Nvidia Doubles Revenue to $96.2 Billion and Guides to $108 Billion — but Free Cash Flow Halves and the Margin Guide Sits at 74%
Data:
Main Theme: “Compute Is Revenue” — Nvidia beat on every headline line, reporting $96.22 billion of revenue up 106% and guiding the current quarter to $108 billion. The stock rose about 4% after hours toward $218 after closing down 1.6%. Core PCE held at 3.3% year-over-year, giving the Fed room to stay on hold, and the index drifted modestly lower ahead of the print.
Wednesday delivered the quarter’s most anticipated report and it cleared the bar comfortably. Nvidia posted revenue of $96.22 billion against a $92.17 billion LSEG consensus — up 106% year-over-year and 18% sequentially — with adjusted EPS of $2.22 versus $2.10 expected, and Data Center revenue of $89.02 billion up 117%. Third-quarter guidance came in at $108 billion ±2%, well above expectations, and explicitly excludes any China Data Center compute revenue.
CEO Jensen Huang framed the quarter in a single line on the call: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.”
But two lines below the headline complicate the picture, and they are precisely the two this publication has flagged for four weeks. Non-GAAP gross margin for the third quarter was guided to around 74%. And free cash flow fell to $21.3 billion from $48.6 billion in the prior quarter, as receivables and inventory ballooned with the ramp, while the company raised roughly $24.9 billion in new debt.
On the macro side, core PCE held at 3.3% year-over-year, unchanged from June, with the monthly core reading in line at 0.2%. Headline PCE rose 0.2% monthly and 3.7% annually — both above expectations. Northlight Asset Management’s Chris Zaccarelli captured the market read: while many of the PCE numbers were worse than expected, the most important measure — core PCE — “held constant and that will give the Fed more time to leave rates on hold.”
🟨 U.S. Equities | Drifting Lower Into the Print
| Index | Session | Detail | Stance |
| S&P 500 | −0.11% | Gyrated between gain and loss all day | Modest decline |
| Nasdaq Composite | Lower | −0.2% just after the open | Modest decline |
| Dow Jones Industrials | Lower | −53 points at the open | Modest decline |
| Nvidia (NVDA) | −1.28% to −1.6% | Then +4% after hours toward $218 | Reversed post-print |
Sector leadership was unusual and worth noting: industrials, technology and energy names rose, while health care and consumer names weighed on the market. That is close to the inverse of Tuesday, when technology and health care led and consumer staples and energy lagged.
Treasury yields were little changed. They initially declined on the monthly PCE readings but ticked up as investors registered the higher-than-expected annual growth of 3.7% headline and 3.3% core. Charles Schwab noted the important caveat: it is the monthly changes that may matter more to the Fed.
Ellen Zentner, chief economic strategist at Morgan Stanley Wealth Management, gave the cleanest summary: “Today’s mild upside inflation surprise and relative economic strength weren’t necessarily what investors — or the Fed — wanted to see. It wasn’t enough to shift the balance for September’s FOMC meeting.”
📰 Macro | Core PCE Holds at 3.3%
| Measure | July | vs Expected | vs June |
| Headline PCE (MoM) | +0.2% | Topped consensus | — |
| Headline PCE (YoY) | 3.7% | Above expectations | — |
| Core PCE (MoM) | +0.2% | In line | — |
| Core PCE (YoY) | 3.3% | Above expectations | Unchanged |
The split between monthly and annual readings is the whole story. The monthly core figure came in in line at 0.2%, which is the number the Fed weights most heavily for near-term policy. The annual readings — 3.7% headline and 3.3% core — both topped expectations and remain well above the 2% target.
Two things follow from that. First, the distortion flagged in advance by Fifth Third did not derail the print. Bill Adams had warned that PPI details “have upward implications for July core PCE,” citing portfolio management services up 6.5% on the month and 22.5% on the year — fees that rise mechanically with asset prices. Core PCE held constant regardless, which is the more reassuring outcome.
Second, and less comfortably, core PCE at 3.3% sits far above core CPI at 2.5%. That gap has run all year and is a substantial part of why the July FOMC produced a 9–3 vote with the first three-way same-direction dissent since September 2016. A benign CPI and a stubborn PCE give both camps on the committee a defensible position, which is exactly why Warsh’s address on Friday carries unusual weight.
🚀 Nvidia | The Numbers
| Metric | Fiscal Q2 2027 | Versus expectation / prior |
| Revenue | $96.22bn | vs $92.17bn (LSEG) — +106% YoY, +18% QoQ |
| Adjusted EPS | $2.22 | vs $2.10 expected; $1.87 last quarter |
| Data Center revenue | $89.02bn | vs $85.7bn expected — +117% YoY; $75.2bn last quarter |
| Q3 revenue guidance | $108bn ±2% | Well above consensus; excludes all China Data Center compute |
| Q3 gross margin guidance | ~74% non-GAAP | The point of contention |
| Free cash flow | $21.3bn | From $48.6bn the prior quarter |
| Supply commitments | $279bn | More than doubled from $119bn — “primarily related to the procurement of memory” |
| New debt raised | ~$24.9bn | — |
| Capital returned | ~$26.0bn | ~$99.0bn remaining on the buyback; 25-cent dividend |
| China | Hopper shipments <1% of Data Center revenue | Excluded from guidance entirely |
This extends a streak to five consecutive quarters of beating expectations on both revenue and adjusted EPS. The Blackwell Ultra architecture drove much of the acceleration, with hyperscale customers ramping deployments faster than modelled. Note that Nvidia has begun including stock-based compensation in its non-GAAP results, which makes direct comparisons with prior fiscal years less clean.
