Since ChatGPT brought generative AI into the public spotlight in late 2022, changes in investment have accelerated even faster: corporate spending on AI hardware and data centers is approaching levels seen during some of the largest investment waves in U.S. history. The market has responded with impressive revenue projections, but a critical question arises—how likely are these forecasts to materialize, and are they worth the capital and time required?

According to information from Zhui Feng Trading Desk, Michael J. Mauboussin of Morgan Stanley Investment Management's Counterpoint Global outlined a clear methodology in a report dated the 10th: evaluating such forward-looking judgments should begin with an initial belief and update that belief as new results emerge, a concept known as the Bayesian formula: "new conclusion = initial judgment (prior probability) × adjustment factor from new evidence (likelihood ratio)."

Using this framework, the report placed two highly watched projections into historical context: OpenAI's revenue forecast rising from $3.7 billion in 2024 to $145 billion in 2029 (representing a 108% compound annual growth rate over five years), and Oracle's cloud business revenue growing from $10 billion in fiscal 2025 to $166 billion in fiscal 2030 (a 75% compound annual growth rate over five years). The conclusion was stark: among U.S. publicly listed companies from 1950 to 2024, no company of similar initial scale has achieved such growth rates.

Compounding the challenge, AI infrastructure is not as simple as "buying a few more servers." Data center construction is fundamentally a large-scale engineering project, and such projects come with their own baseline failure rates: budget overruns, delays, and underperformance relative to expectations are common. The report also interpreted the recent wave of high-volume transactions and expansion announcements through a competitive strategy lens: they may not only aim to meet demand but also serve as signals to competitors and deterrents to potential market entrants—though such preemptive bets carry inherent high risks.

Applying OpenAI's forecast to historical data reveals a 108% compound growth rate that falls into a "blank" area within the sample. The report used a specific reference group: U.S. publicly listed companies from 1950–2024 with initial revenues between $2 billion and $5 billion (adjusted to 2024 dollars), comprising nearly 18,900 company-period observations. The average five-year compound revenue growth rate for this group was only 7.0%, with a standard deviation of 10.6%.

OpenAI's projection implies a jump from $3.7 billion in 2024 to $145 billion in 2029—a 108% compound annual growth rate over five years. The report noted that no publicly listed company has achieved this speed in the past three-quarters of a century. Even using a normal approximation, this result lies nearly 9.5 standard deviations from the mean, making it extremely improbable. Moreover, historical growth rate distributions are not normal and have fatter tails, yet the conclusion remains that such an outcome is "almost unseen."

An interesting detail: since the event has "never occurred" in the sample, the baseline probability becomes zero, making the Bayesian formula difficult to apply directly. The report adopted common heuristic adjustments (such as 3/N or Laplace smoothing), but the resulting initial probability remained below one-tenth of one percent.

While new evidence can increase the probability, the report did not lean toward optimism. It acknowledged that baseline probabilities are not immutable and identified two factors that could raise OpenAI's probability of success from "near zero":

- Diffusion speed: ChatGPT reached 100 million users in two months, compared to nine months for TikTok, 28 months for Instagram, 4.5 years for Facebook, seven years for the internet, 16 years for mobile phones, and 75 years for telephones. Even accounting for population changes, this speed is historically rare. However, the report cautioned that user growth does not equal revenue growth, as many users do not pay. - Short-term revenue growth: OpenAI expects 2025 revenue of approximately $13 billion, representing roughly 250% year-over-year growth—far above the average compound growth rate projected over five years.

However, the report immediately tempered optimism: as companies grow larger, growth rate volatility typically decreases, making it increasingly difficult to sustain high growth. Additionally, OpenAI's projection for 2030 revenue of $200 billion implies a 72.7% compound annual growth rate from 2025 to 2030.

Using a reference group of companies with initial revenues between $10 billion and $15 billion (approximately 3,700 observations), the report again concluded that no company has achieved such growth. Even when lowering the initial revenue threshold to at least $6.5 billion and expanding the sample to over 16,400 observations, the result remained the same.

