From the outside, artificial intelligence looks like limitless growth.  I see headlines online about how revenue projections will skyrocket, hardware demand is going to accelerate, and market announcements read like victory laps. But from inside the data and AI industry, the financial structure underneath American AI looks far less like a stable platform and far more like a precarious tower of cross-investments, interdependence, and capital risk layered on top of one another.  OK, so it may be my pattern-matching skills just talking here, but I know I’m not the only one, so I’ll get back to the patterns that have become so well-developed.  I use AI almost daily (it’s built into almost all applications, no matter if it is desktop, mobile, etc.) but I am always asking myself to:

1. Justify the value of the AI feature I’m using vs. previous means
2. Justify the financial cost to gain the AI feature
3. Justify the amount of resources (human, financial, energy) AI cost for the feature

AI (at least American AI) today is not a free market, but a closed financial loop dominated by a small number of infrastructure owners who are simultaneously suppliers, partners, customers, and investors in each other’s success. GPU manufacturers depend on hyperscalers. Hyperscalers depend on AI vendors. AI vendors depend on enterprise buyers. Everyone depends on regulatory tolerance and cheap capital.  It’s like watching a football league where there’s only an imaginary opposing team and we’re rooting for an unforeseen championship that if done wrong could destroy our civilization, (or at least our way of living, economy and environment.)

The problem that I continue to struggle with is not innovation, but who has leverage.  The AI industry has quietly constructed a capital stack with too many mutual dependencies and too few independent cash flows. When AI profitability hiccups, the financial impact does not land in one place. It cascades across the entire championship and everyone loses and not all players are equally positioned to survive it.  This is the AI Bubble in a foundational nutshell.

The AI Financial Stack (Simplified)

At a technical and financial level, AI infrastructure breaks into four layers:

1. Hardware: GPU vendors, chip manufacturers, energy suppliers.
2. Cloud Infrastructure: Hyperscalers providing compute and storage.
3. Model Builders: Companies training foundation models.
4. Enterprise Software Vendors: Firms trying to commercialize AI into tools, platforms, and products.

There are a lot of big players, no matter if we talk about NVIDIA, AMD, Microsoft, Google, Meta, AWS, Oracle, OpenAI, Anthropic and xAI.  We also have speculators, funding AI financially, technically or by marketing, yet not really vested for the long haul, such as Palantir, JP Morgan Chase, Hugging Face, Arm, Softbank and even analytics companies like Snowflake and Databricks.  Some of the players are more leveraged than others, but also the cross-investment in many of them is highly concerning, as is the risk many of them are taking to get to the finish line as one of the first.  What makes this fragile is that money flows upward in the stack, while risk flows downward…at least that’s how it normally works.

Cloud providers buy hardware in bulk under capex-heavy models. Model builders burn cash on training. Enterprise vendors over-market AI features that generate minimal revenue relative to infrastructure expense. Yet investors value all layers as if sustainable profit exists everywhere, but it doesn’t happen here in the world of AI.  Currently, only the infrastructure layer is reliably monetized and everything above it runs at a margin deficit.

One of the biggest challenges for many to wrap their head around is the inter-investment and revenue sharing of the top companies involved in the AI Bubble, realizing the inter-dependence and risk associated.  Due to this complexity, I’m hoping it may be simpler if we look at just one, vulnerable layer and yes, this is where I’m going to get in trouble, but when has that stopped me before?

Oracle, and Why It’s Vulnerable

It may look like I’m about to pick on Oracle, but I’m not.  I love the Oracle database and Oracle tech, but Oracle is a critical case study because it represents something different than the hyperscalers: an infrastructure-adjacent enterprise software company trying to rebuild growth through AI and cloud expansion, while not actually owning the AI value chain.  It’s not the only one of those involved in the AI Bubble I could make a case from, but it is unique, I’ve observed it closely for years and I know it quite well.

Oracle’s total debt accrued in its goal of achieving AI workloads is roughly around $100-116 billion.  This total includes several recent bond issuances and borrowing around AI/cloud data center expansion.  I included its AI-related capital and borrowing, at least the material portion of its capital expenditures(CapEx) which is forecasted at $50 billion for fiscal year 2026 and is tied to its Oracle Cloud Infrastructure(OCI) build out to support AI workloads.  I also can add another $25-30 billion in bonds and a few market sources that indicated another $38 billion in new debt from financial feeds, (glad to pull up the links for anyone interested…)

Oracle’s exposure is not primarily financial or ideological, but structural.  Oracle is trying to compete in a hyperscaler world without hyperscaler economics the others could leverage.  I know many will argue that Oracle has Oracle Cloud Infrastructure (OCI) but stay with me here:

1. OCI Is Capital-Intensive Without Hyperscaler Advantages

Oracle lacks what AWS, Microsoft, and Google have: a dominant OS, a consumer ecosystem, a first-party AI platform, social data pipelines, and/or an ad-supported SaaS engine.
Because of this, OCI is pure infrastructure risk. Its revenue depends almost entirely on enterprise workloads, including databases, Exadata migrations, and Oracle applications, which is everywhere, but currently under scrutiny by most of its customers.

AI GPU clusters are expensive, short-lived assets, and Oracle is scaling them at a time when cloud budgets are tightening, multi-cloud reduces lock-in, and customers avoid long-term compute commitments.  OCI margins are slimmer than they appear: GPU purchases are up-front costs, power and  cooling rise faster than revenue, hyperscalers undercut pricing, and customers demand portability. Even with Exadata in every hyperscaler, Oracle still rents space in ecosystems it doesn't control.

