A Conversation about the AI Premium: Market Valuations of Artificial Intelligence Adoption

Speaker A: So if you, uh, if you listen to corporate earnings calls these days, you really get the distinct impression that every single company on Earth is like an artificial intelligence pioneer.

Speaker B: Right. Everybody's an expert suddenly.

Speaker A: Exactly. Everyone's special talent is supposedly, you know, synergizing AI workflows. But, um, if you actually look at the data, the stock market is doing something highly counterintuitive. Oh, absolutely, because right now investors are actively penalizing companies for hiring data analysts and aggressively rewarding them for hiring talkers.

Speaker B: Yeah, the disconnect between, uh, corporate public relations and the actual mechanics of market valuation has really never been wider.

Speaker A: It's massive.

Speaker B: I mean, every press release has those magic letters A and I just slapped onto it, but the market, the market is completely ignoring the noise. It's looking strictly at the bare mechanical reality of value creation.

Speaker A: Which brings us to why we're here today. Welcome to the deep dive. Our mission today is to cut through all of that artificial intelligence hype and, and look at the cold, hard financial reality. We are answering one massive question for you. How is the stock market actually valuing AI adoption right now?

Speaker B: And to answer that, we're looking at the really critical research paper titled the AI Premium How Markets Price Artificial Intelligence Adoption and Its Implications for Organizations. It's by Jonathan H. Westover, PhD, and he bases his analysis on some groundbreaking 2026 research from BORI and colleagues. What makes this paper so, uh, definitive is its methodology. They didn't just distribute surveys to executives and ask, you know, about their AI strategies.

Speaker A: Well, everyone lies on those anyway, right?

Speaker B: Exactly. Instead, they audited actual realized AI consumption.

Speaker A: Okay, let's unpack this. Well, I mean, the scale of the data they audited is difficult to wrap your head around.

Speaker B: It really is.

Speaker A: They analyzed 380 trillion tokens of realized AI usage across more than 400 large language models. And, uh, for anyone unfamiliar, a token is essentially like a piece of a word or a fragment of code that an AI processes.

Speaker B: Right. It's the basic unit of AI thought, essentially.

Speaker A: Yeah. So 380 trillion tokens isn't a projection. It is a real time, undeniable ledger of actual API calls. They tracked every late night prompt, every automated script, and, you know, every data query.

Speaker B: Yeah, they looked at exactly who is using the models, how much data they're pushing through them, and crucially, how equity prices respond to that specific value. Measurable usage.

Speaker A: The ultimate lie detector.

Speaker B: Completely. The market is acting as this, uh, this brutal evolutionary filter. Like a sorting hat. It is rapidly selecting winners and losers. And that selection process isn't just happening at the corporate level.

Speaker A: Right.

Speaker B: It is aggressively filtering individual skills, specific tasks, and entire career paths, which is wild.

Speaker A: And, um, the headline finding of this paper focuses on the survivors of that evolutionary filter. Firms with high verifiable AI exposure earn a return premium of 64 basis points weekly over low exposure firms. Yeah, and just for context, that translates to roughly a 33% annualized premium, which is.

Speaker B: I mean, statistically and economically, that is a formidable divergence in valuation.

Speaker A: But wait, I have to assume some of this is just like statistical noise, right?

Speaker B: That you mean.

Speaker A: Well, couldn't this premium simply be because the companies heavily using AI are already, you know, fake, fast growing tech companies or highly profitable tech giants that would be outperforming the market anyway? It feels a bit like the dot com era, where anything adjacent to a computer got a valuation bump. Like putting.com on a billboard in 1999.

Speaker B: Yeah, that is the most logical assumption to make. And honestly, the researchers anticipated that exact critique. What's fascinating here is that the market is surprisingly discerning.

Speaker A: Okay, how so?

Speaker B: Well, they applied incredibly strict financial risk controls, and using something called the FAMA French five factor model.

