Insights from Morgan Stanley's Latest AI Report: The Company That Will Create the Greatest Value May Not Have Been Founded Yet
If you're still debating "which large language model is the most powerful," you might already be a step behind.
Over the past year, we have witnessed a seemingly contradictory phenomenon:
On one hand, OpenAI, Anthropic, Google, and DeepSeek have continuously pushed the boundaries of model capabilities, making AI increasingly intelligent; on the other hand, the cost of invoking large language models has been declining steadily, with inference costs dropping nearly 10-fold within a single year.
According to traditional business logic, plummeting prices would signal that an industry has entered a red ocean of cutthroat competition; yet the AI industry defies this trend entirely — the lower the prices fall, the larger capital investment becomes, and the faster applications proliferate.
Why is this the case?
Morgan Stanley's recently released report The Big Picture: Artificial Intelligence — Ten Investment Truths offers a thought-provoking answer: the real AI story was never about models, but about a global capital cycle that is currently taking shape.
In this report, Morgan Stanley does not predict which company will become the next NVIDIA, nor does it argue over whether GPT or DeepSeek is more capable. Instead, it attempts to answer a far larger question:
Over the next decade, how exactly will AI reshape the economy, capital markets, and industrial landscapes?
AI is not a mere technological upgrade, but a full-scale infrastructure revolution
Over the past two years, discussions about AI have often fixated on model parameters, leaderboards, and benchmark scores.
But Morgan Stanley argues this perspective is far too narrow.
What truly drives the AI revolution is not models alone, but a complete ecosystem made up of infrastructure, algorithms, energy, data, software, and the real economy.
Historically, nearly every major technological revolution has followed the same path: Infrastructure deployment → Cost reduction → Application explosion → New industry emergence.
This pattern held true for the 19th-century railways, the 20th-century power grids, and the internet era.
The report specifically notes that after the U.S. Telecommunications Act of 1996 was enacted, massive amounts of capital flooded into fiber-optic construction. At the time, many viewed this as severe overinvestment, since huge stretches of fiber optic cable sat idle for years.
But what ultimately changed the world was not the fiber optics themselves, but the companies that emerged long after the networks were completed: Google, Facebook, YouTube, Netflix, Uber, and Amazon.
Infrastructure leads the way, and applications follow to capture the value.
Today's AI is repeating this exact historical pattern.
Global tech giants are pouring hundreds of billions of dollars into building GPU clusters, data centers, energy systems, and cloud computing platforms — an effort that may appear to be reckless cash burning.
Yet Morgan Stanley believes these investments are most likely not being made to serve today's ChatGPT, but to support new applications that have not even been conceived in the coming years.
This is the most critical sentence in the entire report:
The AI infrastructure being built today will most likely power applications that do not yet exist to consume its capacity.
The plummeting price of Tokens is not bad news — it means the AI flywheel has begun accelerating
Many observers see the continuous price drops of AI models and worry that industry competition has become excessively fierce.
But Morgan Stanley argues that falling prices are themselves a sign that AI has entered its explosive growth phase.
Citing data from OpenAI CEO Sam Altman, the report points out that AI Token costs fell roughly 10-fold in 2025 alone.
Most people's first reaction is: With prices collapsing, how can companies still turn a profit?
History tells us this is not a contradiction at all.
In the telegraph era, steadily falling communication costs gave rise to a global commercial network; in the internet era, increasingly affordable bandwidth spawned search engines, social media, e-commerce, and streaming services. Today, the rapid decline in AI inference costs similarly means that more businesses can integrate AI directly into their core operational workflows.
Cost reduction does not eliminate demand. On the contrary, it unlocks entirely new categories of demand that were previously unimaginable.
This is precisely the operating logic of the AI flywheel: increased computing power investment → lower inference costs → reduced barriers to enterprise adoption → rapid growth in the number of applications → sustained rise in Token consumption → which in turn drives new rounds of infrastructure investment.
Unlike the internet era, this flywheel is spinning at an unprecedented speed.
The report notes that telegraph services took decades to achieve mass affordability, the fiber-optic internet took roughly 10 years, while AI costs have undergone this massive compression in just one year.
As a result, the inflection point for widespread AI applications may arrive not in 10 years, but in just two or three.
