There is a fascinating story buried underneath the headlines around Leopold Aschenbrenner’s Situational Awareness fund.
The fund reportedly went from roughly $45 billion at the beginning of July to around $10 billion after selling its public equity portfolio to Citadel following margin calls.
Yet one of its biggest AI bets stayed in place.
That was the reported $5 billion stake in Anthropic.
At first glance, that sounds contradictory. How can a portfolio lose tens of billions of dollars while one of its major investments remains untouched?
The answer comes down to something investors often underestimate during a bull market:
Liquidity matters just as much as conviction.
The same AI thesis can behave very differently depending on whether you own it through a publicly traded, leveraged position or through a private investment that is not marked to market every second.
And that distinction is becoming increasingly important as enormous amounts of capital continue flowing into AI infrastructure, private companies and data centers.
The leverage was on the part that could be sold
Situational Awareness, the AI-focused fund Aschenbrenner started after leaving OpenAI, reportedly built concentrated positions in AI infrastructure companies including:
- SK Hynix
- CoreWeave
- Nebius
- Micron
- Bloom Energy
These were not small, passive positions.
The fund reportedly used leverage, which magnified both the upside and the downside.
Then July happened.
Several of those stocks reportedly fell between 35% and 47% during the month.
When highly leveraged positions fall sharply, the problem is no longer simply that the investment thesis has become less attractive.
The lender gets involved.
A margin call can force an investor to sell positions regardless of whether they still believe in the long-term story.
That is what makes leverage so dangerous.
You may be right about the future and still be forced out before you get there.
And then there was Anthropic
The reported $5 billion Anthropic position had no margin attached to it.
That made all the difference.
Unlike the publicly traded positions, Anthropic is private. There is no stock ticker updating every second. There is no exchange immediately repricing the company every time sentiment changes.
There is also no daily public-market mechanism forcing a holder to respond to a falling share price.
So the fund’s public AI positions were forced to absorb the market’s repricing, while the private Anthropic position simply remained private.
One AI thesis. Two completely different financial structures.
That is the important lesson.
A public stock can fall 40% in a matter of weeks.
A private company does not necessarily receive a new valuation every time public markets get nervous.
That does not mean the private valuation is automatically more accurate.
It simply means the price is discovered differently.
Private does not mean immune to losses
This is where investors need to be careful.
It would be easy to look at the Anthropic stake and conclude that the private investment somehow “held up” better.
We do not actually know that.
A private investment can lose significant economic value without showing a new mark.
The difference is that the price is not continuously visible.
Public markets give you constant information, even when that information is painful.
Private markets give you fewer data points.
A public stock might be repriced every second.
A private company might receive a new valuation when it raises another round or when a secondary transaction takes place.
Both prices are real.
But they answer slightly different questions.
One reflects what the marginal buyer or seller is willing to pay right now.
The other reflects the last negotiated transaction.
That distinction becomes critical when portfolios contain both public and private assets.
The denominator effect is quietly changing portfolios
This is one of the most interesting pieces of the story.
Imagine an investor has:
- 85% in public assets
- 15% in private assets
The portfolio falls because public stocks decline sharply.
The private investments have not changed in their latest reported valuation.
Suddenly, the private assets represent a much larger percentage of the remaining portfolio.
Nothing had to be bought. Nothing had to be sold. The allocation changed simply because the denominator shrank.
This is known as the denominator effect.
And it can create a very real problem for investors.
A portfolio manager may have intended to maintain a 15% allocation to private assets.
But after a major public-market selloff, that same private portfolio could suddenly represent a much larger share of total assets.
That can lead investors to sell private positions in the secondary market to rebalance.
There is an interesting twist in the Situational Awareness story.
Instead of selling the illiquid private investment, the fund reportedly sold the liquid public positions to meet obligations.
The result?
The Anthropic stake became an even larger percentage of what remained.
The fund effectively became a private investment firm by subtraction.
Who is actually financing the AI infrastructure boom?
