The Trillion-Dollar AI Question Is Not Adoption. It Is Absorption.

Within a week of each other, two of the most valuable companies in artificial intelligence quietly told the U.S. government they intend to go public.

On June 8, Reuters reported that OpenAI had confidentially filed for a U.S. IPO¹, following Anthropic’s own confidential S-1 a week earlier.² Neither has set final terms or committed to timing. But the signal is unmistakable: the AI race is no longer only a contest for model capability. It has become a contest for public capital, elevating the conversation to increased public scrutiny and public proof.

The numbers strain comprehension. Reuters reported that OpenAI could seek a valuation as high as $1 trillion.¹ Anthropic, which recently raised $65 billion, is reported to be pursuing a listing valued near $965 billion.² These are no longer venture-scale stories. They are infrastructure-scale expectations — the kind of money once reserved for railroads, electrification, and the companies that wire the world.

A valuation is simply the market’s present-day price for future returns. It’s a bet that absorbs accounts for scarcity, narrative, strategic position, access to capital, and the oldest force in markets: the psychological fear of missing the next platform shift. But beneath the spectacle, the question is mundane and unforgiving: will these companies help create enough durable economic value to justify what investors are now willing to pay?

That question will not be answered by model performance or benchmark scores. It will be answered inside organizations where technology becomes output.

A quick note on the audio: the podcast-style version was created specifically for this piece using Google’s NotebookLM. It draws only from the source material behind the article, which makes the result surprisingly thoughtful and grounded. If you haven’t tried NotebookLM before, it’s worth a listen, if only to hear how far these tools have come when the human source material is clear.

0:00

-19:20

And that is where the conversation gets blurry. Companies are already spending on AI. Employees are already using it. Executives are already warning their teams that to sit still is to fall behind. By any ordinary measure, AI has been adopted. So if adoption were the question, the trillion-dollar bet would already be safe.

It isn’t. Which means adoption is the wrong question.

Consider the AI economy as four distinct layers, each harder to reach than the last. Spend is the easy part: subscriptions, licenses, compute, consultants, vendor revenue. Use comes next: prompts, pilots, experiments, activity. Absorption is where it gets difficult — when workflows are redrawn, norms shift, decision rights move, and people actually learn to work alongside the system. Realization is the destination: when all of that finally shows up as something a CFO can see through tangible productivity gains, improved quality, faster speed, lower risk, and better decisions.

Here is what should keep executives and investors awake. Markets can see spend clearly. They can increasingly see use. Absorption is nearly invisible from the outside, and realization — the only layer that justifies the price — is still largely a promise. The market is pricing a layer it cannot yet measure.

The data describe exactly this stall. Stanford’s 2026 AI Index found that 88% of surveyed organizations now use AI in at least one function, and that generative AI reached 53% of the population in three years — faster than the personal computer or the internet at comparable stages.³Adoption, in other words, is not the bottleneck. McKinsey’s 2025 global survey tells the other half of the story: roughly the same share of organizations report using the tools, yet more than 80% say they see no tangible effect on enterprise earnings.⁴

The companies that do see results, McKinsey’s later work found, share one trait — they redesign workflows and pursue genuine transformation rather than simply deploying tools.⁵ RAND’s study of failed AI projects names the same culprits, familiar to anyone who has survived a hard transformation: unclear problem definition, poor fit with how work is actually done, thin data, weak infrastructure, and the reflexive chase after the newest tool instead of the real problem.⁶

That gap between a tool in wide use and the value that never arrives is the entire story. It is the absorption layer: the missing connective tissue between using AI and being changed by it.

Adoption implies the hard part is getting people to pick up the tool. But people have picked it up. Some use it constantly. Some use it quietly, unsure whether their employer approves. Some use it performatively, having been told that fluency is now a credential. None of that is absorption. A tool in a thousand hands is not the same as an organization that has reorganized itself around what the tool makes possible.

Every major cognitive technology has carried this tension. Writing changed memory; the calculator changed arithmetic; search changed recall; GPS changed our sense of direction. Each modified some human habit, and each vastly expanded what people could know, compute, make, and coordinate. The lesson is not that technology improves us on its own. It improves us when the practices around it are rebuilt to match. Writing needed schools, libraries, and citation. The calculator needed a math curriculum that valued reasoning over computation. AI needs its own surrounding redesign, and it needs it in more intimate territory than any tool before it.