CFO Colette Kress supplied the forward number that matters most for the entire complex: capital expenditure among the “top five hyperscalers” is expected to increase to $1.3 trillion next year, from $800 billion in 2026. She also addressed the memory question directly: “Memory scarcity today is being driven in large part by the AI buildout itself.”
The stock closed down about 1.6% on the day and then climbed roughly 4% after hours toward $218, erasing the session’s decline. Nvidia carries a market capitalisation of roughly $5.1 trillion.
📌 Reading the Session
- Nvidia answered the demand question emphatically and left the cash question open. Revenue doubling, Data Center up 117%, a $108 billion guide and hyperscaler capex heading to $1.3 trillion is not a demand problem. But free cash flow halving to $21.3 billion while raising $24.9 billion of new debt is the same cash-conversion pattern that saw Applied Materials fall on an 80% free cash flow collapse and Fabrinet fall 11.3% on negative free cash flow.
- The $279 billion supply commitment is the most consequential single number in the release. More than doubling from $119 billion, primarily for memory procurement, it is simultaneously a lock on future capacity and a confirmation that the memory shortage constraining Apple, Cisco and the entire hardware complex is being caused by Nvidia’s own buildout.
- Core PCE holding at 3.3% is the benign macro outcome, but the CPI–PCE gap is now the central policy problem. Core CPI is at 2.5%; core PCE is at 3.3%. Both hawks and doves on a 9–3 committee can argue from the data, which makes Friday’s Jackson Hole address the only thing that can resolve the September question.
Thursday brings Marvell, Best Buy, Dollar General, Dollar Tree, Burlington and Autodesk, plus jobless claims and pending home sales. Then Warsh speaks at Jackson Hole on Friday.
Companies
Theme: “The Bar Was Margin, and Nvidia Half-Cleared It” — Revenue doubled, Data Center grew 117%, the Q3 guide came in at $108 billion — and gross margin was guided to roughly 74% while free cash flow halved. The stock rose 4% after hours anyway, which is itself the most informative fact about how the market is now weighing growth against cash conversion.
For four weeks this publication has argued that the determining variable in AI earnings reactions has been margin durability and free cash flow rather than revenue. Nine consecutive AI-adjacent companies beat estimates and fell between late July and 18 August on exactly those lines. Nvidia is the first large name to post the same cash-conversion deterioration and still be bought — and understanding why matters more than the headline beat.
🚀 1. Nvidia: What Actually Happened
The headline lines were unambiguous beats. Revenue of $96.22 billion against $92.17 billion expected, up 106% year-over-year and 18% sequentially. Adjusted EPS of $2.22 against $2.10. Data Center revenue of $89.02 billion, up 117%, against an $85.7 billion Visible Alpha consensus — and against $75.2 billion in the prior quarter.
The Q3 guide was the genuine surprise. Management guided to $108 billion ±2%, against a market that had been positioned for something closer to $100 billion, and that figure assumes zero Data Center compute revenue from China — where Hopper shipments accounted for less than 1% of Data Center revenue in the quarter. The China business is therefore pure optionality, not embedded risk.
The two lines that complicate it:
- Non-GAAP gross margin guided to around 74%. For a company whose margin trajectory has been a core part of the bull case, a guide at 74% was read by some outlets as a cut and cited as the reason shares initially wobbled.
- Free cash flow fell to $21.3 billion from $48.6 billion in the prior quarter, as receivables and inventory ballooned with the ramp. Nvidia raised roughly $24.9 billion in new debt and still returned about $26.0 billion to shareholders, with roughly $99.0 billion remaining on the buyback after May’s additional $80 billion authorisation, plus a 25-cent dividend.
The correct framing is that this is a cash-conversion and capital-allocation picture, not a profitability problem — net income and margins both rose. But it is the same working-capital dynamic that punished Applied Materials, whose free cash flow fell 80% to $210 million on record revenue, and Fabrinet, which grew revenue 45% to a record and fell 11.3%.
Why Nvidia was bought anyway comes down to the guide. A $108 billion quarter would represent roughly 12% sequential growth on top of 18%, with China excluded entirely. Applied Materials and Fabrinet had deteriorating cash conversion without a comparable acceleration to justify it. Nvidia has both.
💾 2. The $279 Billion Memory Commitment
This is the most strategically important disclosure in the release and it has received the least attention. Nvidia’s supply commitments more than doubled from $119 billion last quarter to $279 billion, “primarily related to the procurement of memory.”
Kress made the causal link explicit: “Memory scarcity today is being driven in large part by the AI buildout itself.”
That single sentence resolves a question that has run through this publication since 31 July. Apple cut guidance citing an inability to secure enough advanced chips and memory. Cisco guided gross margins below consensus and announced a 3.4–4.0% global price increase on core hardware. SanDisk’s Investor Day drove a 15% rally on a multi-year model of sustained memory capital spending, and Micron, SK Hynix and Kioxia have swung 5–10% in single sessions all month.
