Wall Street’s AI Harvest: The Trillion-Dollar Underwriting Gamble

The Great AI Liquidity Event

On the 200th floor of the financial ecosystem, the air is getting thin. Goldman Sachs ($GS) just printed a staggering 78% jump in quarterly profits, a figure that would have seemed like a typo in any other era. Meanwhile, Morgan Stanley ($MS) is leaning into record-breaking revenues of $21.3 billion. But this isn’t just a cyclical rebound; it is the first major harvest of the AI capital super-cycle. The catalyst? A frantic, high-stakes race to take Anthropic public at a valuation that could touch $1 trillion.

As the lead underwriters—Goldman, Morgan Stanley, and JPMorgan Chase ($JPM)—sharpen their pencils for the Anthropic S-1, they aren’t just selling a company; they are selling a vision of a future where compute is the new oil. However, beneath the celebratory headlines of record equities trading, a more skeptical narrative is emerging. The banks are currently operating in a ‘Capex Trap,’ where the very infrastructure they are financing—billions in Nvidia-powered server farms—has yet to prove it can generate sustainable, long-term enterprise margins.

The Chinese Walls of West Street

The most fascinating boardroom drama isn’t the profit jump itself, but the unprecedented logistical nightmare of the Anthropic mandate. Goldman Sachs and Morgan Stanley find themselves in the ‘dual-mandate’ crosshairs, simultaneously prepping IPOs for Anthropic and its arch-rival, OpenAI. Sources indicate that internal ‘Chinese walls’ have been fortified to levels unseen since the dot-com boom. Deal teams are siloed, resources are restricted, and the risk of trade-secret leakage is a constant ghost in the machine.

This conflict is compounded by the sheer scale of the capital commitments. Anthropic is currently burning through a $1.25 billion-per-month computing agreement with SpaceX. For underwriters, the task of vetting these eye-watering expenses under Section 11 of the Securities Act is a minefield of disclosure liability.

The Compute Capex Flywheel

While the market fixates on the IPO pipeline, the real money is being made in the ‘Compute Flywheel.’ The AI boom is an infrastructure-heavy cycle, requiring tens of billions in debt and equity to fund data centers. Banks are capturing this on both sides of the balance sheet. They underwrite the debt to buy the chips, and the resulting equity issuance creates a permanently larger ecosystem for their market-making desks to monetize.

Morgan Stanley’s performance is a testament to this shift. Their net revenues have climbed steadily from $13.27 billion in Q3 2023 to a projected $17.89 billion by late 2025. This isn't just volatility; it is a structural transformation of the capital markets. The demand for prime brokerage financing—providing leverage to hedge funds rotating aggressively into AI—is proving to be highly ‘sticky’ revenue.

The Fallacy of GPU Collateral

However, an investigative look at the credit side reveals a silent bleed. To keep the AI engine running, some lenders have begun accepting the physical hardware—Nvidia GPUs—as collateral for loans. This is a dangerous gamble. In the LLM landscape, today’s state-of-the-art chip is tomorrow’s paperweight. If the ‘AI monetization gap’ fails to close, and software revenue doesn’t materialize, the secondary market will be flooded with obsolete silicon.

JPMorgan ($JPM) has been more cautious, maintaining a fortress-like balance sheet while its competitors chase the high-beta growth. JPM’s total assets have swelled to over $4.4 trillion, yet their provision for credit losses has also crept up, reaching $4.65 billion in Q4 2025. This suggests that Jamie Dimon’s shop is already bracing for a potential hangover in the leveraged loan market.

The Verdict on Valuation

The institutional sentiment has shifted. We are no longer valuing Goldman and Morgan Stanley on traditional price-to-book ratios alone. They are being re-rated as proxy beneficiaries of the tech boom. By capitalizing on capital-light, high-margin advisory fees to fund the AI capex surge, these banks are successfully defending—and expanding—their valuation multiples. Goldman’s 78% profit surge is a signal that the ‘toll collector’ model is working.

But the sustainability of this velocity is the real question. While the trajectory of AI infrastructure is secular, the current burst of mega-IPOs and trading volume feels like a concentrated windfall. As the Anthropic IPO approaches, the industry is watching to see if Wall Street can actually price a $1 trillion startup that loses billions, or if they are simply blowing a bubble that they themselves will eventually have to pop.

Check out our Interactive Charting Tool.