AI Can’t Afford to Slow Down. It’s Absorbing Capital, Talent & Strategic Resources Faster Than Markets Can Replenish Them. Who Still Knows Where the Money Is Coming From—or What Gets Crowded Out Next?
Goldman Sachs is expanding its AI engineering workforce while helping finance an industry consuming unprecedented amounts of capital. AI-related debt is approaching $500 billion this year, the BIS warns that financing is increasingly opaque and interconnected, and Goldman and 20 other financial institutions are preparing a dollar stablecoin. Separately, FATF says professional money-laundering networks are becoming harder to track. None of this proves illicit capital is funding AI. It raises a more important risk question: as the AI race becomes too expensive to slow, can anyone still see where all the money, talent and strategic resources are coming from—and what happens when the race starts competing with the rest of the economy for them?
Something strange happens when a technological race becomes too important to lose.
At first, companies compete for:
customers.
Then they compete for:
talent.
Then:
chips.
Then:
data centers.
Then:
electricity.
Then:
critical minerals.
Then:
capital.
Eventually the race stops looking like an ordinary technology market.
It starts looking like a contest to control every scarce input required to remain competitive.
That may be where AI is heading now.
Every New Leap Requires Another Resource Round
A frontier AI company makes its model significantly more capable.
That sounds like technological progress.
But underneath the announcement sits another bill.
More training.
More inference.
More GPUs.
More data-center capacity.
More electricity.
More networking.
More engineers.
More financing.
Then the competitor responds.
Another model.
Another chip.
Another data center.
Another financing round.
Another infrastructure commitment.
The economics create a brutal question:
Can anyone in this race afford to be the first company that says, “We have spent enough”?
Reuters captured that tension this week.
Researchers inside frontier laboratories are increasingly questioning whether AI development is moving too quickly.
But Reuters also noted that enormous financial stakes make slowing difficult.
OpenAI needs financing for the compute required to continue advancing its models.
Anthropic is preparing for a potential public offering that could value it at extraordinary levels.
Every major capability gain can strengthen the case for:
the next funding round,
the next infrastructure commitment,
or the next IPO.
This isn't simply technological competition anymore.
It is capital-dependent acceleration.
Goldman Sachs Is Now Hiring More Engineers
Reuters reported that Goldman Sachs opened a dedicated engineering office in Bellevue, Washington, with more than 125 employees focused on AI and cloud transformation.
Goldman already employs more than 12,000 engineers.
That is roughly one-quarter of its workforce.
Pause there.
Goldman Sachs is one of the world's most important financial institutions.
It underwrites securities.
Advises companies.
Structures financing.
Participates in IPOs.
Lends.
Invests.
Moves capital.
Now a quarter of its workforce consists of engineers.
That is not simply an HR story.
It tells us something about where financial power believes the next competitive advantage sits.
AI Isn't Only Replacing Labor. It Is Concentrating Different Labor.
The popular AI labor narrative goes like this:
AI eliminates jobs.
That is part of the story.
But another labor transition is occurring at the same time.
Capital-rich institutions are competing aggressively for scarce technical talent.
AI engineers.
Cloud engineers.
Cybersecurity experts.
Chip designers.
Model researchers.
Infrastructure specialists.
People carrying deep institutional knowledge.
We've already seen extraordinary talent migration between major technology companies.
And that creates an overlooked form of AI risk:
AI Talent Concentration Risk
AI Talent Concentration Risk occurs when the organizations with the deepest capital pools absorb disproportionate shares of scarce technical expertise, increasing their own capability while simultaneously reducing the talent available to competitors, smaller firms, governments and other sectors of the economy.
The strategic advantage comes from both sides.
You gain the person.
Your competitor no longer has the person.
And that can become even more consequential when the person arrives carrying years of accumulated industry knowledge.
Talent Can Function Like Intellectual Infrastructure
A talented engineer does not merely bring:
coding ability.
They bring:
patterns,
experience,
failed approaches,
architecture knowledge,
supplier knowledge,
institutional memory,
and an understanding of what worked and what did not.
No employee should be assumed to carry proprietary information illegally simply because they change jobs.
