Rome’s aqueducts date back to the late fourth century BC, when Appius Claudius Caecus built the Aqua Appia to bring water to a growing city. Over the following centuries the empire raised dozens more - the Aqua Claudia, the Anio Novus and others - at enormous capital cost, borne by the state treasury. The treasury in turn drew on taxes, the provinces and conquests that were already yielding revenue. The aqueducts did not pay for themselves within a generation. They needed constant maintenance and repair, and they kept running for centuries for as long as the state could afford to look after them. When, in the late empire, revenue no longer covered spending, maintenance stopped, channels silted up and, together with the damage done by war, some of the structures collapsed, however well they had been engineered.
The same logic applies today, as Amazon and Intel pour money into infrastructure for artificial intelligence. What separates them is whether a working business already sits underneath, covering the costs until the investment starts to pay off. Amazon can still afford a spell of negative free cash flow, because its foundations are sound. Intel cannot afford the same delay, because its foundations are already shaking. Every test that follows - how long returns have to cover the cost of capital, the quality of the debt, the response to the first warning signs and the conditions for success - exists to judge how long that period can last before systemic pressure on prices sets in.
Amazon reported revenue of $201.9bn, up 12% year on year (13% excluding currency effects) and slightly short of the consensus of about $202.3bn. By segment, North America brought in $112.9bn (+6%), International $43.1bn (+5%, or +8% excluding currency) and AWS $46.0bn (+39%). That is AWS’s fastest growth in almost five years, and the unit now has an annualised revenue run rate of $184bn. Moving Prime Day from the third quarter to the second in 2026 shaved roughly 400 basis points off reported revenue growth, because last year the event fell in the third quarter and this year it did not. Without that shift, overall growth would have been close to 16%.
Segment operating costs stayed under control and margins widened. North America had costs of about $104.2bn and operating income of $8.7bn (a margin of 7.7%, against 4.5% a year earlier). International had costs of about $41.5bn and operating income of $1.6bn (a margin of 3.7%). AWS had costs of $28.2bn and operating income of $17.8bn (a margin of 38.7%). Total operating income reached $28.1bn, a margin of 13.9% and the highest in the company’s history. Net income was $20.7bn ($1.89 per share), slightly below the consensus of about $1.96. The decline from $1.95 a year earlier is mainly due to a one-off pre-tax gain of $9.5bn on the Anthropic investment in the third quarter of 2025.
Advertising revenue, however, came in at $19.64bn (+11%, or +12% excluding currency) against a consensus of about $20.8bn. Management attributes the shortfall to the Prime Day shift (about 850 basis points off advertising growth), the absence of NBA games in the quarter and a tougher comparison with last year, and expects growth to return to the high teens as soon as the fourth quarter. This is a miss in a high-margin business that the market watches as a gauge of the health of the retail platform, and the next quarter will show whether the explanation holds.
What sets this report apart is not record profitability but free cash flow, which has turned negative. Over the past twelve months free cash flow was minus $42.4bn (against a positive $14.8bn a year earlier). In the third quarter alone the outflow was about $33.5bn. Capital expenditure in the quarter reached $77.3bn ($75.9bn net of proceeds from sales and incentives); management reaffirmed its target of about $220bn for 2026 and indicated that 2027 will require even more. This is no longer the start of the heavy-spending cycle. We are in the middle of it, the stretch in which free cash flow has already turned negative.
This spending has been financed mainly with debt. Long-term debt rose from $65.6bn at the end of 2025 to $173.1bn at the end of the third quarter. That is a structural change in how the company funds its growth. Interest costs are already rising and will most likely keep rising. The quality of the new debt, however, remains unclear. If much of it is fixed-rate and long-dated, the pressure on free cash flow is more predictable. If floating-rate or short-dated borrowing dominates, sensitivity to the economic environment rises regardless of revenue growth. The report does not disclose this in detail, and when the information does emerge it may be incomplete. By our rough estimate, if more than 40% of the new debt is floating-rate, the hurdle for AWS growth rises from above 30% to above 35%, to offset the greater sensitivity to interest rates. The central question is whether returns on the new AWS investment will cover these new fixed costs within the available window, before systemic pressure sets in. If AWS growth stays above 30% (or above 35% with a high floating-rate share) and free cash flow starts to improve by the end of 2027, the case for a working, solid foundation is confirmed. If growth falls below 25% while debt keeps rising, it is refuted. Growth of 25-30% is a holding zone, in which free cash flow becomes decisive.