Speaker A: Right, the finance stuff.

Speaker B: Yeah. Economists use this model to isolate specific market phenomena by just stripping out the noise. They computationally account for a company's size, overall market trends, profitability, investment patterns, value versus growth metrics.

Speaker A: All of that.

Speaker B: Exactly. So they essentially level the playing field mathematically.

Speaker A: Wow.

Speaker B: Okay, they completely level it. And once you filter out all those traditional reasons, the stock might go up. That AI premium remains intact at 56.3 basis points weekly.

Speaker A: So it barely drops.

Speaker B: Barely drops. It survives the most rigorous stress tests finance academics can apply. It is an isolated, undeniable AI effect. And the mechanism driving it is what economists call systematic transition risk.

Speaker A: Systematic transition risk meaning what exactly?

Speaker B: Meaning investors are looking at AI and realizing, uh, it's going to fundamentally scramble the economy. It will relentlessly reallocate value from incumbents to upstarts and from slow movers to fast adopters.

Speaker A: Ah, I see. Because the market is this unsentimental lie detector. It's demanding a premium for that uncertainty. Investors don't know exactly who the ultimate winners will be, so they're demanding a higher return for taking on the risk of this massive economic transition.

Speaker B: Yeah, they are pricing in the uncertainty of how value will be reallocated. It functions as a risk premium just as much as a growth premium.

Speaker A: So for you listening to this right now, I mean, think about your own company's stock or your private valuation if you're at a startup, is the market seeing your organization as a future winner, navigating this transition, or, uh, as a legacy dinosaur about to have its value reallocated to a competitor?

Speaker B: Because the market is absolutely keeping score and it isn't handing out awards for participation.

Speaker A: Right. And the scoring system is incredibly specific. Because if the market is being discerning, exactly what kind of AI usage is it rewarding?

Speaker B: Well, the researchers didn't just count those 380 trillion tokens, they categorized them. They drew a very sharp line between what they call the extensive margin and the intensive margin.

Speaker A: Extensive versus intensive.

Speaker B: Right. The extensive margin is broad, shallow usage across an organization. The intensive margin is deep, highly complex usage by a very concentrated group.

Speaker A: You know, I was trying to visualize this intensive versus extensive dynamic when I was reading the source material and I kept thinking about corporate wellness programs.

Speaker B: Oh, interesting.

Speaker A: Yeah. Tell me if this tracks relying on the extensive margin. Like buying a basic $20 a month chatgpt account for all 10,000 of your employees. It's like a CEO buying every single employee a, uh, gym membership for New Year's.

Speaker B: Oh, that's a perfect analogy. Oh, because the optics are great.

Speaker A: The optics are fantastic. We have invested in the health of our entire workforce, but in reality, what happens? 90% of the employees go twice in January, walk on the treadmill for 10 minutes and never log in again? Yeah, it looks good in a press release, but it achieves absolutely nothing for the company's bottom line. The market heavily penalizes that kind of casual consumption.

Speaker B: It does what it actually rewards. The intensive margin is the equivalent of having 100 elite athletes training for the Olympics. The market strictly wants to see the Olympians.

Speaker A: The power users.

Speaker B: Yes, the power users. The valuation premium concentrates entirely on users who generate long, complex prompts, who make automated tool calls, and who utilize the closed source premium frontier models, you know, like GPT4, Claude or Gemini, for actual production tasks.

Speaker A: I was looking at the regional bank case study in the paper, and it is the perfect illustration of that gym membership analogy just playing out in real time.

Speaker B: Oh yeah, that's a great example.

Speaker A: This bank initially rolled out generic AI access to 5,000 employees. They just handed out logins and encouraged everyone to, you know, explore in the market.

Speaker B: Response was a total shrug. In fact, their relative valuation multiples actually compressed.

