Sony Kumar's BLOG

By Sanjeev Kumar

Over the last three years, artificial intelligence has moved from a technical discipline into the centre of global imagination. Suddenly, everyone is an AI expert. Everyone is building an AI company. Everyone is raising capital for AI. Everyone is predicting the singularity.

But beneath the noise, something deeper — and far more consequential — is happening.

We are not just building AI systems. We are building a global financial architecture around AI.

And if we are not careful, this architecture may carry systemic risk larger than the 2008–09 financial crisis.

I say this calmly, academically, and with the same clarity that helped me see the 2008 collapse before it happened.

Because the pattern is repeating — only amplified.

1. Capital Has Flooded Into AI at a Scale Humanity Has Never Seen

AI is now the largest capital magnet in human history.

• $250–300B annual global AI investment
• $1T+ in GPU purchase commitments
• $2T+ in cloud commitments
• $500B+ in corporate debt tied to AI infrastructure
• $852B OpenAI valuation
• $965B Anthropic valuation
• $3T NVIDIA valuation

These numbers are not normal. They are not sustainable. They are not grounded in traditional economics.

They are grounded in narrative.

2. Why the AI Narrative Is Being Sold So Aggressively

The AI narrative is not being sold because AI is close to consciousness or autonomy. It is being sold because the narrative itself attracts capital.

Narrative has become:

• a financing tool
• a valuation driver
• a geopolitical instrument
• a market accelerator
• a hedge against future uncertainty

In other words:

AI narrative is being used to pull forward capital that the underlying economics cannot yet justify.

Narrative is not decoration. Narrative is infrastructure.

And today, narrative is doing more work than economics.

3. Western AI Labs Are Valued As If They Will Dominate the World

OpenAI and Anthropic valuations assume:

• global market capture
• global enterprise dominance
• global cloud reliance
• global model licensing
• trillions in future revenue

But these assumptions are fragile — because the world is changing faster than the narrative.

4. China Has Entered the Arena — And It Is Playing a Different Game

China’s leading AI companies (Baidu, Alibaba, Tencent, ByteDance, iFlytek, SenseTime) are scaling:

• faster
• cheaper
• with state backing
• with lower regulatory overhead
• with access to massive domestic datasets
• with lower energy and infrastructure costs

And now Beijing has made a decisive move:

China is helping developing countries build their own AI infrastructure.

This is not charity. This is geopolitical architecture.

China is exporting AI infrastructure the same way it exported roads, ports, and telecom networks through the Belt and Road Initiative.

This changes the global landscape.
If China captures:

• Africa
• Southeast Asia
• Latin America
• Middle East

…then Western AI companies lose entire continents of future revenue.

Their trillion‑dollar valuations become mathematically impossible to justify.

5. Western AI Expansion Is Being Financed Through Massive Debt

This is the part almost nobody is talking about.

Oracle

• ~$90B total debt
• ~$30B new debt since 2023
• Used for GPU clusters, data centres, energy infrastructure

Amazon

• tens of billions in new debt for AWS AI expansion
• Anthropic commitments structured as cloud credits + debt-backed infra

Microsoft

• $13B+ OpenAI commitment structured as compute credits
• multi-year GPU financing agreements

Meta, Google, Tesla

over $150B in new debt since 2023

used for data centres, GPUs, energy, AI labs

This is industrial-scale leverage.

AI is not software. AI is infrastructure.

Infrastructure requires debt.

6. NVIDIA Has Begun Financializing GPUs — Turning Chips Into Yield-Bearing Assets

This is historically unprecedented.

NVIDIA is partnering with Wall Street to position GPUs as:

• yield-generating assets
• securitisable commodities
• collateral for loans
• tradable financial products

This includes:

• GPU-backed loans
• GPU leasing funds
• securitised GPU pools
• GPU-as-a-service yield products

This is clever. This is rational. This is risky.

Because GPUs are not houses. They depreciate in 2–3 years. They become obsolete overnight. They depend on energy prices. They depend on narrative momentum.

This creates asset-liability mismatch — the same structural flaw that caused 2008.

But this time the underlying asset is far more volatile.

7. Anthropic’s CEO Has Already Said the Quiet Part Out Loud

Dario Amodei has publicly stated that Anthropic may need:

“trillions of dollars in annual revenue”

…to become sustainably profitable.

This is not hyperbole. It is economic reality.

Because Anthropic’s cost structure includes:

• 56% compute cost ratio
• multi-gigawatt energy requirements
• multi-billion-dollar training cycles
• massive safety overhead
• massive cloud commitments

This is not software economics. This is industrial economics.

And industrial economics do not scale cleanly.

8. The Equation That Predicted 2008 — And Predicts Today’s AI Risk

In 2008, the collapse was predictable using a simple structural equation:

$$\text{Systemic}\text{Risk}=L\times V\times N$$

Where:

• L = Leverage
• V = Asset Volatility
• N = Narrative Dependency

Let’s apply it calmly to AI.

Leverage (L) — Very High

GPU-backed securities emerging

$500B+ corporate debt

$1T+ GPU commitments

$2T+ cloud commitments

sovereign wealth financing

GPU-backed securities emerging

Asset Volatility (V) — Extremely High

• GPUs obsolete in 24–36 months
• energy dependency
• model architecture shifts
• China’s cheaper alternatives
• global competition

Narrative Dependency (N) — Maximum

• belief in infinite AI demand
• belief in global dominance
• belief in trillion-dollar revenue
• belief in perpetual scarcity
• belief in Western supremacy

Now multiply:

$${\text{Systemic}\text{Risk}}_{AI}=\text{Very High}\times \text{Extremely High}\times \text{Maximum}$$

This is how we calmly show:

The implosion potential is structurally larger than the 2008–09 crisis.

