Nvidia’s $500 Billion Financing Plan Draws a $180 Sell

Seaport Research Partners kept a Sell rating and a $180 target on Nvidia on Thursday, about 15% below the $212 close. Shares ticked up Friday, rising 1.3% in premarket trading after a 4.4% drop the day before.

Jay Goldberg, the Seaport analyst, says the chipmaker’s aggressive finance bundling is a reaction to competitive pressure. He still thinks Nvidia can post record sales into early 2027. He is pricing the stock as if the same credit that fills those orders will hurt when the cycle turns.

Seaport Holds a $180 Nvidia Sell After Thursday’s Drop

Goldberg wrote that he could be wrong in the coming months if Nvidia uses its supply-chain weight, “but competitive pressures are mounting.” Sell calls on the name are scarce, which is why a $180 target against a $212 close still moves the tape even when Friday buyers show up.

FRIDAY’S TAPE AGAINST THE $180 TARGET

• Thursday close: Nvidia finished at $212, down 4.4%, as AI and chip stocks sold off on fresh doubts about how long spending can last.
• Friday premarket: The stock rose 1.3% before the open, then ticked higher in the regular session as the Seaport note circulated.
• The gap: $180 sits about 15% below Thursday’s close, a discount Goldberg holds even while he grants the company a strong run into early 2027.
• Cash-flow offset: He says Nvidia can afford the new promises, with operating cash flow that could hit $200 billion this year.

That last point is the hinge. Goldberg is not describing a company that cannot pay its bills. He is describing a company that is using those bills to keep other companies buying its boards.

A $500 Billion Memo With Apollo and BlackRock

On August 10, 2026, Nvidia said it had signed memos of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. The six firms would set up independent compute financing platforms designed to mobilize over $500 billion of third-party capital for AI sites over time. The cash is not Nvidia revenue, and the partnerships still need final contracts.

Jensen Huang, Nvidia’s founder and chief executive, put the shift in one line: chips are no longer just products. They are supposed to be factories that a pension fund can underwrite.

NVIDIA has reached an important milestone. We began by building chips; today, we are helping create a new class of productive, investable infrastructure: AI factories. In AI, compute is revenue.

Jensen Huang, founder and CEO, Nvidia newsroom statement, August 10, 2026

David Solomon, chairman and chief executive of Goldman Sachs, said the bank wants “a market for credit backed by NVIDIA compute.” Huang has said Nvidia may offer residual-value support for up to 25% of an opportunity, judged deal by deal. Goldberg thinks the company would like the whole platform even larger, up to $1 trillion, with Nvidia as a cornerstone rather than the sole signer.

https://t.co/pppEbPVtfE

— Jensen Huang (@JensenHuang) August 10, 2026

The structure is meant to solve a problem Huang has already named out loud. Neoclouds, the independent AI clouds that borrow to buy Nvidia systems and rent the compute, have demand he has called “off the charts.” He has also said their constraint is capital, not customers. A platform that lets insurers and private credit funds lend against those racks is how Nvidia answers that shortage without cutting the list price.

What Nvidia’s July 10-Q Committed

Goldberg’s note said Nvidia “over-taxing” its balance sheet after it added roughly $250 billion in commitments last quarter, some to lock down the supply chain and some to support data center construction. The July quarter filing is more precise, and it splits those promises into different buckets.

The 10-Q for the period ended July 26, 2026, says supply commitments rose to $279 billion from $119 billion a quarter earlier, a $160 billion jump in inventory and factory lock-ins for memory and manufacturing. Separate from that stock of supply deals, Nvidia added customer-support contracts and guarantees that total $164.5 billion at the cap, including a $105 billion Ohio package signed in August.

NVIDIA COMMITMENTS IN THE JULY 10-Q

Long-term debt also jumped, to $32.4 billion from $7.5 billion, after Nvidia issued $25.0 billion of senior notes in June 2026. Some of those supply deals can be changed or canceled before firm orders land, the filing says, and extra costs can follow. The Ohio guarantee is a different animal: it is a contingent promise that grows as buildings come online.

The $105 Billion Ohio Guarantee Starts in Fiscal 2029

In August 2026 Nvidia capped credit support at $105 billion for land, power and shell at SB Energy’s PORTS Technology Campus in Pike County, Ohio. The site is slated for about 4.25 gigawatts of IT load under 20-year leases to an OpenAI affiliate, in nine phases. Nvidia pays only if the tenant defaults, and only on defined slices of rent and power, not the whole campus.

Each guarantee generally starts when that phase’s lease starts. The first phase is expected in fiscal 2029. The amount is supposed to shrink across each 20-year term, and the promises end if OpenAI earns a satisfactory credit rating or when a lease runs out. In exchange, the campus is to host Nvidia gear only, with limited exceptions. Nvidia also holds an option, entirely in its hands, to add credit support for about 3.8 more gigawatts as the site grows.

That is residual-value and rent support dressed as an exclusive offtake. It is also, as Goldberg notes, off the $279 billion supply line and off the $500 billion memo. The stack sits in layers, which is why a single “$250 billion added” line in a research note can sound larger or smaller than the filing depending on which layer you count.