The chain is now documented from both ends: Nvidia is locking up $279 billion of memory supply, and the companies that buy the residual are cutting guidance. For portfolio purposes this makes the memory suppliers a structurally advantaged position and every downstream hardware buyer a margin risk — which is precisely the pattern the earnings reactions have expressed all month.
The $1.3 trillion figure is the other number to carry. Kress said top-five hyperscaler capital expenditure is expected to rise to $1.3 trillion next year from $800 billion in 2026 — a 63% increase. Combined with Microsoft’s $678 billion remaining performance obligations, Amazon’s $496 billion AWS backlog, CoreWeave’s $104 billion and Supermicro’s $60 billion order book, the contracted-demand evidence is now overwhelming.
💬 3. Huang Reframes the Argument
Jensen Huang used the call to shift the terms of the debate: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.”
The strategic intent is clear. The bear case has never been that Nvidia cannot sell chips — it is that AI companies are selling to other AI companies in a circular arrangement, with Nvidia investing in firms that then buy its hardware. Critics tie exactly that concern to the cash-conversion picture; Jim Cramer has defended the strategy as strengthening the broader ecosystem.
“Compute is revenue” is an attempt to move the conversation to enterprise monetisation — that customer compute is now billable output rather than speculative capacity. Kiplinger framed the same test independently: the next phase of AI has to demonstrate that trillions invested in GPUs, data centres and models translate into measurable enterprise productivity and revenue, because AI cannot survive on AI companies selling to other AI companies.
Huang separately said AI demand has spread far beyond one anchor lab — an important claim given that customer concentration has been the standing objection to the neocloud layer.
⚠️ 4. The Standing Objections
Four risks remain live regardless of the quarter:
- Circular financing. The 10 August partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilise more than $500 billion of third-party capital are signed memorandums of understanding, not closed deals, with management framing the timeline as unfolding over time. Treat it as a demand tailwind still forming.
- Competition, including OpenAI’s Jalapeño custom chip. Nvidia also announced a $6 billion deal for Poolside’s Model Factory, a wager on open-source models driving chip demand.
- China, excluded from guidance entirely — optionality, but also an acknowledgement that the business is not currently accessible.
- High-bandwidth memory risk, which the $279 billion commitment addresses through procurement but does not eliminate as a cost pressure.
And there is a newer, non-financial one: a growing nationwide backlash against data centre construction, which is now appearing in mainstream coverage as a constraint on the physical buildout.
📉 5. Other Movers
- Dick’s Sporting Goods — the detail behind Tuesday’s record decline emerged: the quarter was hurt by large promotional spending to move merchandise that did not sell. That is inventory misjudgement compounding weak demand, which is a harder problem than demand alone.
- GoDaddy was downgraded to underweight at Wells Fargo with a price target cut to $76, implying 24% downside. Analyst Alec Brondolo’s reasoning is worth noting beyond the single name: “AI Overviews now appear in roughly 48% of searches, up from 10% in December,” and “GoDaddy’s domain pricing remains ahead of peers, causing AI Overviews and LLMs to recommend competitors.” He argues “the top of funnel in consumer internet is quickly shifting to AI Overviews and LLMs.” This is the first concrete sell-side downgrade in this publication’s coverage driven by AI search displacing traditional discovery — a mechanism that reaches far beyond domain registration.
- CrowdStrike and Salesforce also reported after the close, alongside Synopsys, Agilent, HP, Okta and Williams-Sonoma.
📌 Analyst Take
Nvidia produced the cleanest possible answer to the demand question and no answer at all to the funding question.
On demand: revenue up 106%, Data Center up 117%, a $108 billion Q3 guide excluding China, and hyperscaler capex heading to $1.3 trillion. Combined with the four contracted-backlog disclosures already on record, the speculative-capex thesis is no longer defensible on the evidence.
On funding: free cash flow halved to $21.3 billion, $24.9 billion of new debt was raised, receivables and inventory ballooned, and $279 billion of supply was committed. That is an enormous forward cash obligation entered into by a company whose cash conversion is deteriorating — funded, for now, by a balance sheet that can absorb it and a buyback that continues regardless.
The market chose to weight the guide over the cash flow, and that is a defensible judgement — but it is a judgement, not a resolution. The same market punished Applied Materials for an 80% free cash flow decline three weeks ago. The difference is acceleration: Nvidia is guiding to 12% sequential growth on top of 18%; Applied Materials was not.
The most durable takeaway for positioning is the memory chain. Kress stated plainly that memory scarcity is being driven by the AI buildout itself, and Nvidia has locked up $279 billion of it. That makes memory suppliers structurally advantaged and every downstream hardware buyer a margin risk — the exact pattern that has driven Apple, Cisco, SanDisk, Micron and Kioxia all month. Marvell reports Thursday and is the next test of that thesis.
General
Wednesday, August 26th, 2026: The Demand Question Closes, the Cash Question Opens Wider
Nvidia’s fiscal second quarter was the most consequential corporate release of the year and it settled one debate decisively while intensifying another. Revenue doubled to $96.22 billion, Data Center grew 117% to $89.02 billion, and the company guided the current quarter to $108 billion with China excluded entirely. By any measure of demand, the AI buildout is accelerating rather than digesting.