Growth does not automatically create value. The report defined Total Addressable Market (TAM) with constraints—not simply "how much can be sold," but "how much revenue can be generated while creating shareholder value, assuming 100% market share." The key threshold is whether investment returns exceed the cost of capital.

In OpenAI's case, the report highlighted specific constraints:

- 2025 free cash flow is reportedly -$9 billion, with an expected -$17 billion in 2026. Sustaining rapid expansion and heavy investment under these conditions will almost certainly require continuous external financing. - A significant portion of employee compensation consists of stock-based compensation (SBC), estimated to exceed 45% of revenue in 2025. On a per-employee basis, this equates to approximately $1.5 million annually—seven times the SBC intensity seen in large tech companies before their IPOs.

These factors do not directly invalidate revenue projections, but they highlight an often-overlooked issue: even if revenue growth is achieved, capital structure, financing conditions, and dilution costs may determine what shareholders ultimately receive.

Oracle's $166 billion cloud revenue target is supported by signed contracts, but delivery and financing remain hard constraints. The company announced several multi-billion-dollar cloud infrastructure contracts in 2025, significantly boosting its Remaining Performance Obligations (RPO). Management projects cloud business revenue growing from $10 billion in fiscal 2025 to $166 billion in fiscal 2030—a 75% compound annual growth rate over five years. Oracle's cloud business accounted for about 17% of its total fiscal 2025 revenue of $57.4 billion.

The report again applied baseline probability: over the past 75 years, no company with initial revenue exceeding $10 billion has achieved such a five-year growth rate. Even when lowering the initial revenue threshold to above $5.6 billion, none have succeeded.

A more tailored reference group—companies with initial revenues between $8 billion and $12 billion (approximately 4,400 observations)—showed an average five-year compound growth rate of 5.7% with a standard deviation of 9.6%. The report noted that this compares a "business segment" to "entire companies," which is not a perfect match.

Oracle's differentiating factor is its RPO scale, which allows for probability adjustments. However, the report emphasized that adjustments must consider not only contracts but also accompanying financing needs, counterparty risks, and potential delays in infrastructure deployment.

AI data centers represent large-scale engineering projects, and such projects have low baseline success rates. The report cited Bent Flyvbjerg's database of 16,000 large projects, with sobering results:

- 47.9% of projects were completed within budget. - Only 8.5% were completed on time and within budget. - A mere 0.5% were completed on time, within budget, and achieved expected benefits.

The takeaway is clear: do not assume projects will proceed as planned. Key bottlenecks—such as power supply, specialized hardware, and equipment—must be closely monitored. Modular designs tend to have higher success rates, but in a fast-moving AI landscape where competitors are racing for leadership, "thinking slowly and acting quickly" is challenging to execute.

The surge in transactions and expansion announcements may represent a "preemptive deterrence" competitive experiment. The report noted that OpenAI announced approximately 15 infrastructure-related deals in 2025. At the same time, hyperscale cloud providers like Alphabet, Amazon, and Microsoft raised capital expenditure forecasts, while companies like Anthropic and CoreWeave also committed to large investments.

The report compared this trend to the telecom investment boom of the late 1990s and early 2000s, which eventually led to overcapacity and bankruptcies. While current AI demand may still be far from its ceiling—global GenAI adoption stood at only 16% in the second half of 2025—the report suggested that part of the motivation may be strategic signaling: using large-scale capacity commitments to secure market position and deter competitors and new entrants.

Citing Michael Porter's concept of "preemptive strategy," the report also highlighted the risks: committing substantial resources before market outcomes are clear may backfire if it fails to deter competitors, potentially triggering intense battles of attrition. A key differentiator is financing capacity: startup AI companies rely on continuous external funding, while giants like Amazon, Google, and Meta possess stronger cash flows and greater endurance. As of 2025, capital remains available, but the report cautioned that this may change.

The core message of the report is not to be bearish on AI, but to reframe judgment as an updatable probability exercise: start with baseline probabilities to temper enthusiasm, then adjust based on diffusion speed, actual revenue, project progress, and financing conditions. The report explicitly avoids offering investment advice—but it provides a starting point that is harder to ignore: when forecasts fall into historically unobserved territory, optimism requires evidence, and ongoing evidence at that.