Artificial Intelligence:
Bubble or Bedrock?
A Rigorous Assessment of AI's Maturity, Hype Cycle and Long-Term Value

Every major technology wave in history has been accompanied by some version of the same question: Is this real or is it a bubble? The railway mania of the 1840s, the dot-com boom of the late 1990s, the crypto cycle of 2020–2022 — each generated enormous capital flows, euphoric predictions, and ultimately, a reckoning.

Artificial intelligence is now generating capital flows that dwarf all of these precedents. Nvidia's market capitalisation exceeded $3 trillion in 2024. Global AI investment reached $200 billion annually. Enterprise boards that had never discussed AI in 2022 now have AI strategy as a standing agenda item. The question of whether this represents genuine, enduring value creation or speculative excess is not abstract — it is directly relevant to every enterprise investment decision being made right now.

After 18+ years of experience in technology delivery, AI governance, and digital transformation — watching multiple technology cycles from the inside — I've developed a framework for evaluating this question that goes beyond both the uncritical enthusiasm of AI boosters and the reflexive skepticism of bubble-callers. This article is that assessment, applied rigorously to the current AI landscape.

---

The Question That Every Boardroom Is Asking

The bubble question is being asked in multiple forms in boardrooms and investment committees globally:

• "Are we investing in AI because it creates genuine business value, or because our competitors are and we fear being left behind?"
• "Is the capital flowing into AI companies justified by the economic value AI is actually generating, or is this a valuation bubble driven by narrative?"
• "Which AI capabilities are actually mature enough to deliver enterprise value now, versus which are marketing claims built on demonstrations that don't scale to production?"
• "If there is a correction in AI valuations, what happens to the enterprise AI programs we have already committed to?"

These are the right questions. Answering them requires distinguishing between the financial/investment dimension of the bubble question (are AI company valuations justified?), the technology maturity dimension (how capable and reliable are current AI systems?), and the economic value dimension (is AI actually improving productivity and creating measurable value?). The honest answer is different for each dimension.

---

What Makes a Technology Bubble — A Framework

Before evaluating whether AI constitutes a bubble, it is necessary to be precise about what a bubble actually is. The term is used loosely to mean anything from "this is overvalued" to "this is fraudulent" to "this will collapse." These are different claims with different implications.

This definition is important because it is entirely possible for a technology to be both genuinely transformative AND the subject of a speculative bubble. The internet was both. The railway was both. The question is not whether the technology works — it is whether current valuations and investment levels are ahead of demonstrable commercial value generation.

The Five Bubble Indicators

Classic financial bubble analysis identifies five structural indicators. Each is assessable for the current AI environment:

1. Valuation multiples disconnected from earnings: P/E ratios, price-to-sales, and enterprise value multiples far exceeding historical norms for the sector
2. Capital deployment outpacing commercial returns: Investment in infrastructure, development, and capabilities significantly exceeding demonstrated revenue and profit generation
3. Narrative substituting for financial analysis: Investment decisions driven by "AI will transform everything" stories rather than specific, measurable business case analysis
4. Broad participation with limited discernment: Capital flowing to a wide range of AI companies without discrimination between genuinely differentiated and me-too competitors
5. Infrastructure overbuild: Physical or technical infrastructure (data centres, compute, models) being built at a rate that assumes near-term demand far exceeding realistic near-term adoption

---

Historical Parallels — Four Bubbles and What Survived Them

The most useful analytical tool for evaluating the current AI moment is historical precedent — not to predict when or how a correction might occur, but to understand the pattern of what survives technology bubbles and what doesn't.

The historical pattern is consistent: technology bubbles destroy financial capital while often building physical or infrastructure capital that goes on to generate genuine long-run value. The question for AI is which side of this pattern it falls on — and which parts of the AI ecosystem most closely resemble the railways versus the pets.com moment.

---

The Bubble Signals in AI Today

Applying the five bubble indicators to the current AI environment produces a mixed but concerning picture in the financial layer.

Valuation Multiples

Nvidia, the primary infrastructure play for AI compute, traded at price-to-earnings ratios above 60x through much of 2024 — compared to a S&P 500 average of ~20x. AI-focused startups routinely commanded valuations at 50–100x ARR (Annual Recurring Revenue), compared to a normal SaaS multiple of 5–10x ARR in a stable market. OpenAI's reported $80–100 billion valuation as of 2024 represents a multiple of its revenue that could only be justified by assumptions about future dominance that are by no means certain. These valuations are not obviously wrong — but they embed assumptions about AI revenue growth, margin expansion, and competitive moat that are far from guaranteed.

Capital Deployment Outpacing Returns

In 2024, hyperscalers (Microsoft, Google, Amazon, Meta) collectively committed over $200 billion in AI infrastructure capital expenditure. The challenge is that the revenue currently generated by AI services is orders of magnitude below this investment level. Microsoft's Copilot, Google Gemini, and Amazon Bedrock are generating meaningful but not yet transformative revenue. The assumption embedded in this level of investment is that AI monetisation will scale dramatically within 3–5 years — an assumption that may prove correct, but that represents a significant leap of faith relative to current demonstrated returns.