And free cash flow halved, $24.9 billion of new debt was raised, and $279 billion of supply was committed — mostly for memory. The stock rose 4% after hours. Whether that reaction proves correct depends on a question the quarter did not answer: whether the cash generated by AI is arriving fast enough to fund the cash being committed to it.
- The Contracted-Demand Evidence Is Now Complete
Five independent points across the AI supply chain have now disclosed multi-year contracted demand, and Wednesday added the largest of them.
| Layer | Disclosure | Date |
| Hyperscaler (buyer) | Microsoft: $678bn RPO, +84% | 30 July |
| Hyperscaler (buyer) | Amazon AWS: $496bn backlog | 31 July |
| Neocloud (capacity) | CoreWeave: $104bn backlog + $25bn Q3 commitments | 11 August |
| Assembler (hardware) | Supermicro: $60bn fiscal-2026 orders | 11 August |
| Silicon (supplier) | Nvidia: $108bn Q3 guide ex-China; top-five hyperscaler capex to $1.3trn in 2027 from $800bn | 26 August |
The Kress capex figure is the one that reframes everything. Top-five hyperscaler capital expenditure rising to $1.3 trillion next year from $800 billion in 2026 is a 63% increase, disclosed by the company with the best visibility into those orders. That is not a forecast of an AI slowdown.
Huang’s claim that AI demand has spread far beyond one anchor lab addresses the standing customer-concentration objection that has dogged the neocloud layer since CoreWeave listed. Microsoft CFO Amy Hood made the same point on 30 July when she stressed that sequential backlog growth came from customers outside the frontier labs.
The bear case has therefore migrated entirely. It is no longer that demand is speculative. It is that the cash conversion of the businesses meeting that demand is deteriorating while their forward obligations compound.
- The Cash Conversion Problem, Quantified
Nvidia’s own numbers make the case better than any critic could.
| Line | This quarter | Prior quarter / context |
| Free cash flow | $21.3bn | $48.6bn — down 56% |
| Cause | Receivables and inventory ballooned | With the Blackwell Ultra ramp |
| New debt raised | ~$24.9bn | — |
| Capital returned | ~$26.0bn | ~$99.0bn left on the buyback |
| Supply commitments | $279bn | More than doubled from $119bn |
| Net income and margins | Both rose | Not a profitability problem |
This is the identical pattern that has driven the earnings reactions all month, at a vastly larger scale. Applied Materials beat and guided $700 million above consensus and fell more than 5% because free cash flow collapsed 80% to $210 million. Fabrinet grew revenue 45% to a record and fell 11.3% on weaker margins and negative free cash flow. CoreWeave ran $7.7 billion of capital expenditure against negative $4.71 billion of free cash flow. SpaceX spent $18.37 billion in a quarter.
The distinguishing feature is that Nvidia can afford it. A company generating rising net income with a $5.1 trillion market capitalisation, $99 billion of remaining buyback authorisation and access to debt markets at scale is not liquidity-constrained. Working capital absorption during a product ramp is normal and reverses as receivables collect.
The question is systemic rather than company-specific. If Nvidia, the most profitable participant in the chain, is raising $24.9 billion of debt to fund a ramp — and if hyperscaler capex is heading to $1.3 trillion — the aggregate financing requirement of the AI buildout is now large enough to matter to the bond market. That is precisely what strategists cited on 18 August when four sovereign long-bond markets hit multi-decade highs simultaneously, naming corporate issuance from AI companies as a factor raising term premium estimates.
The loop is now visible in both directions: AI issuance raises long yields, and higher long yields lower the present value of AI cash flows.
- The Memory Chain Is Now Fully Documented
Kress’s statement — “memory scarcity today is being driven in large part by the AI buildout itself” — closes a loop this publication has been tracing since 31 July.
| Date | Event | Position in the chain |
| 31 July | Apple cuts guidance, cannot secure enough advanced chips and memory | Downstream buyer, squeezed |
| 5 Aug | SanDisk: 84.6% gross margin, $2.01bn of sequential growth from pricing | Supplier, benefiting |
| 13 Aug | Cisco guides margin below consensus, raises prices 3.4–4.0% | Downstream buyer, squeezed |
| 13 Aug | SanDisk Investor Day +15% on multi-year model | Supplier re-rated |
| 26 Aug | Nvidia commits $279bn, mostly memory; “scarcity driven by the AI buildout itself” | The cause, stated |
For positioning this is now an actionable framework rather than a theme. Any company that buys memory at scale and cannot pass the cost through faces margin compression — Apple and Cisco have both demonstrated it. Any company that sells memory has pricing power for as long as Nvidia is committing $279 billion to procurement.
The caveat is that pricing power is cyclical and the market knows it. SanDisk’s 84.6% gross margin came substantially from price rather than volume, and the stock has swung from +15% to −9% to −6.45% within eleven sessions. The structural demand is real; the timing of the margin peak is not knowable.
- The Inflation Data Gives Both Fed Camps Ammunition
Core PCE held at 3.3% year-over-year with the monthly figure in line at 0.2%, while headline PCE rose 0.2% monthly and 3.7% annually — both annual readings above expectations.