The biggest winners of the future may not even exist yet
This is the perspective that most resonated with AI business commentators across the entire report.
Today's capital markets have directed nearly all their attention to established giants like OpenAI, NVIDIA, Microsoft, and Google.
But Morgan Stanley states: The companies that will reap the largest rewards will not necessarily be the most prominent players today.
The reasoning is straightforward. In every technological revolution, infrastructure builders play an essential role, but the entities that ultimately transform the world are almost always the application-layer players that emerge later.
In the railway era, no one foresaw the rise of modern logistics; when fiber-optic cables were being laid for the internet, no one anticipated short-video platforms, livestream e-commerce, or ride-sharing services. Today's large language models may well be nothing more than the "fiber optics" of this new era.
The real explosion will come from entirely new industries: Agents, robotics, autonomous driving, digital workforces, intelligent manufacturing, AI-powered healthcare, and AI-driven scientific research.
In other words: The large language models we discuss today will likely function as nothing more than the operating systems of the future AI economy.
The application layer that will generate massive, transformative value has not fully emerged yet.
Therefore, AI investment should not focus solely on models, but on the entire industrial chain:
- Chips
- Cloud computing
- Data centers
- Software platforms
- Agents
- Enterprise services
- Robotics
- Smart end-user devices
The report argues that AI in the future will not be a single isolated industry, but a long-term capital theme that spans multiple asset classes and industrial sectors.
Four key risks that demand vigilance
While maintaining an overall optimistic outlook, the report does not ignore risks, and identifies four issues that merit long-term attention.
First, the pace of technological progress could slow down.
If the current large language model architecture gradually approaches its scaling limits, and simply adding more computing power no longer delivers proportional capability improvements, today's massive capital investments could turn into stranded, underutilized assets.
Second, the speed of commercialization could lag behind the speed of capability improvement.
AI capabilities have advanced rapidly, but overall enterprise productivity has not yet seen a corresponding, noticeable improvement.
If AI remains stuck as a "cost center" for a prolonged period, rather than evolving into a genuine new source of revenue, capital markets may reevaluate the valuation of the entire sector.
Third, there are critical reliability risks associated with Agents.
A chatbot that gives an incorrect response only generates a single erroneous sentence. But future Agents will have the ability to directly manipulate enterprise workflows, databases, and business systems. A single error in this context could translate directly into a real-world operational disaster.
The report notes that massive gaps still exist today in areas of Agent liability attribution, legal frameworks, and regulatory oversight.
Fourth, systemic risks stem from the extreme interconnectedness of the industrial chain.
The AI industry appears to span distinct layers including chips, cloud computing, models, and applications, but in reality all segments are deeply interdependent.
If a geopolitical conflict, sharp decline in capital expenditure, or sudden regulatory shift occurs, the impact could rapidly propagate across the entire industrial ecosystem, rather than being contained to a single company.
Conclusion
What truly defines the future is not models, but the capital cycle
If I had to summarize Morgan Stanley's report in one sentence, it would be this:
The real value of AI does not lie in any individual model, but in the global capital cycle that is currently being forged.
Over the past several years, we have grown accustomed to framing AI as a software revolution.
But it is now increasingly clear that it is far more similar to the railways, power grids, and the internet — a new kind of infrastructure revolution.
Model capabilities will continue to improve, Token prices will keep falling, and new applications will continue to emerge in an endless stream.
What truly deserves our long-term attention is not how much any single company's stock price has risen today, but which players can successfully integrate infrastructure, model capabilities, software platforms, industry-specific applications, and the real economy into a self-sustaining, continuously evolving flywheel.
Morgan Stanley concludes the report by reminding readers that no one has prior experience deploying AI on this scale, so all projections must be approached with humility. Smart AI investing is not about betting on a single predetermined outcome, but about building an analytical framework that can adapt continuously as new evidence emerges.
Looking back from where we stand today, we may very well be in the "fiber-optic laying" phase of the AI era.
Over the next decade, the products that will truly change the world may not even have been invented yet; the companies that will create the greatest value may not have been founded.
This is precisely what makes the AI era so full of promise — the greatest opportunities almost always arise at the exact moment when foundational infrastructure is fully deployed, and application innovation is just beginning to take off.
This article originates from the WeChat public account "AI Business Review", authored by AI Business Review, and published with authorization by 36Kr.