Inside every enterprise right now, there is the conversation that is happening, and the one that isn’t.

The conversation that is happening sounds like this: We need to show visible progress on AI. Our competitors are moving. The vendors are embedding AI into every renewal conversation. What is our roadmap? The pressure is real, the timeline is compressed, and the expectation, even when unstated, is action over analysis.

The conversation that isn’t happening is about what an enterprise is actually committing to when it says yes to AI transformation. It involves assumptions about technology that has not yet stabilized and pricing that is structurally unknowable. It also includes regulatory frameworks that are still taking shape even as deployment is underway, as well as vendors who may not survive the rationalization that is coming. At the same time, companies are relying on playbooks that do not yet exist, and they are facing workforce implications that are already appearing in collective bargaining agreements before they have been formally addressed in program plans.

The organizations that succeed in the AI era aren’t the ones who resolve the first conversation fastest. They’re the ones who are having both.

We’ve spent considerable time mapping this terrain, the pressures, the structural uncertainties beneath them, and what history tells us about how organizations have navigated periods of this kind of uncertainty before, and how those that failed to navigate it fared. There are resources at the end of this post for leaders who want to go deeper.

Two Kinds of Pressure

Before mapping the uncertainties, it’s worth naming the pressures because they’re the mechanism by which dangerous assumptions get suppressed.

Business and competitive pressure is real. OpenAI has 900 million weekly active users. McKinsey estimates AI could add $4.4 trillion annually to the global economy. The board reading these numbers doesn’t see a reason to slow down. The board sees a reason to demand a roadmap by next quarter.

Security and sustainability pressure is real and different in character. The argument isn’t about competitive opportunity; it’s about existential risk. Anthropic withheld its most capable model, Mythos, specifically because of cybersecurity concerns. If the most safety-focused AI lab in the world is containing its most capable system due to security risk, the adversarial version of that capability isn’t hypothetical. It’s a matter of timing. The enterprise that frames inaction as prudence may be underestimating what it’s actually choosing.

Both pressures are legitimate. But they share a specific effect: they narrow the space in which assumptions are examined. Urgency and fear are among the most effective assumption-suppression mechanisms in enterprise decision-making. They don’t eliminate the assumptions. They make them harder to surface before they’re embedded.

Six Structural Uncertainties and More Than Twenty Points of Exposure Beneath Them

Beneath the pressure, six structural uncertainty dimensions define the terrain every CIO is navigating. They are not independent; they compound. Across those six dimensions we have catalogued more than twenty discrete points of exposure where assumptions are routinely made without being examined. What follows is one striking point from each. The full inventory is at the end.

Technology. The binding constraint in AI infrastructure is not silicon; it is power. The International Energy Agency projects AI data centers will consume 945 TWh of electricity by 2026, more than Japan’s total annual consumption. The vendor landscape hasn’t rationalized, and the capability you’re paying a premium for today is likely open-source table stakes within 18 to 24 months. Your multi-year business case is built on a foundation with multiple simultaneous sources of instability — technology, vendors, and model pricing are each moving in ways that compound rather than offset each other. This point and two others are mapped in detail in the full inventory.

Environmental. Most enterprise AI programs have modeled regulatory risk. Fewer have modeled what is arriving through a different channel entirely: organized labor. The AFL-CIO has moved AI governance from philosophical concern to active collective bargaining. AI deployment constraints and worker notification requirements are appearing in contracts across manufacturing, healthcare, and logistics right now. An AI program that is technically approved and commercially funded may encounter workforce constraints that are legally binding before it reaches go-live.

Pricing. As AI becomes more efficient and cheaper per unit, total consumption accelerates and total cost rises. Enterprise AI spend tripled over the same two years that per-token costs fell 280 times. The mechanism is agentic systems, RAG pipelines, and always-on agents, each multiplying the unit count while the unit cost falls.

If your business case is built on declining per-unit AI costs, you may be solving the wrong equation. The pricing section of the full inventory surfaces four distinct points of exposure, including the energy cost variable that almost no multi-year AI financial model has accounted for.

Implementation. The support cost assumption in an AI deployment has a problem that doesn’t exist in traditional enterprise software: there is no empirical baseline. With an ERP, decades of industry data exist on support cost ratios and escalation rates. With AI, particularly agentic systems with probabilistic outputs, the number in your plan is invented, not derived.

It enters the business case as a commitment nobody can actually defend, and it draws on program contingency long before the genuine risks materialize. BCG research finds 74 percent of organizations fail to scale AI value beyond the pilot stage. Support cost reality is one of the primary reasons why.

Business Model. Three fundamentally different AI investment strategies are available: efficiency, elevation, and transformation. Each requires different investments, different governance, and different definitions of success. Most organizations pursue elements of all three in the same business case, without ever establishing which one the program is actually being held accountable to deliver. That works until the program hits a decision point where those strategies point in opposite directions, which they will. Without a prioritized objective on record, there is no basis for making that call.

Talent and Governance. It is easy to assume that a leader who delivered a successful ERP or process reengineering program is the right person to lead an AI transformation, and in most governance structures, that assumption would never be formally challenged. Traditional transformation leadership was built for deterministic systems. AI transformation leadership requires a fundamentally different posture: comfort with probabilistic outputs, the ability to detect silent model drift in production, and genuine independence from the commercial relationships that most consistently suppress the hardest questions.

What The Best-Prepared Leaders Do Before They Commit to AI

None of this is an argument for inaction. The pressures driving AI investment are real, and waiting for certainty is not a viable strategy in an environment where the technology is moving faster than the conventional enterprise decision cycle.

The argument is for something more precise: committing with a clear inventory of what you’re assuming, who owns each assumption, what would have to be true for each one to hold, and what your program does when it doesn’t. Those assumptions will be tested. The question is whether your program surfaces them before commitment or discovers them after.

The organizations that navigate this era well are not the fastest movers. They are the ones that move with clarity at the decision point and with the capacity to understand what is being assumed in each commitment before options collapse and the only choices left are expensive ones. That capacity has to be built deliberately. The question is whether your program has a mechanism for it.

Two Resources for Enterprise Leaders Who Want to Go Deeper

The AI Era Uncertainty Map is the diagnostic companion to this post with a full inventory of more than twenty uncertainty points across all six dimensions and key diagnostic questions for each. It’s designed to be used as a working reference before a board conversation, a program gate, or a major vendor commitment. If you want to move through this environment with a structured map in hand, this is the starting point.

Download the AI Era Uncertainty Map →

The Crucial Path is UpperEdge’s research into what it means to sustain optionality across an enterprise AI commitment, drawing on what history tells us about how organizations have navigated high-stakes technology uncertainty before, and what separated those that preserved their ability to course-correct from those that didn’t. If the Uncertainty Map tells you what terrain you’re in, this research tells you how to move through it without closing off your options prematurely.

Access The Crucial Path research →

UpperEdge is an independent advisory firm for enterprise leaders navigating AI transformation. We don’t sell software, services, or implementation work, which means the only thing we’re optimizing for is your outcome.