Guidance for the fourth quarter is revenue of $240bn-247bn (growth of 12-16%), with the top of the range slightly below the analysts’ average estimate. The gap is small, but it suggests that even at a record margin the company sees constraints on volume or pricing that are not visible in past numbers. Early warning signs over the next two quarters would be continued weakness in advertising revenue or AWS growth slowing below 35% as early as the first quarter of 2027. The market is already watching the tension between margin and free cash flow, so it may start to re-rate the stock at the first sign. If, after such a signal, the company announced a 10-15% cut to 2027 capital expenditure, that would show flexibility and limit the de-rating. If it did not, the first signal would become confirmation that the payback window is shrinking, and the re-rating would accelerate. Advertising revenue needs to grow fast. Management expects AWS growth above 40% in both 2027 and 2028.
The red flags are negative free cash flow in the middle of a heavy-spending cycle, the rapid rise in debt and interest costs, the unclear quality of the new debt, the slight miss on advertising revenue and fourth-quarter guidance whose top end sits slightly below the analyst consensus. The opportunities remain in the acceleration of AWS, the record operating margin, gains in retail market share and progress in newer businesses (pharmacy, Amazon Leo, Zoox). The tension, however, is between record profitability and negative free cash flow funded by debt, whose cost must be covered by the new capital within a defined window, provided the nature of the financing becomes clear in time and the company responds to the first signs by cutting capital expenditure.
The share price closed at $240.35 on October 30th (down 1.4% on the day and 5.6% over two days), giving a market capitalisation of about $2.6trn. The analyst consensus remains Buy, with an average target of about $331 (upside of about 38%). After the report several analysts trimmed their targets precisely because of capital spending (Morgan Stanley to $335, Bank of America to $317, Scotiabank to $300). Source for the average target and rating: LSEG via TradingView, as of October 30th 2026.
Any strategy should recognise that Amazon remains a core holding in an AI-oriented portfolio, but one with heavy capital spending and a shifting balance sheet. It makes sense to hold it as such, with discipline over its weight in the portfolio. For asymmetry, long-dated call options or call spreads limit the risk if pressure on free cash flow persists. Trimming on a break below key support levels or on signs of slowing AWS growth is the disciplined approach. Capital preservation comes first: heavy debt and negative free cash flow increase sensitivity to macroeconomic shocks and to any delay in the investments paying off. Amazon has more flexibility to defer part of its AWS investment if demand slows, because it already has a working business, and it can fund part of 2027 with debt it has already raised, without new financing, provided it does not have to raise capital expenditure further still. This is not investment advice but a framework for your own judgement.
Intel, for its part, reported revenue of $16.5bn (+21% year on year), in line with expectations. By segment, Client Computing and Physical AI Group reached $8.1bn (-5%), Data Center and AI $7.3bn (+78%, its strongest growth in 15 years according to the company) and Intel Foundry $6.7bn (+58%, most of it intra-company). Data Center and AI posted operating income of $3.0bn, a margin of about 42%, while the client division earned $2.3bn. Foundry remained loss-making at $1.85bn, though the loss is narrowing. Revenue from purpose-built silicon more than doubled year on year and rose by more than 50% sequentially - a strong performance that does not yet change the overall picture. Non-GAAP earnings per share were $0.52 (above a consensus of about $0.40), and GAAP EPS was $0.60.
Gross capital expenditure for 2026 is unchanged at $20bn, with a strong increase expected in 2027 as production ramps on 18A and Intel 3 and preparation for 14A continues. Funding was bolstered by a $23bn share sale in August, which lifted cash and short-term investments to about $54bn. Operating cash flow in the quarter was $5.1bn, and adjusted free cash flow was positive at about $1.6bn. Management, however, states explicitly that 2027 will be an investment year, probably without positive free cash flow.