This is where the story gets even more interesting.
Anthropic is not only raising capital and building AI models.
It needs enormous amounts of computing infrastructure.
That means data centers.
That means electricity.
That means long-term leases.
And increasingly, it means complicated financing structures involving banks, developers and other technology companies.
A group of banks led by Morgan Stanley is reportedly in talks to lend $15 billion to Nexus Data Centers for a campus in Hubbard, Texas.
The reported financing includes roughly:
- $14 billion bridge financing
- A revolving facility
- A site with a natural gas plant
- Around 1.6 gigawatts of generating capacity
- Anthropic as the reported tenant
The interesting part is not simply the size of the facility.
It is who is taking on the credit risk.
Anthropic needs the capacity. Someone else may carry the financing risk.
Google has reportedly issued guarantees covering billions of dollars of Anthropic’s lease and power-purchase obligations across four leases.
In exchange, Google would reportedly receive around 20% equity in the data center and power project.
That creates an unusual structure.
The data center developer is the borrower.
Anthropic is the tenant.
Google already has an equity interest in Anthropic and would also receive equity in the infrastructure project.
And the financing sits outside Anthropic’s own cap table.
That matters.
Because the AI company can expand its infrastructure footprint without necessarily funding every dollar of that infrastructure directly from its own balance sheet.
The growth belongs to Anthropic. The financing structure is shared across several parties.
That is increasingly becoming a defining feature of the AI infrastructure race.
The next phase of AI may be a credit story, not just an equity story
For years, the AI investment conversation was largely about equity.
Who owns the model?
Who has the best chips?
Who will win the AI platform race?
Who will become the next trillion-dollar company?
But as AI infrastructure gets larger, another question becomes harder to ignore:
Who is financing all of it?
A data center can require billions of dollars before it generates meaningful revenue.
Power infrastructure is expensive.
GPU infrastructure is expensive.
Long-term leases require enormous commitments.
And frontier AI companies are making infrastructure commitments at a scale that requires more than traditional venture capital.
That brings banks, infrastructure investors, strategic companies and credit markets into the picture.
The reported Anthropic structure is one example.
Nvidia has also reportedly been involved in discussions around debt guarantees tied to OpenAI’s Ohio campus.
The structures are different, so the deals should not be treated as identical.
But the broader trend is worth watching.
Credit support can finance the next stage of AI capacity without requiring every dollar to come through a fresh equity round.
The government is becoming a shareholder too
There is another development that deserves attention.
The Commerce Department announced letters of intent with seven companies on July 29 under the CHIPS Act.
The proposed funding totals up to $874 million.
But there is a twist.
The government would receive minority, non-controlling equity stakes in the companies receiving the funding.
Two of the companies are publicly traded:
- GlobalFoundries, up to $300 million
- Aeluma, up to $30 million
Five are private:
- Kepler, up to $245 million
- Multibeam, up to $140 million
- Extropic, up to $75 million
- Thintronics, up to $50 million
- OBSIDIA Semiconductors, up to $34 million
There is an important caveat here.
These are letters of intent, not completed transactions.
The figures are maximum amounts.
Diligence is still underway.
Definitive agreements have not been signed.
And the specific terms of the government equity positions have not been published.
So there is still plenty that can change.
But the structure itself is noteworthy.
Federal support for private semiconductor companies could increasingly involve equity ownership rather than simply grants.
That puts a new kind of shareholder onto private-company cap tables.
And the eventual details will matter enormously.
What rights will the government receive?
Will there be information rights?
Transfer restrictions?
Consent provisions?
What happens if the company is eventually acquired?
Those questions cannot be answered yet because the definitive agreements have not been published.
The private market has another problem coming
The AI story is also changing the venture secondary market.
PitchBook estimates the US venture secondary market reached $121.7 billion over the trailing twelve months through Q2 2026.
That is an enormous market.
And much of it involves individual holders selling individual positions.
But there is a potential shift coming.
