The AI Bubble: Why it should not change your strategy
- MULTIPLAI
- 4 days ago
- 9 min read
Updated: 2 days ago
Investment cycles come and go. Enterprise value does not.
There is a growing debate about whether AI is heading for a correction similar to the dotcom burst of the early 2000s. The signs are familiar: record capital inflows, valuations disconnected from near-term revenues, infrastructure buildout running ahead of proven demand, and a rising chorus of scepticism from analysts and investors. The debate is legitimate. But for most enterprises, it is also largely beside the point.
The bubble question, to the extent one exists, is concentrated at a specific layer of the AI value chain. Understanding where the pressure sits, and where it does not, is what separates organisations that are building durable AI advantage from those that are either over-investing in hype or, just as damagingly, standing still because the noise has made them cautious.
Where the bubble risk actually sits
The dotcom parallel and where it breaks down
The dotcom boom produced genuine, lasting value: it gave us Amazon, Google, and the modern internet. But it also destroyed enormous amounts of capital in companies that built infrastructure and services for a market that did not yet exist at the assumed scale. The correction was severe, but the underlying technology continued to reshape the world. The organisations that survived and adapted captured more value in the decade after the crash than many had imagined possible during the peak.
The AI dynamic today bears a structural resemblance. Capital is concentrating at the infrastructure and model layer in ways that may not be sustainable at current levels. Global AI spend is projected to rise from nearly $1.5 trillion in 2025 to over $2 trillion in 2026 [1] Hyperscaler capital expenditure alone is on track to exceed $600 billion, a 36% year-on-year increase, pushing capital intensity to between 45% and 57% of revenue for the major cloud providers.[2] [3] [4]
At the foundation model layer, the cost dynamics are striking. OpenAI reported $20 billion in annualized revenue at the end of 2025, rising to $25 billion by February 2026 - yet the company projects a $14 billion net loss for 2026 and does not expect to turn cash-flow positive until 2030.[5] [6] Anthropic has committed to data centre infrastructure costing $50 billion and saw its annualized revenue surge from $1 billion at the start of 2025 to $30 billion by April 2026; a trajectory that defies early scepticism, though the underlying cost structure remains substantial.[7] [8] These dynamics reflect a race for strategic position funded by the expectation of future economics that are only beginning to arrive for some players, and have not yet arrived for others.
If a correction comes - a consolidation at the model layer, a pullback in hyperscaler spending, a repricing of foundation model access - it will be real and it will be visible. But it will be concentrated where the capital is: in data centres, GPU supply chains, and the foundation model providers themselves.
Why enterprise AI value creation will continue
The value is real and largely uncaptured
The case for AI as a genuine enterprise value driver does not depend on any particular model provider surviving, on hyperscaler spending continuing at current rates, or on the current generation of tools remaining dominant. It depends on something far more durable: the structural opportunity to automate, augment, and improve how organisations operate.
That opportunity is not theoretical. Individual users in leading organisations are already seeing productivity gains of up to five times.[9] Average ROI per dollar invested in generative AI stands at 3.7x for those who have genuinely deployed it.[10] . The technology works. The value is real.
What is equally real is how little of that potential has been captured at scale. Enterprise-wide AI deployment remains the exception, achieved by fewer than one in fifteen organisations.[11] Significant ROI from generative AI is reported by less than a third of those running it.[12] Nearly half of AI transformation programmes are abandoned before reaching production. The share of companies scrapping the majority of their AI initiatives more than doubled in a single year, from 17% to 42%.[13] Revenue growth from AI remains an aspiration for most: 74% of organisations hope to grow revenue through AI in the future, compared to just 20% already doing so.[14]
This gap between potential and realised value is not caused by the infrastructure bubble. It is caused by something enterprises can directly address: weak data foundations, absent governance structures, strategies built around tool access rather than business outcomes, and operating models that were not designed to capture AI value at scale.
The reasons are well-documented. Individual productivity gains are real, but they are not translating into process change. Usage remains optional rather than embedded. AI outputs are not being integrated into core systems or decision-making. The result is a growing class of AI super-users delivering outsized individual results, while the organisations around them remain structurally unchanged.[15] [16]
A bubble correction at the infrastructure and model layer would not close this gap. If anything, it might widen it temporarily, as organisations use market uncertainty as a reason to delay decisions they should have made already. The enterprises that continue to build the right foundations will simply have more of an advantage when conditions stabilise.
What a correction would and would not change
Separating signal from noise
It is worth being precise about what a market correction in AI infrastructure and foundation models would actually mean for enterprise buyers.