Speaker A: Right, let's translate that for a second. When valuation multiples compress, it basically means investors became completely unwilling to pay a premium for the company's future earnings. The market looked at their broad AI rollout, saw A bunch of experimental waste and essentially downgraded their growth potential.

Speaker B: Exactly. And the leadership at the bank noticed this multiple compression and executed a hard pivot.

Speaker A: Thank goodness they did.

Speaker B: Yeah, they revoked the broad generic access and built a highly specialized 150 person AI ops team.

Speaker A: Out of 5,000 employees, right?

Speaker B: A tiny fraction. They gave this specific group access to the most powerful frontier models, dedicated budgets and high stakes production targets. We're talking complex fraud detection and credit risk analysis.

Speaker A: Yeah, the Olympic training facility.

Speaker B: Exactly. And within a few quarters, 75% of the entire bank's AI token volume was being generated by just 3% of its employees.

Speaker A: The 100 Olympians. Yeah. And when they made that shift from broad experimentation to serious production grade deployment by those power users, the bank's valuation multiple expanded again. And they outperformed their peers.

Speaker B: And a core reason for that outperformance is a fundamental shift in how those Olympians actually use the technology. The paper documents the rapid rise of agency agentic AI. Yeah, this is a critical evolution in the data. Agentic usage means the AI is no longer just a passive oracle answering questions, it is acting as an independent agent.

Speaker A: Oh wow.

Speaker B: It calls external tools, executes multi step workflows, and coordinates actions across various software platforms without any human intervention.

Speaker A: So it's not a user typing, hey AI, write an email to my boss. It's like an automated trigger where the AI looks at a live database, finds a statistical anomaly, generates a custom diagnostic report, opens an email client drafts the message and schedules it to send to the executive team at 8am tomorrow. All while the human operator is literally asleep.

Speaker B: Exactly. The researchers found that by the end of their sample period, this agentic usage grew to account for over 50% of all tokens processed.

Speaker A: That's huge.

Speaker B: The era the chatbot, you know, the simple conversational interface is ending. The era of the AI participant which actively executes business processes has arrived.

Speaker A: But I mean, if AI is now an active participant executing complex workflows, it is inevitably stepping on somebody's toes. So within these highly valued companies, we have to ask, whose jobs are most impacted by this high level agentic deployment?

Speaker B: And this brings us to the market implied occupation exposure map, which is where they map specific job skills directly to market valuation.

Speaker A: Here's where it gets really interesting, because when you look at the occupations the market is penalizing. It contradicts decades of career advice.

Speaker B: It really does.

Speaker A: The market is actively penalizing analytical, scientific and operations control roles. I mean, I was told my whole life to learn math, learn to code, become a data Analyst get into stem.

Speaker B: We all were right.

Speaker A: So are we really seeing data showing that a, uh, highly trained scientist analyzing lab data is facing a bigger headwind in the stock market than like a salesperson pitching a client or a tech doing system installation? That feels completely backward.

Speaker B: It feels backward, but the data proves exactly that. To understand the mechanics of why, we have to examine how the researchers broke down the O NET database.

Speaker A: Okay.

Speaker B: One ad categorizes the granular day to day skills required for thousands of different jobs. When researchers cross referenced these skills with the market data, they found the stock market heavily rewards what they term interactive cognitive complexity.

Speaker A: Meaning human to human skills.

Speaker B: Yes. Social skills load positively on the market exposure map, showing about 0.16 standard deviations of positive correlation. And tasks categorized under interaction and communication load the highest at a positive 0.36 standard deviations. Meanwhile, on the other end of the spectrum, tasks focus purely on information use and deductive reasoning load negatively. They show negative 0.29 and negative 0.27 standard deviations.

Speaker A: So the market is signaling that artificial intelligence is rapidly commoditizing pure analytical processing and information synthesis completely. If your primary value proposition at work is, you know, taking a raw data set, crunching it, finding a pattern, summarizing it in a document, you are performing a function that a frontier model can do infinitely faster and cheaper.