Zaccarelli at Northlight identified the market-relevant point: core PCE “held constant and that will give the Fed more time to leave rates on hold.” Zentner at Morgan Stanley Wealth Management identified the policy-relevant one: “Today’s mild upside inflation surprise and relative economic strength weren’t necessarily what investors — or the Fed — wanted to see. It wasn’t enough to shift the balance for September’s FOMC meeting.”
The unresolved problem is the divergence between the two inflation measures:
- Core CPI: 2.5% year-over-year, the lowest since February.
- Core PCE: 3.3% year-over-year, unchanged and well above target.
That 80 basis point gap has persisted all year, and it is why the July FOMC produced a 9–3 vote with the first three-way same-direction dissent since September 2016. Hammack, Kashkari and Logan can point to core PCE at 3.3% and argue for a hike. The majority can point to core CPI at 2.5%, a monthly core PCE in line, and a consumer that just produced the weakest confidence expectations since January. Both readings are defensible.
This is why Friday matters disproportionately. With no forward guidance in the July statement and analysts treating every meeting as live, Warsh’s Jackson Hole address is the only scheduled opportunity to indicate which measure the committee is actually weighting — and the Associated Press has noted pressure rising on him to hike.
- A New Risk Channel: AI Search Displacing Discovery
Wells Fargo’s downgrade of GoDaddy to underweight, with a price target cut to $76 implying 24% downside, deserves attention well beyond the single name.
Analyst Alec Brondolo’s stated reasoning: “AI Overviews now appear in roughly 48% of searches, up from 10% in December.” He argues “GoDaddy’s domain pricing remains ahead of peers, causing AI Overviews and LLMs to recommend competitors,” and that pricing needs to align with the industry as discovery shifts to AI. His broader claim: “We believe the top of funnel in consumer internet is quickly shifting to AI Overviews and LLMs.”
A move from 10% to 48% of searches carrying AI Overviews in eight months is a structural change in how consumer internet demand is originated. The immediate implication for GoDaddy is narrow — an LLM recommending a cheaper competitor. The general implication is not: any business whose customer acquisition depends on ranking in traditional search results now faces a discovery layer that can be re-priced against it in real time.
This is the first concrete sell-side downgrade in this publication’s coverage driven by AI search displacement, and it is a category of risk that does not appear in any of the AI capex or demand analysis. It is worth screening portfolios for exposure to search-dependent customer acquisition — travel, insurance comparison, e-commerce aggregation, local services and domain or hosting businesses in particular.
📊 Global Macro Sentiment Summary — Wednesday, August 26th, 2026
| Narrative Channel | Core Fundamental Trigger | Net Portfolio Posture |
| Index Structure | S&P −0.11%; Nasdaq and Dow modestly lower; industrials, tech and energy up, health care and consumer down | 🟨 Drifting into the print |
| Nvidia — demand | Revenue $96.22bn (+106%) vs $92.17bn exp; Data Center $89.02bn (+117%); Q3 guide $108bn ex-China | 🟩 Emphatic beat |
| Nvidia — cash | FCF $21.3bn from $48.6bn; $24.9bn new debt; $279bn supply commitments (from $119bn) | ⚠️ Conversion deteriorating |
| Nvidia — margin | Q3 non-GAAP gross margin guided ~74% | 🟨 The point of contention |
| Nvidia — reaction | Closed −1.6%, then +4% after hours toward $218 | 🟩 Guide outweighed cash flow |
| Hyperscaler capex | Kress: top-five capex to $1.3trn in 2027, from $800bn in 2026 | 🟩 +63% |
| Memory | Kress: “Memory scarcity today is being driven in large part by the AI buildout itself” | 🟩 Suppliers advantaged, buyers squeezed |
| Inflation | Core PCE 3.3% YoY unchanged, monthly in line at 0.2%; headline 3.7% YoY, above expectations | 🟨 Fed can hold |
| Policy divergence | Core CPI 2.5% vs core PCE 3.3% — an 80bp gap arming both FOMC camps | ⚠️ Jackson Hole decides |
| Rates | Yields initially fell on monthly readings, ticked up on annual; little changed on the day | 🟨 Neutral |
| New risk channel | Wells Fargo: AI Overviews in ~48% of searches, up from 10% in December | ⚠️ Discovery displacement |
Compliance and framing notes. Nvidia has begun including stock-based compensation in its non-GAAP results, which makes comparisons with prior fiscal years less clean — note this when citing EPS growth. Sources differ on whether the ~74% Q3 gross margin guide constitutes a cut; present the figure rather than the characterisation. And note that the 10 August partnerships to mobilise $500 billion of third-party capital are signed memorandums of understanding, not closed deals.
Upcoming News
Thursday, August 27th, 2026 — Theme: “Marvell Tests the Memory Thesis, the Dollar Stores Test the Consumer” — Jobless claims and pending home sales lead a light macro calendar, while Marvell, Best Buy, Dollar General, Dollar Tree, Burlington and Autodesk report into the two open questions of the week.
Thursday is a corporate session with two distinct tests. Marvell is the first custom-silicon read after Nvidia disclosed $279 billion of supply commitments and named the AI buildout as the cause of memory scarcity. Dollar General and Dollar Tree are the lowest-income consumer read after Dick’s Sporting Goods cut full-year profit guidance in its worst session on record. Then Warsh speaks at Jackson Hole on Friday.
