuation of about $1 trillion. The same filing counts more than 1,900 customers above $1 million and more than 11,900 above $100,000. Customers spending $100,000 or more generated about 74% of revenue in the first half of 2026.

Enterprises are wiring AI into how they run the business, and many are paying eight figures to do it.

At the same time, a stranger market is forming. AI labs and data brokers are paying companies for their old Slack messages, emails, and CRM history. Business Insider reported in October that brokers have moved beyond buying archives from failed startups and are now making six-figure offers to operating companies, some with 48-hour deadlines. One broker, micro1, pays midsize companies between $100,000 and $2 million for anonymized workplace data. In August, Google agreed to pay $10 million for bankrupt Spirit Airlines’ internal data.

And this week Disney and Adobe announced a multiyear alliance to build AI capabilities trained on Disney’s own intellectual property.

Put those three stories together and a pattern emerges.

Everyone is racing to own the model. The scarcer asset is what the model knows right now.

What the Slack Data Rush Tells Us

Why would an AI lab pay millions for a company’s Slack history?

Because public internet data is largely exhausted, and models built to do real work need to see how real work happens. Brokers describe the ideal package as the full chain: the client email, the internal debate, the options rejected, the fix chosen, the reply sent. William Almeida, micro1’s chief revenue officer, told Business Insider that without Slack, CRM, or email, the rest of a corpus loses much of its value.

That’s a remarkable admission. The industry is saying context is worth more than volume.

data to contextBut notice what that data is. It’s history. It teaches a model how a sales team handled a renewal in 2023. It cannot tell the model that your largest customer named a new CFO yesterday, that a Tier 2 supplier just missed payroll, or that a competitor opened an office in your best territory last week.

Brilliant Reasoning, Stale Facts

Ask a frontier model whether a major customer is likely to cut spending next quarter. You’ll get a thoughtful answer. It understands churn drivers, contract structures, spending patterns, and retention plays.

Then ask whether it knows that:

  • The customer quietly announced a restructuring yesterday
  • Its CFO changed last month
  • One of its largest customers is in financial distress
  • A competitor has started winning in the market it used to own
  • Its suppliers are showing early signs of operational strain

Any one of those could flip the recommendation. None of them is in the model.

A model can reason brilliantly and still decide badly, because it is reasoning about a world that no longer exists.

Three Layers, One Decision

The enterprise AI stack is settling into three layers.

three layers, one decisionModels that reason. Foundation models, plus specialized models like the ones Disney is building. Enterprises will increasingly choose among several based on cost, performance, and security. That competition is good for buyers and hard on differentiation.

Applications and agents that act. The workflows that turn reasoning into outcomes: a sales agent prioritizing accounts, a procurement agent finding alternate suppliers, a portfolio tool testing whether an investment thesis still holds.

Intelligence that keeps decisions current. The layer that watches the outside world continuously, connects signals across companies, people, suppliers, and markets, decides which changes matter, and hands that context to the people and agents making the call.

The model reasons and the application acts, but only the intelligence layer knows what is true right now.

The first layer gets more crowded every quarter. Durable advantage is forming in the third.

Why This Layer Is Hard to Copy

If several models can reason equally well, what makes one enterprise AI deployment better than another?

Part of the answer is workflow design. Part is proprietary internal data, which is exactly why labs are buying Slack archives. But the hardest piece to replicate is a continuously refreshed picture of the world outside your walls.

That picture compounds. Every day of monitoring adds signals. Every signal makes the next pattern easier to see. A system that has watched your accounts, suppliers, and markets for a year knows things a model trained last quarter never will.

It also avoids the questions now hanging over the Slack market. Privacy advocates warn that even anonymized workplace archives can expose employees, and Slack’s own API terms bar third-party apps from using API data to train large language models. External decision intelligence is built from developments in the market, so it never depends on what employees said in private.

The Proof Is Already in the Numbers

Earlier this year, SBI Growth Advisory and Polaris I/O analyzed more than 58,000 business evolution signals across 46 enterprise accounts over 12 months. Compared with traditional account management on the same accounts, teams acting on those signals:

  • Generated 4x the qualified opportunities
  • Converted 71% of them, versus 20%
  • Closed average deals of $2.6 million, versus $350,000
  • Closed in 144 days, versus 272

The timing finding matters most. Teams that engaged within 30 days of a business change helped shape the requirements. Teams that showed up after 90 days got an RFP and competed on price.

That’s the cost of deciding on stale information, measured in dollars and days.

Where It Shows Up

Revenue. Most commercial signals start outside the CRM: an expansion, a new executive with a technology mandate, a competitor stumbling. The goal is to move from tracking pipeline to seeing the conditions that create it.

Supply chain. Disruption rarely starts with your direct supplier. It starts two or three tiers out, with a distressed manufacturer, an export restriction, or a regional energy shortage. The earlier those dependencies are visible, the more options you have.

Private equity. Diligence is rigorous before close and mostly periodic after it. But the assumptions behind an investment thesis start changing the day the deal closes. Continuous diligence asks, every day, whether those assumptions still hold, long before the change shows up in quarterly results.

Public sector. National security, industrial base resilience, and mission readiness depend on spotting meaningful change early and trusting the signal. The bottleneck is usually making sense of what has already been collected.

Why Transactional AI Needs a Live Feed

As AI moves from answering questions to taking action (approving, prioritizing, buying, or flagging), the cost of stale context rises sharply. A chatbot with outdated facts gives you a wrong answer you can ignore. An agent with outdated facts executes the wrong decision.

recency is the advantageThat is why decision intelligence companies like Polaris I/O will matter more as models get better, not less. Better reasoning on the wrong facts just produces confident mistakes faster.

The roughly 250 customers paying Anthropic eight figures a year are paying for better decisions, and a decision is only as good as what the decision maker knows at the moment of choosing.


The Polaris I/O View

At Polaris I/O, we’re building for a world that never stops changing. We monitor the accounts, suppliers, markets, and environments that matter to our customers, connect individual signals into emerging patterns, and show teams where and when to act.

Powerful models still matter, and their rise makes the intelligence feeding them more valuable.

In business, investing, and national security, the advantage goes to whoever recognizes meaningful change first.

The future of enterprise AI will be decided by the quality of the intelligence informing every decision those models make.

See what matters. Spot what is forming. Steer the response.

Your turn: What decision in your business is being made today on information that’s already out of date?


Sources and further reading


Kunal Mehta
Kunal Mehta

Kunal Mehta is President and Chief Commercial Officer, Technology, at Polaris I/O. Before joining in 2026, he was CEO of TPG Technologies, a Partner in B2B Commercial Excellence at Bain & Company, and an Operating Principal at TCV, where he stepped in twice as CMO for portfolio companies. He has led revenue teams at Infoblox, VMware, Informatica, and HP Software, and serves on the board of Cofense. At Polaris I/O he writes about the signals that show up before a market moves, from defense production to critical supply chains.

All author posts
Related Posts

The decision intelligence platform for what comes next. Nothing gets by you.

Schedule a Demo

Privacy Preference Center