Backstops Are a Discount by Another Name

Goldberg’s note rests on four points, and only one of them is “AMD exists.” The rest are about how Nvidia is paying, in credit rather than in price, to keep a wide set of buyers in the market.

GOLDBERG’S FOUR PRESSURE POINTS

• Physical build: About 55 gigawatts of data center capacity is announced for the next year, but he says most observers doubt there are enough electricians, gas turbines and grid ties to finish more than about 30 gigawatts.
• Balance-sheet load: He counts about $250 billion of new commitments to lock supply and help build sites, plus the $500 billion funding platform he thinks could stretch toward $1 trillion, and he calls those liabilities pro-cyclical.
• Circular credit: Nvidia invests in and lends to suppliers and customers. He says calling neocloud and data center spending an investment is fair, and that it still functions like a discount because backstops and guarantees cut the buyer’s cost of capital without a list-price cut.
• The neocloud bind: Nvidia wants many small, dependent clouds rather than three giant ones, Goldberg says, and it is also squeezing those same clouds’ finances. If CoreWeave, Nebius and their peers fail, he expects both lower sales and thinner margins.

The CoreWeave file is the cleanest specimen of the third and fourth points working at once. A September 2025 CoreWeave filing showed Nvidia agreeing to buy unsold capacity through April 2032, an order first valued at $6.3 billion. Nvidia takes unused racks off the cloud’s hands, which helps the cloud borrow, which helps the cloud buy more Nvidia systems. Goldberg’s line is that this is still a price cut, just one that never shows up on a quote sheet.

The risk here is that this kind of liability is pro-cyclical. It will boost things when times are good, but when the cycle turns down, the debt will amplify the pain. At some point, probably still years away, when compute demand gets saturated, the neoclouds will put their backstops to Nvidia at just the moment when the company’s cash flows get pinched by declines in sales.

Jay Goldberg, analyst, Seaport Research Partners note to clients

Huang’s public answer is that the neoclouds’ problem is a shortage of capital, not of work. That is the same fact. It just leads him to build a bigger credit pipe instead of a smaller one.

After Custom Silicon, the Old Customer Mix Breaks

The reason Nvidia is willing to sit behind its own buyers is that the old mix of customers is trying to sit behind itself. Amazon and Google this past week each put out open-source tools, TorchNeuron and TorchTPU, that let PyTorch jobs move onto Trainium chips and TPUs with a configuration change instead of a rewrite. The software moat that made switching painful is the thing those tools are built to thin.

DIGITIMES, in a mid-October note on 2027 high-end cloud AI accelerator shipments, has ASICs overtaking GPUs on unit volume for the first time. It puts Nvidia still first at about 35% of those units, with Google near 31% after a sharp TPU ramp. Merchant GPUs keep growing. They just grow slower than the in-house silicon at the same accounts that already buy a large share of Nvidia’s data center boards.

Goldberg reads the finance bundling as a response to that pressure. If hyperscalers can train and run more work on their own chips, Nvidia needs a second bench of buyers who cannot print their own accelerators. Those buyers need cheap credit. Cheap credit, at this scale, is something Nvidia and six Wall Street firms can organize. The irony is that the customers most in need of that credit are the ones whose default would send the backstops home.

55 Gigawatts of Plans, 30 Gigawatts of Power

The near-term governor on all of this may not be a rival chip. Goldberg’s first point is that the announced build is larger than the trades that can actually pour the slabs. He puts announced capacity for the next year at about 55 gigawatts and the plausible finish, given electricians, turbines and utility hookups, near 30 gigawatts.

If he is right, a slice of today’s supply lock-ins is a bet on sites that will slip. Nvidia’s own 10-Q already warns that customers delay new architectures when land, power, shell or capital is missing, and that the company has started guaranteeing those inputs for select partners for that reason. The Ohio package is the largest of those bets, and it does not even begin to accrete until fiscal 2029.

A slipped site does not automatically cancel a GPU order. It can, though, leave a neocloud with debt, empty raised floor, and a capacity contract Nvidia has promised to eat. That is the mechanical path from a turbine shortage to a line on Nvidia’s P&L, and it is why Goldberg treats the grid as a chip problem.

Neoclouds Send the Bill Back as Sales Slow

The $500 billion memo is still a memo. The $105 billion Ohio cap is years from its first phase. The $279 billion of supply deals is the piece that is already live, and even that book can move if Nvidia reschedules. Goldberg grants all of that, grants the cash flow, and still keeps $180.

His wager is about sequence, not solvency. Nvidia is building a customer base that depends on Nvidia’s own credit, because the customers who do not need that credit are busy designing around CUDA. While token demand is tight, the backstops barely show. When demand fills, the same names that borrowed against residual value will, in his telling, present those claims in the quarter when GPU sales are already falling.

Friday’s small bounce after Thursday’s 4.4% drop says the market is not on that clock yet. The $180 target says Seaport is.

Disclaimer: This article is news reporting and analysis for information only. It is not investment advice, a recommendation to buy or sell Nvidia or any other security, or a forecast of returns. Readers should consult a licensed financial adviser or other qualified professional who can consider their own holdings, time horizon and risk tolerance before making any investment decision. Ratings, price targets, commitment figures and contract terms are those published in the cited notes and filings and may be revised, restated or closed on later dates.