Nvidia earnings blowout! Set to lead tech stocks’ rebound! B…

Weekly Outlook — Let’s Start With Some Other Developments

US retailer Target received a government refund of nearly $1 billion ($994 million pre-tax) after a court ruled parts of Trump’s tariffs unlawful. The windfall roughly doubled its second-quarter operating income. Other companies — including Estée Lauder, whose controlling family is connected to Warsh — benefited similarly. The timing, ahead of midterms, has not gone unnoticed.

The structure here is worth observing: tariffs are paid by US companies, price increases are absorbed by consumers, and the refunds flow back to corporations. A textbook case of the burden falling on ordinary people while the relief goes to the top.

Treasury Extends Buybacks to 10–30 Year Bonds

The US has moved to suppress long-end yields, with the 30-year briefly falling below 5.18%. In substance this is yield curve control — using market operations to paper over fundamental problems. The obvious questions are where the money comes from and how long it can last.

The buyback program has pushed both yields and the dollar lower, with the dollar hitting a three-month low. Gold continues to rise. When interest rates are artificially suppressed, the dollar’s loss of value has to show up somewhere — and it’s showing up in gold. Markets are effectively voting against the dollar’s credibility.

Bessent’s long-bond buyback is a short-term fix with long-term costs. Buying long duration at elevated prices while funding the operation short-term means raising its own financing costs. It isn’t sustainable. With the national debt now above $40 trillion, compounding interest will accelerate the problem — this is a well-worn concern, but that doesn’t make it less real.

At the same time, a surge in AI-related bond issuance from US tech giants is widening credit spreads and becoming one of the primary drivers pushing Treasury yields higher. AI companies competing with Treasuries for the same pool of capital creates a two-sided trap: the AI narrative supports debt repayment expectations, but the sheer volume of issuance raises borrowing costs for everyone.

Nvidia’s latest earnings delivered revenue and net income more than doubling year-on-year. The data centre business remains the absolute core — accounting for over 90% of total revenue — and next quarter’s guidance came in above this quarter’s actual result.

On China: the Financial Times reported that small quantities of Nvidia H200 chips have been approved for sale into mainland China, with ByteDance and Tencent each receiving around 10,000 units. Beijing has indicated it wants the chips to remain outside mainland China. Any near-term imports should be read as transitional, not structural.

Chinese AI majors are also investing at an aggressive pace. Alibaba is raising approximately $10 billion through a share placement at a 3.6% discount to fund AI infrastructure expansion, with quarterly capex already running materially higher. All of this signals that companies remain very optimistic about the durability of AI demand.

That said, there are things beyond the earnings headline worth paying attention to.

Morgan Stanley estimates Nvidia’s total credit exposure will reach approximately $200 billion by end-2028, of which around $170 billion consists of guarantees and contingent commitments that are currently off-balance-sheet but would be recognised in a downturn. A Wall Street Journal analysis found that nine major tech companies hold roughly $3 trillion in off-balance-sheet commitments — largely AI-related leases and procurement contracts. Reported capex figures significantly understate real liabilities. Meta and others have structured arrangements where they effectively finance and lease to themselves, keeping the obligations off the books. Meanwhile, private credit non-performing loans have risen to their highest level since 2017, with major funds already flagging problem loans. The industry faces its biggest challenge in nearly a decade. Given how deeply private credit is intertwined with AI data centre financing, this is a plausible ignition point for the next credit event.

Nvidia’s off-balance-sheet guarantees appear to be collateralised against GPU inventory — pledged to private lenders but not appearing on the balance sheet. The issue is that these are long-dated commitments, while GPU generations turn over quickly. Cards that are cutting-edge today could depreciate sharply within three years. If a forced liquidation of inventory were ever required to repay debt, the assets might not cover the liabilities.

The key moment to watch is when Nvidia’s chips stop selling — meaning when capex starts declining — meaning when hyperscalers conclude that data centre returns are falling short of expectations. That is when the stress becomes visible. But for the next one or two quarters, the numbers should remain strong, and Nvidia will likely continue to lead tech higher.

Even if an AI infrastructure bubble eventually bursts, the scale of the damage would almost certainly be smaller than 2008. That crisis was concentrated in real estate and banking — sectors that touched virtually every household. This cycle’s vulnerabilities are concentrated in a handful of large AI companies. There is no need for investors to panic.