🔴 Calendar — Thursday, August 27th, 2026
Times in ICT (Hanoi). ET is ICT minus 11 hours.
| Time (ICT) | Currency | Event / Indicator | Consensus | Impact |
| 19:30 | USD | Initial Jobless Claims | — | 🔴 High |
| 19:30 | USD | Continuing Claims | — | 🟠 Med |
| 21:00 | USD | Pending Home Sales (July) | — | 🟠 Med |
| 21:30 | USD | EIA Weekly Natural Gas Inventories | — | 🟢 Low |
| Before open | — | Best Buy (BBY), Dollar General (DG), Dollar Tree (DLTR), Burlington (BURL), Royal Bank of Canada, Toronto-Dominion | — | 🔴 High |
| After close | — | Marvell Technology (MRVL), Autodesk (ADSK) | — | 🔴 High |
| All day | — | Jackson Hole Economic Symposium begins | — | 🔴 High |
- Marvell — The First Test of the $279 Billion Disclosure
Marvell reports after the close and is the most direct read on whether Nvidia’s memory commitment helps or hurts the rest of the silicon complex.
The setup: Nvidia disclosed $279 billion of supply commitments, more than doubled from $119 billion, “primarily related to the procurement of memory,” with CFO Kress stating that “memory scarcity today is being driven in large part by the AI buildout itself.”
For Marvell that cuts two ways. It supplies custom silicon and interconnect to hyperscalers building the same infrastructure — so top-five capex rising to $1.3 trillion next year from $800 billion is directly supportive. But it also competes for the same constrained memory and packaging capacity that Nvidia has just locked up at unprecedented scale.
Watch the gross margin line above all. This has been the determining variable in every AI-adjacent reaction this season: nine consecutive companies beat and fell between late July and 18 August on margins and cash flow; Analog Devices beat and rose on a 52% adjusted operating margin. Marvell fell roughly 8% on 24 August and around 10% during the week of 17 August, so positioning is lighter than a fortnight ago.
Autodesk reports the same evening as a read on design software and, indirectly, on construction and manufacturing activity.
- Dollar General and Dollar Tree — The Lowest-Income Consumer
These are now the most important consumer reads remaining, and the bar has been set low by everything that preceded them.
| Date | Company / data | Result |
| 7 Aug | The Trade Desk | −21% on an advertising revenue miss |
| 7 Aug | Under Armour | Full-year guidance cut |
| 11 Aug | On Holding | Worst day on record |
| 14 Aug | July retail sales | −0.6%, control group −0.4% |
| 20 Aug | Walmart | US comps +2.6% vs 3.8% exp; ticket +1.1% |
| 25 Aug | Dick’s Sporting Goods | Worst day on record; full-year profit outlook cut |
| 25 Aug | Consumer Confidence | 89.4, expectations −7.8% to a seven-month low |
| 27 Aug | Dollar General, Dollar Tree | The lowest-income cohort |
Why these two matter more than the rest. The Michigan survey found sentiment declines concentrated among older, lower-income and non-college households — precisely the dollar stores’ customer base. Real average hourly earnings have been negative for four consecutive months and the savings rate is at a four-year low.
The interpretive framework: dollar stores normally benefit when consumers trade down, in the same way Walmart does. Walmart already missed on 20 August. If the dollar stores also disappoint, the trade-down channel has been exhausted across the entire income distribution — which would mean the weakness is a spending-power problem rather than a mix-shift problem.
Best Buy is the discretionary electronics read, and it arrives with a specific complication: the memory cost inflation Nvidia just documented flows into consumer device prices. Watch for commentary on component costs and pricing.
A detail worth carrying from Dick’s: the quarter was hurt by large promotional spending to move merchandise that did not sell. That is inventory misjudgement layered on weak demand. If the dollar stores report the same combination, it signals the sector broadly misread the second half.
- Jackson Hole Begins
The symposium opens Thursday, with Chair Warsh speaking Friday. Its importance is structural rather than scheduled: the July FOMC statement contained no forward guidance, consistent with Warsh’s stated aversion to signalling the policy path, which has led analysts to treat every meeting as effectively live.
Wednesday’s data sharpened the question rather than resolving it. Core PCE held at 3.3% year-over-year while core CPI sits at 2.5% — an 80 basis point gap that gives both camps on a 9–3 committee a defensible position. Hammack, Kashkari and Logan dissented in July in favour of a hike, the first three-way same-direction dissent since September 2016.
What to listen for: whether Warsh signals which inflation measure the committee weights; whether he addresses the long end, given that the Treasury intervened in its own bond market this month and Secretary Bessent said yields “don’t reflect the underlying fundamentals” — a stance that works against a Fed content to let markets do part of its tightening; and any framework language, which first Jackson Hole addresses are conventionally used to establish.
- Carry-Over Into Thursday
- Nvidia beat on every headline line — revenue $96.22bn (+106%) versus $92.17bn expected, adjusted EPS $2.22 versus $2.10, Data Center $89.02bn (+117%) — and guided Q3 to $108bn ±2% excluding China. The stock closed down about 1.6% and rose roughly 4% after hours toward $218.
- Free cash flow fell to $21.3bn from $48.6bn, roughly $24.9bn of new debt was raised, and supply commitments more than doubled to $279bn, primarily for memory. Q3 non-GAAP gross margin was guided to around 74%.