But when companies recruit large numbers of people from competitors, something perfectly legitimate still happens:
knowledge migrates.
AI may make that knowledge easier to:
organize,
connect,
model,
test,
and synthesize.
One employee has one piece.
Another employee has another.
Another understands a supplier.
Another understands architecture.
Another understands deployment.
AI can increasingly help connect fragmented knowledge into a more complete map.
That creates:
AI Knowledge Aggregation Risk
AI Knowledge Aggregation Risk occurs when individually lawful fragments of expertise, experience and institutional knowledge become substantially more strategically valuable after AI systems combine them into patterns or capabilities that no individual contributor possessed alone.
That is different from stealing a trade secret.
It is about the increasing value of aggregation.
Now Look Beyond Talent
The same concentration is happening across almost every strategic AI input.
Critical minerals.
The United States, Europe and China are competing for access.
Chip supply.
Nvidia sits at the center of an extraordinary global infrastructure buildout.
Data-center capacity.
OpenAI and a small number of hyperscalers are becoming anchor customers for enormous projects.
Electricity.
AI data centers are competing with industrial users and households for grid capacity.
Debt.
Hyperscalers have issued more than $200 billion of debt in 2026, more than double the prior year's total.
And Reuters reported AI-related debt issuance had reached nearly $500 billion by early August, representing about one-fifth of higher-rated U.S. issuance this year, compared with roughly 1% in 2024.
This is no longer isolated resource competition.
It is systemic.
This Is AI Resource Capture Risk
AI Resource Capture Risk
AI Resource Capture Risk occurs when the race for frontier AI causes leading companies and their financial backers to absorb disproportionate shares of scarce capital, engineering talent, compute, energy, minerals and infrastructure capacity in order to maintain competitive position.
The key word is:
maintain.
Because today's capability advantage may not last long.
A company spends billions to move ahead.
A competitor catches up.
The company must spend again.
The result can become:
capital
→ capability
→ temporary advantage
→ competitor response
→ additional capital
→ additional capability
→ another temporary advantage.
That is an expensive loop.
What Happens When Improvement Itself Creates Another Funding Requirement?
This is where the economics become fascinating.
Normally, technological improvement should eventually reduce dependence on outside capital.
A company develops a successful product.
Revenue grows.
Margins improve.
Cash flow begins financing future growth.
But frontier AI still requires enormous infrastructure investment.
The Bank for International Settlements now estimates that the world's five largest technology companies will spend more than $1 trillion on AI in 2025 and 2026, while global AI investment could rise toward $4 trillion by 2030.
The BIS also says AI financing is increasingly shifting toward:
debt
and
private credit
rather than being funded entirely from corporate earnings.
And it specifically warned that much of that financing is:
opaque and interconnected.
That phrase should matter enormously to AI investors.
Opaque and Interconnected
Consider what those words mean together.
Opaque means outsiders may struggle to understand:
who ultimately carries the exposure,
what collateral supports it,
how leverage moves between vehicles,
and where losses migrate.
Interconnected means the same institutions may appear across:
lenders,
equity investors,
infrastructure funds,
suppliers,
customers,
guarantors,
and counterparties.
The risk is not automatically fraud.
The risk is:
observability.
Can investors actually see the complete financial architecture supporting AI expansion?
This Is AI Capital Observability Risk
AI Capital Observability Risk
AI Capital Observability Risk occurs when AI financing spreads across public debt, banks, private credit, infrastructure vehicles, strategic partnerships and other funding structures faster than investors and regulators can reconstruct where leverage, concentration and ultimate economic exposure actually sit.
This is where the view from above becomes important.
Looking at one transaction:
reasonable.
One data-center loan.
One private-credit deal.
One IPO.
One strategic investment.
One chip financing agreement.
One bond issue.
But step upward.
How many are ultimately supported by:
the same AI customers,
the same projected compute demand,
the same suppliers,
the same infrastructure assumptions,
and the same expected future AI productivity?
That's the risk.
AI Is Already Competing With Everyone Else for Capital
This does not occur in an infinite financial system.
Capital has alternative uses.
Governments borrow.
Small businesses borrow.
Home buyers borrow.
Industrial companies borrow.
Infrastructure companies borrow.
