Briefings After Hours

Weekly Market Recap (August 24-28, 2026): Nvidia Delivered the Verdict – the Market Gave a Complicated Answer

Aug 30, 2026

By Ieva Bružaitė | Founder, Orion Capital | August 2026

Executive Summary

Nvidia earnings 2026 delivered the answer investors had been waiting for. Two weeks ago, this briefing named August 26 the day that would settle the AI debate. It has now happened, and the honest answer is: settled, but not simply. Nvidia reported the strongest quarter of its history — $96.2 billion in revenue, up 106% year over year, guidance raised above consensus to $106-110 billion — and the stock rose 8.7% on Thursday, its best post-earnings reaction in four quarters and the clearest single piece of evidence yet that AI demand is accelerating rather than plateauing.

And yet the broader semiconductor sector fell roughly 3% across the same five sessions. Marvell beat estimates, raised guidance to 50% growth, and was sold nearly 10%. Friday brought a new Fed chair’s first hawkish turn at Jackson Hole, a bond market pricing rate-hike odds up from 34% to 57% in a single speech, and the 30-year Treasury yield sitting within a whisker of 2007 levels while US gross federal debt crossed $40 trillion for the first time. Nvidia answered the summer’s central question. The market spent the rest of the week asking several new ones.

The Week in Numbers

Index	Friday Close	Week
S&P 500	7,711.76	+0.5% — first back-to-back gain after a choppy August
Nasdaq Composite	26,402.42	+0.9% — held gains despite Friday’s chip selloff
Dow Jones	53,559.99	+0.5% — first winning week in three
Nvidia (NVDA)	post-earnings +8.7% Thu, -4.5% Fri	+~1% net on the week — the verdict was split
VanEck Semiconductor ETF (SMH)	–	-3% on the week despite Nvidia’s blowout
US 10-Year Yield	4.72-4.73%	Up on Warsh’s hawkish tone
US 30-Year Yield	5.21-5.27%	Near 2007 highs; US gross debt crossed $40 trillion this week

Nvidia Earnings 2026: The Report That Delivered

Metric (Q2 FY2027)	Result	Context
Revenue	$96.2B, +106% YoY, +18% QoQ	Beat consensus by ~5%; fourth quarter of accelerating, not slowing, growth
EPS	$2.22 vs $2.09 expected	More than double year-ago EPS
Q3 guidance	$106-110B	Above Street’s ~$104B consensus – the number that mattered most
Gross margin guide	73.5-74.5%	Narrower than expected – explicitly due to rising memory prices
FY2028 revenue growth outlook	~70%	Sharply above prior analyst projections
Stock reaction	+8.7-9% Thursday	First positive post-earnings reaction in four quarters; added ~$440B market cap

Every question this briefing posed in its August 26 preview received a direct answer. Can Blackwell keep up with demand? CEO Jensen Huang’s own framing supplied the clearest possible response: demand is growing beyond 70% but is constrained by supply — not the other way around. Is the buildout broadening or narrowing to one company? Huang noted that a year ago a single AI lab was driving the buildout; today the number of companies requiring large GPU clusters has expanded dramatically, a claim reinforced within hours by Amazon and Nvidia jointly announcing that AWS will purchase two million Nvidia GPUs alongside the company’s new Vera CPU. What does the guide say about durability? A $106-110 billion third-quarter guide, comfortably above the roughly $104 billion the Street had modelled, was the single number that mattered most — and it beat.

One detail deserves particular attention for readers of this platform: Nvidia’s gross margin guidance narrowed to 73.5-74.5%, below what analysts had modelled, and management was explicit about why — rising memory prices. This is the memory war thesis we traced through SK Hynix in July arriving, in real time, on the income statement of the world’s most valuable company. The HBM shortage is no longer a story about Korean memory makers; it is now a cost line inside Nvidia’s own results, and it will very likely appear again in every AI hardware company’s margins this autumn.

The Rest of Semiconductors: Priced for Perfection, Again

If Nvidia was the week’s proof of demand, Marvell was the week’s reminder about expectations. Reporting Wednesday, Marvell posted record revenue of $2.74 billion, up 37% year over year, beat on both revenue and earnings, and raised its fiscal 2028 outlook to roughly $18 billion — 50% growth. The stock fell as much as 10% anyway. The mechanism was almost identical to SK Hynix in July: shares had risen 184% year-to-date and roughly 30% in the month before the print, at a forward multiple near 58 times earnings, leaving essentially no room for a merely excellent quarter. A 90-basis-point sequential compression in gross margin — the same rising-input-cost story as Nvidia’s, but received without Nvidia’s offsetting growth surprise — was enough to trigger the sell-off.

The pattern repeated across the group. The VanEck Semiconductor ETF fell roughly 3% on the week even as its largest holding posted the best earnings reaction of its year; Applied Materials and Broadcom both traded lower in sympathy with Marvell’s post-earnings move. The lesson, restated for a third time this summer: strong results and rising share prices have become two separate questions. The first is now being answered clearly and consistently in the affirmative. The second depends entirely on what was already in the price before the report.