AI does not stop at memory or arithmetic, it enters language, judgment, synthesis, expertise, creativity, and — most contentiously — decision-making. In most organizations, those are precisely the places where people locate their professional worth. Which means the anxiety around AI is not a failure of nerve. When leaders experience AI as future productivity and employees experience it as future displacement, the obstacle is no longer technical. It is trust, identity, incentive, and voice, and those forces decide whether people experiment honestly, hide what they are doing, quietly (or, increasingly, openly) resist, or actually learn to work differently.

This is why the human-AI interface is not a user interface. It is the operating relationship between people, machines, and work.

We can already see that relationship working and failing in instructive ways. In a large study of customer-support agents, Erik Brynjolfsson, Danielle Li, and Lindsey Raymond found that access to a generative-AI assistant raised productivity by nearly 14% on average, with the largest gains going to the newest and least-experienced workers.⁷ The mechanism is the interesting part: the system appeared to distribute the tacit knowledge of the best agents to everyone else, compressing the learning curve rather than replacing the worker. AI as apprenticeship, not replacement.

The Harvard Business School–BCG “jagged frontier” study shows the other edge. Across 758 consultants, those using GPT-4 finished more tasks, worked faster, and produced higher-quality work..., so long as the task sat inside the model’s capability frontier. On a task deliberately placed outside it, AI users were more likely to get the answer wrong.⁸ The same tool, in the same hands, lifted performance in one moment and degraded it in the next. The difference was not the technology. It was whether the human knew where its competence ended.

That may be the single most important finding for any executive making a bet on AI. The tool rewards people who know where to trust it, where to test it, and where judgment must stay human and accountable. It punishes those who mistake fluency for accuracy and delegate past its frontier.

Even AI’s most celebrated triumph follows the rule. The 2024 Nobel Prize in Chemistry recognized, in part, Demis Hassabis and John Jumper for AlphaFold, which cracked the decades-old problem of predicting protein structure.⁹ A profound achievement, but not one of a tool floating above human systems. Its value came from being embedded in scientific practice: research communities, validation methods, and the existing, albeit, slow machinery of discovery.

Across every case, the pattern holds. AI creates value where the task boundary is clear, the feedback loop is tight, and the human role is well understood. That is a different problem than access. It is a different challenge than rollout.

The companies that win this phase will not ask, “Where can we deploy AI?” They will ask harder questions. Where does AI actually change the work? Where does it improve a decision, and where does it merely add a new burden of verification? Where does it quietly threaten someone’s status? Where does time saved become better work and where does it just become invisible slack that no one captures?

And one question almost no AI roadmap asks: who inside the organization can make a new behavior legitimate?

That’s what I aim to make abundantly clear through this publication: behavior does not spread evenly through a company. It travels along trust networks. A credible engineer, analyst, nurse, manager, or frontline supervisor can normalize a new way of working faster than any executive mandate. This is not a metaphor; it is one of the most durable findings in the study of how innovations spread, which has long distinguished between authority figures, opinion leaders, change agents, and innovation champions precisely because each does a different kind of social work.¹⁰ The org chart tells you who has power. The trust network tells you where change will actually begin.

That is the human system beneath AI value. And it is why the IPO moment is so revealing.

OpenAI and Anthropic may well be worth extraordinary sums because they sit near the center of a new economic infrastructure. The valuations may be directionally right. They may also be early, fragile, and built on assumptions that public markets are about to test far less politely than private ones did. Both can be true at once.

What the market cannot price — yet! — is the part that decides everything. The next phase of AI will not be settled in model releases, benchmark scores, or capital raises. It will be settled in workflows, review loops, handoffs, incentives, trust networks, and the unglamorous redesign of how people and machines learn to work together.

The market has placed its bet. Now organizations must build the bridge from spend and use to realization. That bridge is absorption. And it is built out of people.

Jason L Zimmerman is the founder and principal of 3Fold Collective, a consulting firm focused on the human systems beneath successful strategy. His work helps leaders understand how trust, influence, behavior, and operating norms shape whether big ideas actually become results, including the rapidly evolving work of AI augmentation.