AI companies borrow.
Reuters has reported that massive AI investment and heavy government borrowing are both contributing to unusually strong demand for capital, potentially helping keep real interest rates structurally higher.
That creates another AI externality.
The AI company does not need to borrow from you directly to affect your financing costs.
If hyperscalers absorb enormous amounts of bond-market capacity, the marginal cost of capital elsewhere can rise.
This Is AI Capital Crowding Risk
AI Capital Crowding Risk
AI Capital Crowding Risk occurs when unusually large AI financing requirements compete with governments, households and non-AI businesses for finite investor balance-sheet capacity, potentially raising borrowing costs or reducing access to capital elsewhere in the economy.
The AI company receives the compute.
Everyone may participate in the financing consequence.
That is why AI risk is becoming macroeconomic.
But What If Traditional Capital Eventually Isn't Enough?
This is where the question becomes legitimate.
NOT as an allegation.
As a stress test.
Suppose AI investment really does approach several trillion dollars.
Suppose model competition does not slow.
Suppose every capability leap triggers:
more compute,
more data centers,
more power,
more chips,
more acquisition,
more talent,
and more financing.
Where does all of that capital come from?
Banks?
Bond markets?
Private equity?
Private credit?
Sovereign funds?
Pension funds?
Retail investors through IPOs?
Foreign capital?
Digital financial markets?
Probably some combination.
And that means the financial perimeter around AI keeps expanding.
Goldman and Other Major Banks Are Building New Digital Money Rails
Reuters reported September 1 that 21 financial institutions, including:
Goldman Sachs,
Bank of America,
Citi,
and Deutsche Bank
plan to create a company that will issue a dollar-pegged stablecoin in 2027.
The group also intends eventually to support other G7 currencies.
A separate consortium of 37 financial institutions is developing a euro stablecoin.
Stablecoins can move value quickly across digital networks.
They can reduce settlement friction.
They may eventually become important infrastructure for mainstream finance.
That does not make them suspicious.
But their rise changes the architecture through which money can move.
And whenever a financial architecture changes rapidly, risk intelligence should ask:
What new visibility problems arrive with the efficiency?
Faster Money Creates a Provenance Problem
This is where FATF becomes relevant.
Not because FATF said AI companies are laundering money.
It didn't.
Reuters reported that the Financial Action Task Force warned that criminal misuse of informal money-transfer networks has become a growing global problem.
Professional laundering networks increasingly exploit systems outside normal banking oversight.
FATF is asking governments and private companies to improve detection and international cooperation.
These are separate developments.
But place them beside each other:
AI requires extraordinary amounts of capital.
Finance is increasingly digitizing.
Stablecoins can move money globally.
Private credit can be opaque.
Informal financial networks are becoming harder to police.
AI startups and infrastructure companies are increasingly moving enormous amounts of money through private markets before reaching IPO.
That does NOT prove illicit money is entering AI companies.
But it does justify a question:
How good is AI investment capital provenance?
AI Capital Provenance Risk
AI Capital Provenance Risk occurs when the speed, complexity and internationalization of AI financing make it increasingly difficult for investors, regulators or counterparties to determine the ultimate origin, path, concentration and beneficial ownership of capital flowing into strategically important AI companies and infrastructure.
This is not an accusation.
It is a control question.
If we are going to finance one of the most strategically important technologies in history through increasingly diverse global capital channels, we should know:
Where did the money originate?
Which entity ultimately supplied it?
What intermediaries touched it?
What claims exist against the same asset?
Who benefits when the company reaches public markets?
Those are entirely legitimate questions.
Then the IPO Changes the Financial Perimeter Again
This is where the repeated AI IPO pattern matters.
Private AI companies raise enormous sums.
Infrastructure companies obtain anchor AI customers.
Valuations rise.
Then public listings are discussed.
Reuters reports investors are currently waiting for Anthropic's S-1 because it will finally allow them to compare its widely discussed annualized revenue with actual GAAP revenue and examine its compute costs.
Its potential valuation has been discussed at around $2 trillion.
That transition matters.
Private capital carries the early-stage risk.
An IPO potentially transfers exposure into:
mutual funds,
retirement accounts,
institutional portfolios,
index funds,
and ordinary public investors.