Speaker B: Yeah, the human premium, the, the scarcity that commands high wages has completely shifted. AI is driving the cost of cognitive processing and deduction towards zero. Therefore, the new premium lies in the friction of the physical world and the messy reality of human consensus. AI handles the analytics, humans handle the persuasion, the instruction, and the physical installation of systems in the real world.

Speaker A: But I mean, if individual analytical workers are being commoditized, that dynamic has to extend to entire business sectors built around information processing.

Speaker B: Oh, it does.

Speaker A: I'm thinking specifically about the software industry. What happens to SaaS companies?

Speaker B: Well, the transition risk we discussed earlier is starkly visible in the software sector. The paper noted that while retail and consumer durable companies are seeing positive stock exposure to AI, several software application firms are experiencing severe negative exposure.

Speaker A: Yeah, I want to dig into the real world example of the mid size enterprise SaaS firm mentioned in the paper. Because when generative AI first hit the mainstream, the SaaS firm did what everyone else did. They quickly bolted a conversational chatbot onto

Speaker B: their existing dashboard and the market gave them a brief reward. Their stock popped initially, but by late 2025, their market exposure completely flipped to negative. The frontier models had advanced so rapidly that they could natively replicate the core analytical functionality of the SaaS firm's ENT entire software suite.

Speaker A: Let's visualize how that actually happens to a software company. Imagine you are an enterprise paying $100,000 a year for a specialized software dashboard. That dashboard pulls your customer data, analyzes churn risk, and generates a visual report for your sales team.

Speaker B: Right.

Speaker A: Two years ago you absolutely needed a SaaS company to build and maintain that user interface. Today, an agentic AI workflow can just ping your raw database directly, write its own python script to analyze the churn risk and email the sales team a summary.

Speaker B: Exactly.

Speaker A: It bypasses the specialized software dashboard entirely. You don't need to pay for a user interface. The AI can just talk directly to the database.

Speaker B: The SaaS firm's core product was turning into a commodity overnight. But they survives because they recognized this multiple compression early and executed a brutal pivot.

Speaker A: What did they do?

Speaker B: They actively communicated to investors that they were moving away from selling AI as a feature within their application. Instead, they pivoted their entire engineering team toward workflow orchestration and backend infrastructure.

Speaker A: Uh ah, so they realized AI was going to replace the car, so they decided to own the toll road.

Speaker B: Exactly that. They positioned themselves as the underlying infrastructure that Agendic AI uses to connect different enterprise databases together, rather than being the standalone application that the AI easily replicates.

Speaker A: And by communicating this transparently, they survived the market correction that decimated their competitors.

Speaker B: Yeah.

Speaker A: So what does this all mean? I mean, moving from high level macroeconomic transition risk and software industry warnings down to ground level tactics, what should you listening actually do about this on Monday morning?

Speaker B: Well, the paper concludes with a very clear playbook for how organizations should structure their AI deployment. The primary recommendation is the barbell strategy.

Speaker A: Like in weightlifting. Heavy investments on the ends, nothing in the middle.

Speaker B: Exactly. On one end of the barbell you have your mission critical production workloads. These are the agentic workflows that directly touch the customer or create core enterprise

Speaker A: value, high stakes stuff.

Speaker B: Right. For these, you pay the premium. You buy the absolute best, most expensive closed source frontier models. You give your Olympians the best gear available.

Speaker A: Okay. And on the other end of the

Speaker B: barbell, you have your experimental internal low stakes learning applications. For these, the paper recommends using open weight models like Llama or Mistral because they're cheaper, cheaper and they allow you to download the AI and run it locally on your own company servers. This provides total data privacy. Your engineers can tinker with the model, fine tune it on internal company jargon and build prototypes without paying Massive per token API costs to a major tech company.

Speaker A: It's secure, cost effective experimentation.