Three of the most heavily traded private names have been:
- SpaceX
- OpenAI
- Anthropic
SpaceX is now public.
OpenAI and Anthropic have reportedly explored potential listings, although neither has filed to go public and neither is guaranteed to pursue an IPO.
If these companies eventually leave the private market, the secondary market will lose some of its biggest sources of activity.
That raises a bigger question:
What replaces them?
The next private-market giants need more than a big valuation
It is tempting to simply look for the next company worth $10 billion, $20 billion or $50 billion.
But that is not enough.
A company needs several things to support a deep secondary market:
- Employees who have held shares long enough to become sellers
- Enough shares available to absorb large transactions
- An issuer willing to permit transfers
- Enough information for buyers and sellers to agree on a price
Very few private companies satisfy all four.
That is why replacing SpaceX, OpenAI and Anthropic in the secondary market could be harder than simply finding three companies with large valuations.
Liquidity is not the same thing as valuation.
And that distinction is becoming increasingly important.
SpaceX is about to provide a real-world test
SpaceX is now moving into a new phase.
Its first quarterly results as a public company are scheduled for August 4.
Then, on August 6, a major lock-up tranche is expected to lift, involving roughly 911.5 million shares according to the reporting cited in the source material.
That could give investors an important first look at what happens when a company with a huge pre-IPO shareholder base finally becomes more liquid.
Will insiders and early investors sell aggressively?
Will they hold?
Will demand absorb the additional supply?
The answer could tell us a lot about how large private-company shareholder bases behave after restrictions disappear.
Meanwhile, capital keeps moving into AI
The funding data tells its own story.
Xsight Labs: $300M+
Xsight Labs reportedly raised more than $300 million at a $2.8 billion post-money valuation.
The company develops Ethernet switching and DPU silicon, which sits between AI accelerators and the rest of the infrastructure.
It is a reminder that the AI infrastructure opportunity extends well beyond GPUs.
Moonshot AI: $3.5B
Moonshot AI reportedly raised $3.5 billion at a $35 billion valuation.
The round was reportedly oversubscribed after initially targeting between $1 billion and $2 billion.
The company’s Kimi K3 release helped drive the fundraising.
But there is an important consideration for investors outside China.
Most US accredited investors cannot directly access the company.
That means this valuation is being established within a different liquidity pool from the one available to most US private-market investors.
Commonwealth Fusion Systems: $1B
Commonwealth Fusion Systems raised $1 billion in equity financing.
The company did not disclose a lead investor or valuation.
Its backers reportedly include pension funds, sovereign wealth funds and infrastructure and industrial corporate partners.
That investor mix is notable.
The story is no longer purely about venture capital.
Large-scale infrastructure ambitions are increasingly attracting infrastructure-style capital.
Simile: $200M Series B
Simile raised $200 million at a $2 billion valuation, just five months after its $100 million Series A.
The company builds what it calls “agentic twins” of real people to test products and brands.
The valuation doubled in roughly twenty weeks.
That is an extremely fast repricing for a company operating at the applications layer.
DataBahn: $40M Series B
DataBahn raised $40 million in Series B funding, bringing total funding to $59 million.
Its business is less glamorous than the headline AI companies.
It filters and routes enterprise telemetry before it reaches destinations that charge based on volume.
But that is precisely why it is interesting.
AI creates enormous amounts of infrastructure spending, and businesses that help customers control those costs can benefit too.
Not every AI winner needs to build a model.
A few other moves worth watching
The broader market is also seeing meaningful consolidation and leadership changes.
Nscale agreed to acquire Anyscale for a reported $1.65 billion.
Anyscale is known for the Ray compute orchestration framework.
The reported price has not been confirmed by either company, and the transaction is expected to close in the second half of 2026 subject to regulatory approval.
Okta is acquiring Permiso Security for reportedly just under $200 million.
The transaction is expected to be mostly cash, although Okta has not disclosed the terms.
Permiso had raised around $29 million, making the reported acquisition price roughly seven times the capital it had raised.