The capital does not become “clean” in a criminal-law sense simply because a company goes public.
But the risk becomes financially distributed.
That distinction is important.
This Is AI Risk Distribution Through Liquidity
AI Risk Distribution Through Liquidity
AI Risk Distribution Through Liquidity occurs when private AI valuations and concentrated early-stage financial exposure migrate into broader public markets through IPOs, debt issuance or securitized financial structures, spreading the consequences of future valuation errors across a much larger investor base.
The stronger version of the question being asked is...
Not:
Is an IPO laundering money?
But:
What risks, assumptions and sources of capital become harder to see once privately financed AI expansion enters liquid public markets?
That is much more difficult to dismiss.
Public Markets Can Legitimize a Valuation Without Validating the Economics
Suppose a private AI company reaches a $500 billion valuation.
Then $1 trillion.
Then $2 trillion.
An IPO occurs.
Shares trade.
Index funds eventually buy them.
That market process can create enormous legitimacy.
But public liquidity does not answer:
whether projected demand was right,
whether compute costs remain sustainable,
whether margins ultimately justify valuation,
or how much upstream financing depended on circular AI relationships.
Liquidity validates:
market willingness to trade the asset.
Not necessarily:
the long-term economic thesis.
This Is Valuation Legitimacy Transfer
Valuation Legitimacy Transfer
Valuation Legitimacy Transfer occurs when an AI company's transition from opaque private markets into highly visible public markets causes investors to interpret market liquidity and institutional participation as evidence that the underlying valuation assumptions have been economically validated.
Those are different things.
A traded price is real.
The assumptions underneath it can still fail.
Now Add the Resource-Hoarding Problem
Capital is only one scarce input.
The same competition is occurring across:
talent,
minerals,
chips,
energy,
data centers,
supplier relationships,
and intellectual property.
That is why recent stories that initially look unrelated actually belong together from an AI-risk perspective.
Europe is struggling to build critical-mineral independence from China.
Major technology companies are reserving enormous data-center capacity across continents.
Suppliers are pivoting toward AI infrastructure.
Frontier labs are recruiting hundreds of employees from competitors.
Hyperscalers are issuing historic amounts of debt.
Banks are building AI engineering teams.
Banks are also building digital financial rails.
Each story individually looks reasonable.
From above, they reveal something different.
The AI Race Is Becoming a Resource-Control Race
The competitive objective is no longer simply:
build the best model.
It is:
control enough compute.
Secure enough capital.
Reserve enough power.
Lock in enough chips.
Find enough minerals.
Hire enough engineers.
Secure enough data-center capacity.
Acquire enough customers.
Reach the next funding milestone.
And do it before the competitor does.
That is much closer to strategic resource competition than ordinary software development.
This Creates AI Strategic Resource Enclosure
AI Strategic Resource Enclosure
AI Strategic Resource Enclosure occurs when leading AI companies, financial institutions and governments secure outsized shares of scarce inputs through long-term contracts, acquisitions, talent concentration, financing commitments or supply agreements, reducing the flexibility available to competitors and the broader economy.
It is not necessary to physically own a resource.
Control can come from:
priority contracts,
exclusive access,
anchor-customer relationships,
pre-purchases,
financing,
or strategic investment.
That is how ecosystems become enclosed.
And This Can Become Self-Reinforcing
The company with the most capital can acquire:
more compute.
More compute enables:
better models.
Better models attract:
customers.
Customers support:
higher valuation.
Higher valuation attracts:
more capital.
More capital acquires:
more compute,
more talent,
and more infrastructure.
That is:
AI Capital Reflexivity.
The leader's advantage produces resources that reinforce the advantage.
But Reflexivity Has a Weak Point
The loop depends on one thing:
continued belief that tomorrow's AI economics will justify today's resource consumption.
If that belief weakens:
capital becomes more expensive.
Debt terms tighten.
Infrastructure projects slow.
IPOs become harder.
Talent compensation becomes harder to sustain.
Suppliers renegotiate.
The circle can reverse.
Reuters is already seeing early signs in credit markets.
Lenders are charging more.