Narrative Over Analysis

Enterprise AI investment decisions in 2023–2024 were frequently driven by competitive anxiety rather than business case rigour. CIOs and CEOs who could not articulate a specific ROI case for their AI investments nonetheless committed to them because "we cannot afford to be left behind." The quality of business case analysis in AI investment decisions compares unfavourably to the analysis applied to previous major enterprise technology investments of comparable scale.

Indiscriminate Capital

The AI venture capital market of 2023–2024 showed classic bubble characteristics: capital flowing to nearly any company with "AI" in its pitch deck, regardless of differentiation. Applications of a thin wrapper around an API call — adding minimal proprietary value but capturing "AI startup" valuation multiples — represent a category of enterprise that will not survive a return to rigorous venture investing.

---

The Bedrock Signals — Why This Is Different

Despite the bubble indicators in the financial layer, there are fundamental differences between AI and the clearest bubble cases — differences that suggest the underlying technology is bedrock, even if current valuations are bubble-elevated.

The most important bedrock signal is the quality of the use cases already in production. AI is not a promise backed by a PowerPoint — it is code running in production environments, generating measurable business outcomes, across a wide range of industries. Unlike the internet at a comparable phase of development (1996–1997), current AI already has widespread, functional, revenue-generating enterprise deployments. That is fundamentally different from the dot-com era's "build it and hope the business model emerges" dynamic.

---

Where AI Sits on the Gartner Hype Cycle

Gartner's Hype Cycle is a useful framework for contextualising where specific AI technologies sit relative to their maturity and commercial value realisation — though it must be applied at the level of specific AI applications rather than "AI" as a monolithic category.

The critical insight from hype cycle analysis is that different AI technologies are at different points on the curve simultaneously. The hype cycle does not describe "AI" — it describes specific AI applications and subcategories, each of which has its own maturity trajectory.

Peak of Inflated Expectations

Generative AI and large language models are at or near the peak — characterised by maximum media coverage, maximum investment, maximum hype, and a growing gap between promise and demonstrated enterprise production performance. This does not mean LLMs are not valuable — it means the current expectations embedded in valuations and enterprise planning assumptions are likely higher than will prove realistic in the 2–3 year horizon.

Trough of Disillusionment (Entering)

AI agents — autonomous AI systems that take multi-step actions in the world without human intervention — are entering the trough. Early deployments have produced high-profile failures alongside impressive demonstrations. Enterprise adoption has been slower than projected, and the gap between agentic AI demonstrations and production-reliable agentic AI systems is becoming more apparent.

Slope of Enlightenment

Traditional ML, predictive analytics, and narrow AI for specific business processes (fraud detection, demand forecasting, quality control) are firmly on the slope of enlightenment — past the hype peak, past the disillusionment, delivering consistent, measurable business value in production environments across thousands of enterprises.

Plateau of Productivity

Computer vision for specific industrial applications (quality inspection, inventory management, security), NLP for customer service routing, and recommendation systems are at or near the productivity plateau — mature, reliable, and generating consistent ROI in production environments across industries.

---

AI Maturity by Domain — A Practitioner's Map

Based on 18+ years of working with enterprise technology and AI governance programs — and drawing on published maturity assessments from Gartner, IDC, and peer-reviewed research — the following maturity assessment reflects where specific AI domains actually are, rather than where marketing positions them.

Technology Readiness Level (TRL) is used as the assessment metric — TRL 9 indicates full production maturity; TRL 1–3 indicates research stage; TRL 4–6 indicates validation and early deployment; TRL 7–8 indicates near-production and early production.

---

The Productivity Paradox — Where Is the Economic Value?

The most intellectually interesting tension in the AI bubble debate is the productivity paradox: AI is demonstrably capable and demonstrably improving productivity in controlled settings — so why is the macroeconomic productivity data not showing the transformative uplift that AI advocates predict?

The Diffusion Gap

Technology productivity gains typically appear in macroeconomic data with a 10–15 year lag from widespread adoption. The internet's productivity impact on GDP was not clearly measurable until the early 2000s, despite broad consumer adoption beginning in the mid-1990s. Economists Paul David and Robert Gordon's work on the "productivity paradox" of electrification showed that the economic productivity gains from electricity did not appear in GDP data until 40 years after its widespread industrial adoption — because organisational and business model changes needed to realise productivity gains from new technologies take much longer than the technology adoption itself.

If the same pattern applies to AI, the absence of clear macroeconomic productivity signal as of 2025 is not evidence that AI is not generating value — it is evidence that we are in the diffusion and organisational adaptation phase that precedes the productivity realisation phase.