1. Reuters reported OpenAI’s confidential IPO filing on June 8, 2026, noted that no timeline or terms were disclosed, reported a potential valuation up to $1 trillion, and included OpenAI’s reported user, subscriber, and revenue figures. https://www.reuters.com/technology/openai-files-us-ipo-after-anthropic-ai-giants-head-public-markets-2026-06-08/

2. Anthropic announced on June 1, 2026 that it had confidentially submitted a draft S-1 to the SEC, with share count, price, and timing not yet set. Anthropic’s most recent raise was approximately $65 billion; reporting places a potential public-listing valuation near $965 billion — a reported target, not a settled price. Anthropic / press reporting, June 2026. https://www.anthropic.com/news/confidential-draft-s1-sec;https://seekingalpha.com/news/4601609-openai-confidentially-files-draft-ipo-to-sec-one-week-after-anthropic?share_source=shared_news

3. Stanford HAI, The 2026 AI Index Report (April 2026), economy chapter: organizational AI adoption reached 88%; generative AI is used in at least one business function at 70% of organizations; generative AI reached 53% population-level adoption within three years, faster than the personal computer or the internet. https://hai.stanford.edu/ai-index/2026-ai-index-report

4. McKinsey, “The state of AI: How organizations are rewiring to capture value” (Global Survey, March 12, 2025): 78% of respondents use AI in at least one business function and 71% regularly use generative AI; more than 80% say their organizations are not seeing a tangible enterprise-level EBIT impact from generative AI. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-how-organizations-are-rewiring-to-capture-value

5. McKinsey, “The state of AI in 2025: Agents, innovation, and transformation” (November 2025): AI high performers are markedly more likely to pursue transformation and to redesign workflows — the attribute most associated with realizing enterprise-level EBIT impact. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai

6. RAND Corporation, The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed (2024): leading causes include unclear problem definition, optimizing the wrong metric, poor workflow fit, insufficient data, weak infrastructure, chasing technology over a real user problem, and applying AI beyond its current capability. https://www.rand.org/pubs/research_reports/RRA2680-1.html

7. Erik Brynjolfsson, Danielle Li & Lindsey R. Raymond, “Generative AI at Work,” NBER Working Paper No. 31161 (2023; published in The Quarterly Journal of Economics, 2025): across 5,179 customer-support agents, AI access raised productivity by ~14% on average and by 34% for novice and low-skilled workers, with the system disseminating the best practices of more able workers and moving newer workers down the experience curve. https://www.nber.org/papers/w31161

8. Fabrizio Dell’Acqua, Edward McFowland III, Ethan Mollick, et al., “Navigating the Jagged Technological Frontier,” Harvard Business School Working Paper 24-013 (2023; Organization Science, 2025): in a preregistered field experiment with 758 BCG consultants, those using GPT-4 completed 12.2% more tasks, worked 25.1% faster, and produced 40% higher-quality work on tasks inside the model’s frontier, but were less likely to reach the correct answer on a complex task outside it. https://www.hbs.edu/faculty/Pages/item.aspx?num=64700

9. The Royal Swedish Academy of Sciences, Nobel Prize in Chemistry 2024: awarded with one half to David Baker and the other half jointly to Demis Hassabis and John M. Jumper, recognized in part for AlphaFold2’s solution to protein-structure prediction, now used widely across scientific research. https://www.nobelprize.org/prizes/chemistry/2024/

10. Everett M. Rogers, Diffusion of Innovations (5th ed., Free Press, 2003): the foundational synthesis distinguishing opinion leaders, change agents, and innovation champions as distinct roles through which new practices gain legitimacy and spread within social systems. https://www.simonandschuster.com/books/Diffusion-of-Innovations-5th-Edition/Everett-M-Rogers/9780743258234

Still with me? Excellent. That means either the article worked, or your coffee is stronger than mine. For a related rabbit hole, I’d read The Physics of Organizational Trust and Rapid Transformation next. It gets into the part most transformation decks politely ignore: change doesn’t move because people saw the slide. It moves when the people they already trust make the new way feel safe enough to try.

The Physics of Organizational Trust and Rapid Transformation

·

May 17

At a Glance: Traditional business transformations often fail because leaders mistakenly treat behavioral change as a simple matter of communication rather than a complex social shift. While companies like Tesla succeed through integrated, product-centric structures, legacy organizations struggle to replicate this agility because employees prioritize soc…