They are demanding stronger safeguards.
Some projects face delays because of:
chips,
electricity,
permitting,
and political opposition.
That does not mean the boom is ending.
It means financial discipline is starting to re-enter the system.
This Is the AI Resource Burn Problem
Imagine the frontier race continues for another decade.
Each generation requires:
greater infrastructure,
higher compensation,
more energy,
more advanced semiconductors,
more financing,
and more geographically distributed capacity.
At what point does the race itself become economically destabilizing?
Not because AI fails.
Because keeping AI competitive consumes too much of everything else.
AI Resource Burn Risk
AI Resource Burn Risk occurs when maintaining competitive frontier capability requires recurring increases in capital, compute, talent, energy and physical infrastructure faster than the economic returns generated by prior generations can replenish those resources.
That is one of the most important failure modes to watch.
Because a technology can be extraordinary—
and still have an unsustainable race structure.
Can AI Afford to Slow Down?
Researchers increasingly ask whether it should.
Financially, there is another question:
Can it?
Suppose OpenAI slows.
Anthropic doesn't.
Suppose Anthropic slows.
Google doesn't.
Suppose American firms slow.
China doesn't.
Every actor faces the same strategic fear:
If I stop spending first, I lose.
That creates a classic race dynamic.
Even actors who recognize the risk remain incentivized to accelerate.
This Is AI Capital Lock-In
AI Capital Lock-In
AI Capital Lock-In occurs when competitive pressure makes continued investment rational for each individual participant even when the aggregate capital requirement may be increasingly difficult for the broader system to sustain.
The participant cannot stop.
The market cannot easily stop.
The investor fears missing the next winner.
The government fears losing technological leadership.
The supplier fears losing the customer.
Everyone keeps moving.
And That Is Why New Financial Rails Matter
The question is not whether stablecoins are secretly funding AI.
We do not have evidence of that.
The interesting question is whether the financial system itself is evolving because increasingly global, digital and capital-intensive industries demand:
faster settlement,
more liquidity,
greater cross-border flexibility,
and new financing structures.
If AI becomes one of the largest capital-consuming sectors on Earth, those rails will eventually intersect with it.
So governance has to arrive before the problem.
The Financial-Control Questions Should Be Asked Now
If AI companies increasingly raise money through:
international banks,
private funds,
sovereign investors,
digital assets,
stablecoin settlement,
cross-border infrastructure vehicles,
and public markets,
investors and regulators need much stronger answers to:
Who ultimately provided the capital?
What other AI exposures does that lender hold?
How much leverage exists outside visible bank balance sheets?
How many entities are financing the same underlying infrastructure demand?
What AML controls apply when digital assets enter investment structures?
Can the beneficial owner always be reconstructed?
Can capital be followed after it crosses chains, jurisdictions and intermediaries?
How quickly can suspicious funding be frozen?
What happens when private AI companies become strategically important enough that governments cannot allow them to fail?
Those are not conspiracy questions.
Those are risk-intelligence questions.
The Biggest Risk May Be That Everyone Is Looking at One Resource at a Time
Finance specialists see:
debt.
Cybersecurity teams see:
models.
HR sees:
talent.
Governments see:
minerals.
Utilities see:
power.
Data-center companies see:
capacity.
Regulators see:
stablecoins.
AML investigators see:
money flows.
Investors see:
IPOs.
But the AI race connects all of them.
That is the bird's-eye view.
The System Looks Different From Above
From above, the chain looks like this:
AI capability increases
→ competitive pressure increases
→ compute requirement increases
→ infrastructure requirement increases
→ capital requirement increases
→ financial channels expand
→ talent concentration increases
→ supply contracts tighten
→ governments intervene
→ valuations rise
→ public-market exits become more important
→ more capital becomes available
→ next model generation begins.
That is not proof of wrongdoing.
It is proof of an increasingly resource-intensive system.
And every increasingly complex system develops blind spots.
The Strategic Questions
Investors, regulators, boards and governments should now ask:
How much additional capital does each frontier model generation require?
At what point does marginal capability improvement stop justifying marginal infrastructure cost?
How much AI financing has moved outside traditional bank balance sheets?
How concentrated are AI exposures across private-credit firms?