- Kress: top-five hyperscaler capex rising to $1.3 trillion next year from $800 billion in 2026, and “memory scarcity today is being driven in large part by the AI buildout itself.”
- Core PCE held at 3.3% year-over-year with the monthly figure in line at 0.2%; headline PCE rose 0.2% monthly and 3.7% annually, both annual readings above expectations.
- CrowdStrike, Salesforce, Synopsys, Agilent, HP, Okta and Williams-Sonoma also reported after Wednesday’s close.
- The Road to September
| Date | Event | Why it matters |
| Fri 28 | Warsh speaks at Jackson Hole | Pressure rising on him to hike; the only scheduled chance to frame September |
| 4 Sept | August payrolls; annual nonfarm payroll revisions | May–June were already revised down a combined 103,000 |
| 15–16 Sept | FOMC decision and dot plot | ~70% probability of a hold priced |
The annual payroll revisions on 4 September deserve advance attention. The July report already cut May and June by a combined 103,000, taking the trailing twelve-month average to roughly 34,000 a month. An annual benchmark revision on top of that could materially change the labour picture the FOMC is working from — and it lands eleven days before the meeting.
Snapshot
Wednesday, August 26th, 2026 — Theme: “Revenue Doubled, Cash Flow Halved” — Nvidia reported $96.22 billion of revenue up 106%, Data Center up 117%, and guided the current quarter to $108 billion excluding China. Free cash flow fell to $21.3 billion from $48.6 billion and supply commitments doubled to $279 billion. The stock rose 4% after hours; core PCE held at 3.3%.
Wednesday closed the AI demand debate and opened the financing one wider. Nvidia beat every headline estimate and guided far above consensus, with CFO Kress disclosing that top-five hyperscaler capital expenditure is heading to $1.3 trillion next year from $800 billion in 2026. But free cash flow halved, $24.9 billion of new debt was raised, and the company committed $279 billion of forward supply — mostly memory. On the macro side, core PCE held at 3.3% year-over-year with the monthly reading in line, which Northlight’s Chris Zaccarelli said “will give the Fed more time to leave rates on hold.”
🏛️ The Bottom Line
The S&P 500 slipped 0.11%, gyrating between gain and loss all session, with the Nasdaq and Dow also modestly lower. Industrials, technology and energy names rose while health care and consumer names weighed on the market. Treasury yields were little changed — they initially declined on the monthly PCE readings, then ticked up as investors registered higher-than-expected annual growth.
July PCE: headline rose 0.2% month-on-month and 3.7% year-over-year, both above expectations. Core PCE rose 0.2% monthly — in line — and held at 3.3% year-over-year, unchanged from June. Ellen Zentner of Morgan Stanley Wealth Management: “Today’s mild upside inflation surprise and relative economic strength weren’t necessarily what investors — or the Fed — wanted to see. It wasn’t enough to shift the balance for September’s FOMC meeting.”
Nvidia reported fiscal Q2 2027 after the close. Revenue of $96.22 billion beat the $92.17 billion LSEG consensus, up 106% year-over-year and 18% sequentially. Adjusted EPS of $2.22 beat $2.10. Data Center revenue reached $89.02 billion, up 117%, against $85.7 billion expected and $75.2 billion in the prior quarter. The Blackwell Ultra architecture drove much of the acceleration, extending a streak to five consecutive quarters of beating expectations.
Third-quarter guidance came in at $108 billion ±2% with non-GAAP gross margin near 74%, and explicitly excludes any Data Center compute revenue from China, where Hopper shipments were less than 1% of Data Center revenue in the quarter.
The cash lines: free cash flow fell to $21.3 billion from $48.6 billion in the prior quarter as receivables and inventory ballooned with the ramp; the company raised roughly $24.9 billion in new debt; and it returned about $26.0 billion to shareholders with approximately $99.0 billion remaining on the buyback, plus a 25-cent dividend. Supply commitments more than doubled from $119 billion to $279 billion, “primarily related to the procurement of memory.”
CFO Colette Kress said capital expenditure among the “top five hyperscalers” is expected to increase to $1.3 trillion next year from $800 billion in 2026, and that “memory scarcity today is being driven in large part by the AI buildout itself.” CEO Jensen Huang told the call: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.”
Nvidia closed down about 1.6% on the day and rose roughly 4% after hours toward $218, against a market capitalisation of approximately $5.1 trillion. CrowdStrike, Salesforce, Synopsys, Agilent, HP, Okta and Williams-Sonoma also reported after the close.
Elsewhere, Wells Fargo downgraded GoDaddy to underweight with a price target cut to $76, implying 24% downside, with analyst Alec Brondolo noting that “AI Overviews now appear in roughly 48% of searches, up from 10% in December” and that the top of funnel in consumer internet is “quickly shifting to AI Overviews and LLMs.” Detail also emerged that Dick’s Sporting Goods’ quarter was hurt by large promotional spending to move merchandise that did not sell.