Where the Value Is Being Created Now

While macroeconomic data is ambiguous, microeconomic evidence of AI value creation is concrete and growing:

• Software development productivity: GitHub reports AI Copilot users complete tasks 55% faster than non-users, with adoption across millions of developers globally. Google reports 25% of new code at Google is written by AI, reviewed and accepted by human engineers.
• Customer service: Klarna reported its AI assistant handles the equivalent of 700 human agents, resolving 2.3 million conversations in the first month with customer satisfaction scores matching human agents
• Drug discovery: Google DeepMind's AlphaFold predicted the structures of 200 million proteins — a problem that would have taken thousands of years of laboratory work using conventional methods. Applications in drug discovery are accelerating materially
• Healthcare diagnostics: AI diagnostic tools for radiology, pathology, and ophthalmology are in clinical production, demonstrating detection accuracy at or above specialist consultant level for specific conditions
• Legal and professional services: AI document review, contract analysis, and legal research are demonstrably reducing the time for specific professional service tasks by 30–70%

---

The Investment Landscape — Follow the Capital

Capital flows are among the most informative signals available for technology maturity assessment — not because markets are always right, but because the scale and composition of investment reveals something about the collective judgment of sophisticated market participants.

The Numbers That Define the Moment

The scale of AI capital deployment in 2024–2025 is genuinely unprecedented:

• Hyperscaler capex: Microsoft, Google, Amazon, and Meta collectively committed over $250 billion in AI-related capital expenditure in 2024/2025 — more than the entire global IT industry invested in cloud infrastructure during the first decade of cloud computing
• Venture capital: AI companies raised $67 billion in venture funding in 2023 and over $100 billion in 2024 — more than all other technology sectors combined
• Nvidia revenue: Nvidia's data centre revenue reached $47 billion in fiscal year 2024, up from $15 billion in FY2023 — a growth rate that makes it one of the fastest-scaling companies in corporate history
• Foundation model funding: OpenAI raised $6.6 billion at a $157 billion valuation in 2024. Anthropic raised $7.3 billion. xAI raised $12 billion. These rounds are unprecedented for pre-revenue or early-revenue AI companies.

What This Capital Is Buying

Unlike dot-com era capital, which was frequently invested in marketing, "eyeballs," and speculative growth, AI capital is being deployed on physical infrastructure with genuine asset value:

• GPU clusters and data centres with real tangible asset value regardless of AI revenue trajectory
• Research talent building capabilities that will have durable competitive value
• Dataset and model development representing genuine intellectual property
• Cloud computing infrastructure that serves AI workloads but also general cloud computing demand

This is structurally more similar to the railway infrastructure investment of the 1840s — physical assets that retain value even if the specific companies that built them fail — than to the dot-com era's investment in perishable marketing and customer acquisition costs.

---

Sector-by-Sector Maturity Assessment

AI maturity is not uniform across sectors. The following assessment reflects genuine production deployment evidence rather than pilot or proof-of-concept activity.

---

What Will Burst — and What Will Survive

Drawing on the historical precedents and current maturity evidence, it is possible to make a calibrated assessment of which elements of the current AI ecosystem are most at risk in a correction and which are most likely to survive and create durable long-term value.

What Is Most Likely to Burst

• AI application startups with no proprietary data or model moat: Companies whose entire value proposition is a thin interface layer on top of a foundation model API. When foundation models become cheaper, better, and more accessible, the "AI chatbot for X vertical" companies with no proprietary data advantage will find their differentiation evaporates
• Current foundation model valuations: The specific financial valuations of frontier AI labs embed assumptions about future revenue, competitive moat, and regulatory environment that are far from certain. Not that the companies will fail — but that the valuations will compress
• GPU/infrastructure valuations at current levels: Nvidia's valuation embeds assumptions about perpetual AI compute demand growth that may not persist as model efficiency improves (smaller, more efficient models requiring less compute) and as competition in AI chip design intensifies
• Enterprise AI projects without clear ROI: The many enterprise AI initiatives launched in 2023–2024 without a clear, measurable business case will face budget scrutiny in 2025–2027 as CFOs demand evidence of return on AI investment

What Will Survive and Create Durable Value

• Foundation model infrastructure: The models themselves — their capabilities, training, and ongoing improvement — represent genuine intellectual property and platform infrastructure that will continue to generate value regardless of which specific company owns them
• AI applications with proprietary data advantage: Companies and enterprises that have used AI to create or leverage proprietary data assets that provide durable competitive advantage — personalised medical diagnostics, proprietary financial models, enterprise knowledge graphs
• Mature AI applications in production: Fraud detection, recommendation systems, demand forecasting, quality inspection — these are past the hype cycle and delivering consistent ROI. They will survive any correction because they are demonstrating value today
• The AI governance and compliance layer: As AI regulation matures globally, the professional services and software that helps organisations comply with AI governance requirements will generate durable value regardless of which specific AI capabilities survive the hype cycle correction
• Physical AI infrastructure: Data centres, energy infrastructure, and chip design capabilities represent physical assets that will retain value across AI capability cycles

---

Enterprise Strategy — Investing Wisely Across the Hype Cycle

For enterprise technology leaders navigating the AI investment landscape, the bubble/bedrock analysis produces a clear strategic framework: invest in the bedrock, approach the bubble with discipline, and position for the post-correction landscape rather than the peak.

---

Key Takeaways