How much of supposedly diversified AI investment ultimately depends on the same frontier companies?
How much higher are borrowing costs becoming for non-AI businesses because of AI capital demand?
How much technical talent is becoming concentrated inside a handful of firms?
What happens when AI reconstructs strategically valuable knowledge from fragments brought by hundreds of employees?
How much chip capacity is contractually reserved years into the future?
How much power capacity has been effectively enclosed by data-center commitments?
How much critical-mineral capacity is already subject to strategic agreements?
How much capital entering AI can be traced to its ultimate beneficial owner?
How do AML controls work if future investment increasingly moves through tokenized financial rails?
What happens when regulated banks and less transparent private-credit markets finance different layers of the same AI ecosystem?
Do public IPOs distribute AI risk faster than they reveal its underlying economics?
Can public-market investors reconstruct where early-stage capital originated and how many claims already exist on future AI cash flows?
What happens if AI companies need another trillion dollars before existing infrastructure has generated adequate returns?
And the largest:
What does the AI race do when the next breakthrough requires more capital, talent, energy and infrastructure—but nobody in the race can afford to be the first one to stop spending?
Strategic Conclusion
The Goldman Sachs engineering office looks ordinary in isolation.
It isn't.
One of the world's most powerful financial institutions now employs more than 12,000 engineers.
It is expanding its AI workforce.
It helps finance the companies and infrastructure driving the AI boom.
It is participating in capital markets being increasingly reshaped by AI borrowing.
And it is joining 20 other financial institutions to build new digital money infrastructure.
Separately, FATF is warning that illicit financial networks are becoming increasingly sophisticated.
Separately, the BIS is warning that AI financing is increasingly:
opaque and interconnected.
Separately, AI-related debt has approached $500 billion in one year.
Separately, frontier companies are pursuing valuations potentially reaching trillions.
Separately, technology firms are competing for:
talent,
chips,
minerals,
electricity,
and data-center capacity.
No evidence proves these developments form one coordinated scheme.
But that does not make the pattern irrelevant.
Quite the opposite.
The absence of a conspiracy can make the systemic risk more important.
Because every participant can be behaving rationally:
Goldman wants better technology.
Banks want faster settlement.
AI firms want more compute.
Investors want higher returns.
Engineers want better opportunities.
Governments want AI leadership.
Suppliers want large customers.
And each rational decision can push the same system toward greater:
concentration,
opacity,
resource consumption,
financial interconnectedness,
and dependence on continued expansion.
That is AI Resource Capture Risk.
It creates AI Capital Absorption Risk when enormous AI financing begins competing with the rest of the economy.
It creates AI Capital Observability Risk when debt and private-credit exposure become difficult to map.
It creates AI Capital Provenance Risk when increasingly global funding architectures make the ultimate origin and path of capital harder to reconstruct.
It creates AI Talent Concentration Risk and AI Knowledge Aggregation Risk when scarce expertise accumulates inside a handful of institutions.
It creates AI Strategic Resource Enclosure when capital-rich actors secure disproportionate access to the physical resources required for the next generation.
And ultimately it creates AI Resource Burn Risk:
the possibility that maintaining leadership requires resources faster than prior AI investment generates the economic returns needed to replenish them.
That's where the uncomfortable question begins.
Maybe AI produces extraordinary productivity and eventually pays for everything being built around it.
That is possible.
But markets are already financing that future today.
With debt.
Private capital.
Infrastructure commitments.
Potential IPOs.
And increasingly complex financial channels.
So the question investors should be asking is not simply:
How big will AI become?
It is:
How much of the financial system will have to reorganize itself to keep AI growing long enough to become that big?
And then one question regulators should ask before the capital flows become too complex to reconstruct:
When trillions of dollars are racing toward the most strategically important technology on Earth, how certain are we that we can still follow the money?
I write about AI failure intelligence, ROI exposure, high-stakes decision architecture, and the hidden pathways through which AI incidents become financial and institutional consequences.
Follow me and subscribe to my work if you are responsible for investing in, acquiring, governing, insuring, or protecting strategically important AI systems and need to understand what technical failure can become after it leaves the engineering team.
Comments