2. Oracle Consumes AI; It Doesn’t Produce It

Oracle does not own a dominant foundation model like OpenAI, Copilot, or Gemini, nor does it control the AI training stack or developer ecosystems that shape pricing and influence. Oracle 23ai/26ai improves the database story, but Oracle still sells infrastructure and applications around AI, not the AI itself.  The economics are upside-down as Oracle pays the infrastructure bill, AI vendors capture the margin, customers expect AI “included,” and Oracle absorbs the compression.

Enterprise buyers already view Oracle licensing as a target for cost reduction, making this an even harder sell.

3. Oracle’s Enterprise Base Adopts AI Slowly

Oracle’s customers (which include government, finance, healthcare, insurance, and manufacturing) are risk-averse, compliance-heavy, and slow to adopt new technology. They resist exactly the models Oracle needs to monetize AI: per-call pricing, embedded AI licensing, deeper platform dependency, and vendor-controlled data flows.  This slows ROI at a time when markets expect rapid AI revenue. That mismatch creates earnings pressure and drags Oracle into regulatory and governmental scrutiny during a period when agility matters most.

Who Can Pivot More Easily in an AI Correction?

Companies with structural resilience, such as GPU vendors, hyperscalers with ad revenue, platforms with consumer lock-in, data monopolies, and defense-aligned firms can raise prices, shift investment, pass costs through, or secure government funding.  Oracle cannot raise prices without accelerating customer churn, and the market is full of alternatives. This doesn’t even touch the debt load Oracle must assume to fund its current AI and cloud timelines, but enough about poking holes in the AI Bubble at Oracle’s expense.  It’s not fair to them, as they are just part of the huge AI bubble problem, not the source of it.

Then There’s the Impending AI Bailout

So, if I haven’t convinced you why the AI Bubble is a real thing, the other is me looking into the crystal ball and seeing the future of a government bailout of AI.  If AI infrastructure collapses, governments will have to intervene because:

• AI underpins government systems, (don’t even get me started on how dependent they are on Oracle.)
• AI now touches healthcare infrastructure
• AI powers national cyber operations
• AI is treated as strategic capital, not tech optionality

Although some news will report on AI companies and projects receiving a bailout, most of it will be in smaller initiatives that will be used to save our economy, (because the few companies that are all cross-invested are over 20% of the total U.S. stock market value…) If you doubt how volatile the markets are right now, Oracle suffered a major hit just this week.

What will most likely occur if the bubble does burst is the following:

• AI incentive bills
• Infrastructure subsidies
• Tariff adjustments
• “Innovation protection policies”
• Government cloud contracts
• Defense modernization programs

I wanted to make sure none of my pattern matching skills were off on the big picture I was seeing, so I spent a long time researching this with and without AI to understand the weak points and opportunities in the AI bubble.  Over 90% of what I thought I was seeing, at a high and mid-level is concerning to many in the financial/tech sector and something everyone should think about as we build out the future of AI.

The most sobering thing I confirmed is who really pays for the above, and in the end, it’s we, the taxpayer.

• Tariffs are inflation that is passed onto the consumer, which is the American taxpayer.
• Government contract funnels reward size, not efficiency.
• And consolidation accelerates this problem.
• No one breaks up the monopolies, even though you’d think we’d have learned from history.
• A few AI companies will make it big.
• A very large group will go under, which is the norm in an economic bubble.
• And the average taxpayer gets hollowed out paying for it all in the end.

Who Becomes the AI Cost Absorbers?

In every tech bubble, someone becomes the cost sink.

• ISPs in the dot-com era.
• Device manufacturers in mobile.
• Miners in crypto.

Oracle sits closer to the cost sink than most believe and it’s not that I wish bad on any company, I just think it’s important to learn from history and make smarter decisions in the future.  There’s always those that think they can either make a big win, never minding the devastation in their wake or that they don’t need to think about how to do things the right way.  It’s always about the race to the finish for these groups and we do have an AI race well underway.

Oracle has a lot of financial capital and a lot of data gravity, but Oracle does not control the gravity of AI in this race, which makes it the weakest link.

The Bubble Reality Check

AI is not sustainable as it is today. It is subsidy-driven, capital-heavy, and dependent on public tolerance and public and private data (yes, I said private data, too.)  The cross-investment structure looks like stability if you don’t look to close, but should not be confused as such.  It is feedback risk disguised as partnership and has more in common with a house of cards someone has glued together vs. a true foundation you’d want to build a real home on.   If that glue lets go and the house gives way, the companies closest to infrastructure, without data power or distribution control will feel the impact first.  This will happen not because they failed, but because they were positioned incorrectly when the glue (gravity) shifted.

American AI won’t crash in a headline, but most likely deflate quietly through:

• Margin erosion
• Contract renegotiation
• Government intervention
• Quality/Value reduction
• Product bundling
• Reduced innovation

And the public will pay for it:

• Once with a choice.
• Continually with their money.
• Continually with their data.

Pattern matching or opinion, this is where my editorial lead me to this week.

What are your thoughts on the AI Bubble? 

As valuable as AI is becoming and how much integration is geared to every aspect of our day-to-day life, do you worry about the AI Bubble and the financial situation created to compete? 

Do you feel it’s more about the revenue of a few vs. the benefit of all?