9. Rising Government Debt and Debt‑Servicing Costs Make Any Crisis Unmanageable

Across the Western world, government debt has reached levels that make crisis response almost impossible — not just because the debt is high, but because the cost of servicing the debt has become unsustainable.

United States

• Federal debt: ~$34.8 trillion
• Annual interest payments: $1.1 trillion
• Debt-to-GDP: ~123%

The U.S. now spends more on interest payments than on defense.

United Kingdom

• Public debt: ~£2.7 trillion
• Debt-to-GDP: ~103–105%
• Annual interest payments: £110–120 billion

The UK now spends more on debt interest than on policing, transport, or housing.

Governments have no fiscal shock absorbers left.

If an AI-driven financial correction occurs, they cannot intervene.

This is why the risk is larger than 2008.

10. And Here Is My Deepest Worry — The Human One

As someone who did get the 2008–09 financial crisis right, I do genuinely worry that we are blindly falling in love with AI financing

I say this without emotion, without drama, and without any desire to sensationalise the moment. I say it as someone who has seen this pattern before.

In 2006–07, the world fell in love with mortgage financing. Today, we are falling in love with AI financing in exactly the same way — blindly, aggressively, and without structural awareness.

And therefore, it is important to calmly, academically, structurally consider all the downside risks — because if the AI financial bubble collapses:

The AI heroes of today may become the villains of tomorrow.

Not because they intended harm. Not because they acted maliciously. But because the system around them was built on:

• leverage
• narrative
• fragility
• unsustainable debt
• geopolitical competition
• financialisation of physical assets

And when systems collapse, society looks for someone to blame.

But the deeper danger is this:

• Western governments will not have the fiscal capacity to respond
• corporate debt tied to AI infrastructure will become unserviceable
• GPU‑backed financial products will unwind violently
• sovereign wealth funds will face catastrophic losses
• pension systems will be destabilised
• banks will face liquidity shocks
• global markets will fracture
• social trust will collapse
• political institutions will weaken
• public services will fail

And the most painful truth is this:

Average people are not getting richer from the AI boom.

They are spectators, not beneficiaries. They are watching wealth concentrate at the top while their own lives do not improve.

If the system collapses under the weight of AI financialization, society may implode under its own pressure — and in that vacuum, the world may turn, out of desperation, to the very AI systems that contributed to the collapse.

Not because AI is ready. Not because AI is safe. But because society may not have any functioning alternatives left.

This is how heroes become villains. Not through intention. Through systemic failure.

11. A pragmatic conclusion

AI is real. AI is powerful. AI is transformative.

But the financialisation of AI — not AI itself — is where the systemic risk lies.

Western AI companies are valued as if they will dominate the world. China is building an alternative AI ecosystem across the developing world. Corporations are issuing massive debt to finance AI infrastructure. GPUs are being turned into yield-bearing financial assets. Valuations assume trillions in future revenue. Narratives are driving capital faster than economics can sustain. Government debt is rising to levels that make crisis response impossible. Debt-servicing costs are consuming national budgets. Society is not benefiting from the boom. At least not yet.

If we do not manage this carefully, the correction will not be small. It will not be local. It will not be contained.

It will be global, structural, and five times larger than 2008–09.

And in that collapse, the AI heroes of today may be remembered as the villains of tomorrow — not because they intended harm, but because the system around them was built on leverage, narrative, and fragility.

I do think that it is time the world starts to prepare for downside risk. Not out of fear — but out of responsibility.

And we must also acknowledge a simple, uncomfortable truth: society has not gotten rich from the AI boom — not yet

For all the noise, all the headlines, all the valuations, all the GPU shortages, all the capital inflows, all the trillion‑dollar narratives — the average person on the street has not become richer.

Not in London. Not in New York. Not in San Francisco. Not in Berlin. Not in Bangalore. Not anywhere.

Wages have not risen because of AI. Savings have not increased because of AI. Housing has not become more affordable because of AI. Public services have not improved because of AI. Financial security has not strengthened because of AI.

The AI boom has created paper wealth, not shared prosperity.

It has enriched:

• investors
• founders
• early employees
• GPU suppliers
• cloud providers
• financial intermediaries

But it has not yet enriched society.

And this matters — deeply — because when a financial bubble grows without lifting the population beneath it, the social fabric becomes fragile.

People tolerate risk when they benefit from the upside. They do not tolerate risk when they carry only the downside.

This is why I would say:

This is not fear — not at all. It is a genuine attempt to bring a bit of pragmatism to the table. And it is time the world starts to prepare for downside risk.

Not because collapse is inevitable. Not because AI is dangerous. Not because innovation should stop.

But because prudence is not pessimism. Pragmatism is not fear. Structural awareness is not negativity.

It is simply responsible.

And if we want AI to truly uplift humanity — not just capital markets — then we must ensure that the financial architecture around AI is stable, sustainable, and grounded in reality.

Because society has not gotten rich from the AI boom yet. And if the bubble bursts before society benefits, the consequences will not be technological — they will be human.