Speaker B: Exactly. What the barbell strategy avoids is the muddy middle. Spending premium API dollars on casual employee experimentation or relying on cheap, underpowered models for critical production workflows.

Speaker A: So that covers the technical architecture, but the human element is equally vital. I mean, if you are a leader, you cannot hide these market signals from your workforce. If I'm a data analyst and I read the onet data in this paper, I am genuinely concerned for my livelihood.

Speaker B: And you should be. The paper explicitly stresses transparency and trust. You cannot manage a workforce transition of this magnitude in secret. The text highlights a major healthcare system that dealt with this exact vulnerability head on.

Speaker A: Right. This health system audited their own operations and realized that a massive chunk of their workforce, specifically those handling clinical documentation, medical coding and diagnostic imaging workflows, had strongly negative market exposure.

Speaker B: Yeah, AI is simply getting too good at analyzing X rays and writing medical charts.

Speaker A: It's wild. And leadership could have easily hidden this data, utilized the AI to automate the workflows, and quietly initiated rolling layoffs to cut costs.

Speaker B: Which is what a lot of companies do.

Speaker A: Yeah, but instead they opted for radical transparency. They went to the staff and explained that the market was signaling the impending automation of their core analytical tasks.

Speaker B: But they didn't fire them.

Speaker A: No. They funded a comprehensive three year retraining program. They explicitly aimed to transition these highly skilled analytical workers into positively exposed roles.

Speaker B: Right.

Speaker A: They move them into patient interaction, complex care coordination and family counseling. The human to human empathy and persuasion tasks that the market currently values at a premium.

Speaker B: It's brilliant. They took the smartest, most analytical minds in the hospital and systematically retrained them to be the most empathetic, persuasive communicators in the hospital.

Speaker A: Which is so hard to do.

Speaker B: It is incredibly difficult. But their transparent approach, maintained institutional trust, prevented a collapse in morale, and retained decades of valuable medical knowledge within the organization.

Speaker A: That is a profound lesson in change management. And it leads to a direct challenge for you, the listener. You need to audit your own skills this week. Look at your calendar for the last month. Are you leaning entirely on deductive reasoning? Is your whole professional value proposition rooted in crunching numbers, finding patterns and writing summaries? Or are you actively cultivating your capacity for persuasion, teaching, and integrating disparate systems in the messy, physical real world?

Speaker B: Because the market has already made its choice on which of those profiles commands a premium.

Speaker A: Exactly. The stock market aggregates truth faster than any single individual can. Right now, its clearest signal is Demanding that we become more fundamentally human, which

Speaker B: is such a fascinating paradox.

Speaker A: It really is. So just to wrap this all up, the market isn't rewarding Shallow gym membership AI hype. The 64 basis point premium is reserved strictly for deep, complex agentic deployment by power users. Purely analytical software products and traditional STEM skills are facing a relentless headwind of commoditization. Meanwhile, interactive, persuasive and physical skills are commanding a massive premium. Ironically, the rise of supreme artificial intelligence is making uniquely human interactive skills more financially valuable than ever before.

Speaker B: And you know, if we connect this current reality to the broader horizon, it leaves us with one incredibly provocative question that the research doesn't even attempt to answer.

Speaker A: Oh, what's that?

Speaker B: Well, if the market is currently rewarding human persuasion and empathy simply because AI is busy mastering the analytics, what happens to our economy in three, five, or ten years if AI crosses that final threshold?

Speaker A: Wow.

Speaker B: What happens when an agentic model becomes demonstrably better at emotional persuasion, empathy and human interaction than we are? If the ultimate interactive skill is fully automated, where does the market look next

Speaker A: for human value when the unsentimental truth telling machine decides that even our humanity can be replicated at scale? Man, that is a fascinating thought to mull over.

Speaker B: It really is.

Speaker A: Well, thank you for joining us on this deep dive. Go audit your calendar, build those interactive skills, and we will catch you next time.