Jackson Hole: A New Chairman’s First Hawkish Signal

Friday brought the event this briefing flagged three weeks ago as the summer’s second great catalyst: Kevin Warsh’s first keynote address as Fed Chair, delivered at the Kansas City Fed’s Jackson Hole symposium and titled, pointedly, “In Our Time.” Investors had been pressing Warsh for months to articulate his “reaction function” — the conditions under which he would move rates in either direction — and Friday was billed as his clearest opportunity to do so.

He offered philosophy rather than a formula, and the philosophy leaned hawkish. “While this summer’s [inflation] readings were better than expected, they do not tell me that underlying trends have meaningfully improved,” Warsh said, adding that the Fed should not “indulge a regime in which market participants are looking primarily to the Fed for their next trade.” He committed, in his words, “to a discipline, not to a decision.” Markets read the discipline as tilted toward higher rates: the probability of a September hike priced into futures markets jumped from roughly 34% to 57% within hours of the speech, short-term Treasury yields rose, and all three major indexes turned lower into Friday’s close, with semiconductor stocks — Nvidia and Intel among them — leading the Nasdaq’s decline.

The bond market supplied the week’s quietest but most consequential data point. The 30-year Treasury yield closed near 5.21-5.27%, within reach of levels last seen in 2007, while US gross federal debt crossed $40 trillion for the first time in history. Treasury Secretary Scott Bessent responded by announcing that the government will double its long-bond buyback program to $4 billion per session from $2 billion, targeting the 10-, 20- and 30-year segments specifically — an unusually direct attempt to cap the borrowing costs that a near-two-decade-high long yield imposes on every asset priced against it, Nvidia included.

Earnings Season Closes at Full Strength

Largely overshadowed by Wednesday and Friday’s drama, the second-quarter earnings season quietly finished among the strongest on record: more than 85% of S&P 500 companies beat analyst estimates, with year-over-year earnings growth topping 50% — or roughly 30% once one-time hyperscaler investment gains are excluded, still comfortably the best pace in years. Meanwhile, the Commerce Department’s second estimate confirmed second-quarter GDP growth at a 1.5% annualised rate, decelerating from 2.1% in the first quarter — a reminder that the earnings boom and the broader economy are, once again, telling two different stories at two different speeds.

Orion Capital Perspective: A Verdict With Two Clauses

We wrote two weeks ago that Nvidia’s evening would settle the demand side of the summer’s argument while leaving the slope — how fast, at what margin, at what price — as the genuinely open question. That is almost exactly what happened, in the clearest possible form. Demand: settled, decisively, in the affirmative. A 106% revenue quarter, a raised guide above consensus, and a hyperscaler doubling down within hours are not ambiguous signals. Price and margin: still live, and now visibly harder. Memory costs are compressing gross margins across the AI hardware stack, a new Fed chair has tilted hawkish in his very first public test, and the long bond — the discount rate applied to every dollar of future AI earnings — sits near its highest level in almost two decades.

This is not a contradiction; it is what a maturing capital cycle looks like. The receipts-versus-promises framework that guided this platform’s coverage all summer holds up once more, but its lesson has sharpened: producing a receipt is no longer sufficient on its own, because the ruler itself — the valuation, the multiple, the rate at which future cash flows are discounted — is now moving too. Nvidia cleared a very high bar by a wide enough margin to be rewarded outright. Marvell cleared its bar and was sold regardless. The difference was not the quality of the business. It was the distance between the result and what had already been priced.

What We Watch Next

September 15-16 FOMC meeting – now the live question after Friday’s repricing of hike odds; watch whether incoming inflation data pushes the probability further in either direction
Memory prices and margins across the AI hardware stack – Nvidia’s own guidance confirmed the mechanism this platform traced through SK Hynix in July; expect the same story from AMD, Broadcom and the hyperscalers’ cloud segments this autumn
The 30-year Treasury yield and the Treasury’s buyback response – whether Bessent’s doubled buyback program meaningfully caps long-end yields, or merely slows their rise
Marvell’s October 6 investor day – management’s chance to name firm FY2028 custom-silicon programs and reframe the sell-off as a valuation reset rather than a demand concern
Broadening AI capex evidence – further hyperscaler commentary on GPU purchase commitments following the AWS-Nvidia two-million-unit agreement

Conclusion

The week investors had circled all summer arrived and did exactly what it promised: it produced a verdict. What it did not produce was simplicity. Nvidia proved, as thoroughly as a single earnings report can prove anything, that the AI buildout remains real and still accelerating. The same week proved, through Marvell, through a hawkish Fed debut, and through a bond market flirting with 2007-era yields, that proving demand is no longer the hard part of this cycle. Pricing it correctly is. That is the argument markets will spend the autumn having — and, fittingly, it is the same argument this platform has been having since July: the market pays for receipts, but increasingly, it is auditing the ruler it uses to grade them.

Disclaimer: This publication is independent research and reflects the views of the author. It is provided for informational purposes only and does not constitute investment advice or a recommendation to buy or sell any security. Investors should conduct their own analysis before making investment decisions.

References

About the Author

Ieva Bružaitė is the Founder of Orion Capital, an independent research platform covering AI, semiconductors and long-term investing.

Ieva Bružaitė - Founder of Orion Capital

← Back to Orion Capital