Scale AI named Francis deSouza as CEO.
The former Google Cloud COO and Illumina CEO replaces interim CEO Jason Droege.
That change answers one of the major leadership questions around the company after Alexandr Wang left for Meta.
Zoox received the first US approval for paid driverless deployment.
The Amazon subsidiary was cleared by NHTSA to deploy up to 2,500 vehicles a year through 2028 without human controls.
That gives the rest of the autonomous-driving industry a significant precedent to watch.
The bigger picture: AI is becoming a balance-sheet game
Put all these developments together and a bigger pattern starts to emerge.
The AI race is no longer just about building better models.
It is about capital structures.
It is about who owns the infrastructure.
Who guarantees the leases.
Who lends the money.
Who takes the equity.
Who carries the credit risk.
Who is allowed to sell.
And who is forced to hold.
The Situational Awareness story is a perfect example.
The fund reportedly had enormous exposure to the same broad AI theme in both public and private markets.
Yet the public positions were vulnerable to leverage and margin calls.
The private Anthropic stake was not.
That does not prove the private investment was better.
It shows something more fundamental:
The structure of an investment can determine how long you get to stay invested.
The questions investors should be asking now
The most interesting part of this story may not be what happened this week.
It is what happens next.
Here are the questions worth watching:
1. Do other AI-focused funds have the same portfolio structure?
Situational Awareness combined leveraged public AI infrastructure positions with a large private Anthropic stake.
We do not yet know how common that structure is.
If more disclosures emerge, they could show whether this was an isolated case or part of a broader pattern.
2. How much of Anthropic’s infrastructure is being financed by third parties?
Reporting has described more than a dozen preliminary US lease agreements.
The Hubbard project is one of the deals where financing details have emerged.
The question is whether that structure is typical or unusual.
3. What will the government’s private-company stakes actually look like?
The Commerce Department’s letters of intent could become a template for future semiconductor funding.
But the definitive agreements will matter more than the headlines.
The rights attached to those shares could determine how meaningful the government’s ownership really is.
4. What happens to venture secondaries if the biggest names go public?
SpaceX has already moved out of the private market.
If OpenAI and Anthropic eventually follow, the secondary market will need new sources of liquidity.
Finding companies with large valuations is easy.
Finding companies with enough shares, sellers and transfer flexibility is much harder.
5. How much leverage is hiding underneath the AI boom?
This may be the most important question of all.
The AI infrastructure buildout requires extraordinary amounts of capital.
Some of that capital will come from equity.
Some will come from debt.
Some will come from guarantees.
Some will come through leases and project financing.
And when markets are rising, those structures can look almost invisible.
They become much more important when the tide starts going out.
The takeaway
There is a popular way of looking at the AI boom:
AI companies are raising billions because investors believe they will become enormous businesses.
That is true, but it is only part of the story.
The more interesting question is how the system around those companies is being financed.
A private Anthropic stake can remain untouched while leveraged public AI positions are liquidated.
A data center can be financed by banks while an AI company becomes the tenant.
A technology company can guarantee obligations and receive equity in the infrastructure behind its partner.
The US government can provide semiconductor funding while taking minority stakes in private companies.
And a private-market portfolio can become a much larger percentage of an investor’s assets simply because public stocks fell.
This is why the next phase of the AI trade may be less about who has the biggest valuation and more about who has the strongest balance sheet, the best financing structure and the ability to survive volatility.
The AI boom is attracting an extraordinary amount of capital.
But capital has different forms.
Equity can take dilution. Debt needs to be repaid. Guarantees can become liabilities. Leverage can force selling. And private valuations can stay unchanged on paper while the underlying market around them moves dramatically.
That is the part of the AI story worth watching now.
As Warren Buffett’s famous line puts it:
“It’s only when the tide goes out that you learn who has been swimming naked.”
The question for AI investors is simple:
When the tide eventually goes out, who will still have enough liquidity to keep swimming?