📉 Reference Levels for the Thursday Open (August 27th)
Derived from recent session closes and range extremes — not vendor-published levels. Verify against your own charts.
| Asset | Support | Resistance | Operational Bias |
| S&P 500 | 7,652 → 7,600 | 7,674 → 7,798.99 (record) | 🟨 Awaiting the Nvidia read-through |
| Nasdaq Composite | 25,980 → 25,900 | 26,180 → 26,803 | 🟩 After-hours strength |
| Dow Jones | 53,417 → 53,277 | 53,839 → 54,349 | 🟨 Range-bound |
| Nvidia (NVDA) | $208 (24 Aug close) | $218 (after hours) → $225 | 🟩 Guide outweighed cash flow |
| US 10Y Yield | 4.60% → 4.55% | 4.71% → 4.73% | 🟨 Little changed |
| US 30Y Yield | 5.10% | 5.25% → 5.34% | 🟨 Off multi-decade highs |
| Brent Crude | $88 → $70 (CBA downside) | $94 → $100 | 🔄 Falling through the week |
| Bitcoin | $70,000 | $80,000+ (three-month high) | 🟩 Risk appetite intact |
| VIX | 14.25 (2026 low) | 15.87 → 20 | ⚠️ Jackson Hole ahead |
📊 Market Sentiment & Bias
AI demand: 🟩 The debate is closed. Revenue +106%, Data Center +117%, a $108 billion Q3 guide excluding China, and hyperscaler capex heading to $1.3 trillion. Five independent points in the chain now disclose multi-year contracted demand.
AI financing: ⚠️ The open question, now larger. Free cash flow halved to $21.3 billion, $24.9 billion of new debt raised, $279 billion of supply committed. If the most profitable participant is issuing debt to fund a ramp, the aggregate financing requirement is large enough to matter to the bond market.
Memory: 🟩 The chain is fully documented. Kress named the AI buildout as the cause of scarcity. Suppliers have pricing power; downstream hardware buyers have a margin problem.
Inflation: 🟨 Enough to hold, not enough to settle. Core PCE at 3.3% unchanged with the monthly figure in line — but an 80 basis point gap to core CPI at 2.5% arms both camps on a 9–3 committee.
Consumer: 🟥 Unchanged from Tuesday. Dick’s cut full-year guidance in its worst session ever, with promotional spending on unsold merchandise compounding weak demand. Dollar General and Dollar Tree report Thursday.
New risk: ⚠️ AI search displacement. AI Overviews in roughly 48% of searches, up from 10% in December. A structural change in how consumer internet demand is originated.
💡 Top Trade Takeaway: “Own the Constraint, Not the Buyer”
Focus: Favour memory and constrained-input suppliers over downstream hardware buyers. Retain Nvidia exposure on the strength of the guide while monitoring cash conversion quarterly rather than annually. Screen portfolios for search-dependent customer acquisition. Await the dollar stores before drawing final conclusions on the consumer. Keep gross exposure moderate into Warsh on Friday.
Logic. Nvidia closed the demand debate in the most emphatic way available. Revenue of $96.22 billion up 106%, Data Center up 117% to $89.02 billion, a Q3 guide of $108 billion that assumes zero China compute revenue, and hyperscaler capital expenditure heading to $1.3 trillion next year from $800 billion in 2026. Set alongside Microsoft’s $678 billion in remaining performance obligations, Amazon’s $496 billion AWS backlog, CoreWeave’s $104 billion and Supermicro’s $60 billion, the speculative-capex thesis is no longer defensible on the evidence.
But the bear case has migrated rather than disappeared, and Nvidia’s own disclosures made it larger. Free cash flow halved to $21.3 billion from $48.6 billion, the company raised $24.9 billion of new debt, and it committed $279 billion of forward supply. That is the same cash-conversion deterioration that saw Applied Materials fall on an 80% free cash flow collapse and Fabrinet fall 11.3% on negative free cash flow. Nvidia was bought anyway because it is guiding to 12% sequential growth on top of 18% — the acceleration justifies the working capital absorption, and those two did not have that offset.
The systemic point is what matters for the coming quarter. If the single most profitable participant in the chain is raising $24.9 billion of debt to fund a ramp while hyperscaler capex heads to $1.3 trillion, the aggregate financing requirement of the AI buildout is now large enough to affect the bond market — which is precisely what strategists cited on 18 August when four sovereign long-bond markets hit multi-decade highs simultaneously, naming corporate issuance from AI companies as a factor raising term premium estimates. AI issuance raises long yields; higher long yields lower the present value of AI cash flows.
The most actionable single line was Kress’s: “memory scarcity today is being driven in large part by the AI buildout itself,” alongside $279 billion of committed supply. That closes a loop running from Apple’s 31 July guidance cut on component availability, through Cisco’s below-consensus margin guide and 3.4–4.0% price increase on 13 August, to SanDisk’s 15% Investor Day rally. The framework is now clean: own the constraint, not the buyer. Memory suppliers have pricing power for as long as this procurement continues; every downstream hardware manufacturer carries a margin risk it cannot fully pass through. Marvell tests this on Thursday.
Calendar discipline: Thursday — jobless claims and pending home sales; Marvell and Autodesk after the close; Best Buy, Dollar General, Dollar Tree and Burlington before the open, with the dollar stores the lowest-income consumer read after Dick’s cut guidance. Jackson Hole begins. Friday — Warsh speaks, with core PCE at 3.3% against core CPI at 2.5% arming both camps on a 9–3 committee. 4 September — August payrolls and annual nonfarm payroll revisions. 15–16 September — FOMC, with roughly 70% odds of a hold priced.
The report belongs to The Concept Trading and Van Hung Nguyen