What would likely change:
Model access pricing, potentially significantly, as providers move toward sustainable unit economics
The number of viable foundation model providers, as consolidation follows over-investment
The pace and scale of new infrastructure investment, creating potential capacity constraints
Investor appetite for AI-adjacent businesses without clear paths to profitability
What would not change:
The fundamental capability of AI to automate and augment enterprise processes
The structural opportunity to generate value from better data, better processes, and better decision-making
The competitive disadvantage facing organisations that have not built the foundations to deploy AI effectively
The direction of travel: AI capabilities will continue to improve, inference costs will continue to fall, and the range of deployable use cases will continue to expand
The dotcom correction did not make the internet less valuable. It made overpriced, poorly-founded internet businesses less viable. The same logic applies here. What will survive and strengthen is not the infrastructure bet or the model layer race. It is the enterprise value creation that good AI strategy, good data foundations, and good operating models make possible.
The risks that are actually within your control
Governance, Foundations, and Strategy
While the infrastructure bubble debate plays out at a level most enterprises cannot influence, there are material risks that sit directly within their control and that are accumulating quietly in many organisations.
Most executives (67%) believe their company has already suffered a data breach caused by unapproved AI tools. More than a third (36%) have no formal plan for supervising AI agents, and a similar proportion admit they could not immediately disable a rogue one if they needed to. Three quarters of executives acknowledge that their company’s AI strategy is, in their own words, ‘more for show than actual guidance.’ [17] Nearly half (48%) describe AI adoption as a massive disappointment - a figure that has grown sharply in a single year.[18]
These are not indicators of a technology failing to deliver. They are indicators of adoption running ahead of the foundations required to make it sustainable. Organisations that have moved fast without those foundations are accumulating governance risk, vendor lock-in, and cost structures that do not reflect the true economics of AI at scale. They also reflect a broader pattern in which organisations move fast because the technology is accessible and the barrier to experimentation is low, without having defined which use cases genuinely create business value and where focused investment will actually make a difference. Only 7% of organisations report their data is fully ready for AI.[19]
A further dimension that often goes unnoticed is pricing transparency. When model access has been priced at levels disconnected from underlying unit economics - as it has been during the investment race - enterprise buyers construct business cases on assumptions that may not remain valid. Organisations that have not built a clear view of their true AI cost base will face difficult recalibrations as pricing normalises, whether or not a broader correction materialises.
What this means for enterprise strategy
Building for durability, not for the cycle
The organisations building durable AI advantage are not the ones making the biggest bets on which model provider will win or which infrastructure stack will dominate. They are the ones that have answered a more fundamental question: how does AI create value for this business, under what conditions, and with what foundations in place to capture it?
That means a differentiated AI strategy tied to specific business outcomes rather than technology access. It means data architectures and governance frameworks that enable AI deployment at scale rather than constraining it. It means operating models built to sustain value creation, not just to run pilots. And it means vendor-neutral positions that preserve flexibility as the model landscape inevitably evolves.
The bubble, if it comes, will be a noise event for enterprises that have built on the right foundations. For those that have not, the correction will not be the cause of their difficulties, it will simply make existing weaknesses harder to ignore.
MULTIPLAI: How we help
At MULTIPLAI, we work with organisations navigating exactly this environment. Our focus is on the foundations that make AI investment durable regardless of how market conditions evolve: a differentiated strategy tied to clear business objectives, robust data and technology foundations, sound governance, and scalable operating models backed by the organisational capability to execute.
We do not start from the technology. We start from the question of strategic value: how AI can drive the business forward, what outcomes are being created, for whom, under what conditions, and with what level of risk. That framing allows us to help clients build AI programmes that are credible to boards, manageable under regulatory scrutiny, and resilient to the market pressures that are reshaping the AI landscape.
Our support does not stop at programme design. We work alongside organisations through delivery itself, because building AI at scale is operationally complex and requires the kind of cross-functional experience that only comes from having done it before.
Coming up in this series
Data Readiness: Why data quality and architecture are the silent prerequisites that determine whether AI delivers and how to assess where your organisation stands.
Security & Sovereignty: How to build AI systems you can actually control, from data residency and access management to regulatory compliance in cross-border environments.
AI Governance: What effective oversight looks like in practice, and why it is one of many determinants of whether AI investments hold up over time.
Vendor-Neutral Technology: Why lock-in is the hidden risk in most enterprise AI deployments, and what a genuinely flexible architecture looks like in a market that is still moving fast.
Differentiated AI Strategy: Why the organisations that capture lasting value are those with a clear, business-led AI strategy - not just access to the latest tools - and how to build one that holds up as the market evolves.
-> Follow MULTIPLAI on LinkedIn to stay updated on upcoming posts in this series.
Comments