The issuance of new shares has a tangible effect. For the fourth quarter the company expects the average diluted share count to reach 5.44bn, against 4.53bn a year earlier (about 20% more). This means that even if Data Center and AI keeps growing, earnings per share will grow more slowly than they would have without the new issue. Merely to bring earnings per share back to where they would have been on the old share count, earnings must rise by about 17-20%. That hurdle demands near-flawless execution in Data Center and AI. The division already runs at a margin of about 42%. At that level, a fall in prices is hard to offset with higher volume and feeds almost directly into lower operating income. Large customers have ever more bargaining power, because they are raising their own capital spending and seeking alternatives. If price pressure comes not only from oversupply but also from customers’ stronger negotiating position, the high margin becomes more vulnerable still. If the company tried to offset pressure from large customers by charging smaller ones more, it could speed their move to other suppliers and narrow the customer base. If volume growth does not offset price pressure, the requirement for 17-20% higher earnings to compensate for the new shares becomes even harder to meet.
Fourth-quarter guidance is weak: revenue of $15bn-16bn (against a consensus of about $17.1bn) and non-GAAP EPS of $0.25 (against about $0.43). The reasons are continued weakness in the PC market and an incident at Fab 28 in Israel, with an impact of about $300m on revenue and 200 basis points on gross margin. The incident affects almost exclusively 10-nanometre server products - just when Data Center and AI needs to deliver strong returns. For the fourth quarter management expects Data Center and AI to grow by about 15% from the third quarter, while client division revenue shrinks by more than 20%.
The PC market is structurally weak. Management expects unit sales to fall by about 12% in 2026 and by another 10% or so in 2027. This is not a short-term dip but multi-year pressure on what was until recently the largest division. External Foundry revenue was $411m in the quarter (less than 3% of total company revenue), part of it from a one-off customer programme. The improvement in yields and costs on 18A is real, but the absence of a large external contract after years of investment shows that big customers still judge the risk of moving significant volume to a new process node to outweigh the benefit. If that persists, capital spending on 14A will be funded on the assumption that internal demand covers a larger share of capacity. That deepens dependence on Data Center and AI just as the PC market contracts. If internal demand does not fill enough capacity, 14A spending may be cut or delayed, which in turn makes Foundry less attractive to external customers and could slow the technology roadmap. If customers see the roadmap slipping, they may scale back their orders too, tightening the vicious circle and making external contracts even less likely. Progress in process is not progress in the business model.
The red flags are the multi-year decline in the PC market, continued losses at Foundry without significant external customers for the new process nodes, the issuance of new shares and the need for 17-20% higher earnings to offset the larger share count at an already high margin, which reduces flexibility in the face of price pressure from oversupply, from large customers’ stronger bargaining power and from the potential loss of smaller customers, the operational risk at Fab 28, management’s expectation of negative free cash flow in 2027 and the risk of fresh financing needs if targets are missed. The opportunities lie in Data Center and AI (the strongest growth in 15 years, higher average selling prices for servers) and in improving 18A production. The pressure comes from the combination of a strong Data Center and AI business with a structural decline in the PC market, accompanied by the issuance of new shares and a still loss-making Foundry.
For Intel, success over the next four quarters would look like this: Data Center and AI sustains growth above 50% year on year for the next two quarters, and external Foundry revenue exceeds $1bn a quarter by the end of 2027. The effect of the new shares could then be offset. If growth falls below 30% and external revenue stays below $500m, the larger share count becomes a burden and the company may need to raise fresh capital as early as 2027, which would intensify the process further. Growth of 30-50% is a holding zone, in which Foundry’s external contracts become decisive.
The shares closed at $65.76 on October 30th after a 28.6% plunge on the day - the stock’s worst day since the 1970s, wiping out close to $140bn of market value. Market capitalisation now stands at about $349bn. The consensus is Hold, with an average target of about $98 (upside of about 49%), but after the report many analysts cut their targets (JPMorgan and UBS to $80) and Bank of America downgraded the stock to Neutral. Part of Intel’s re-rating has already begun with the October 30th sell-off. Source for the average target and rating: MarketBeat, as of October 30th 2026.
This position calls for heightened caution. If it is part of a portfolio, it should be smaller and carry a clearly defined risk. After such a collapse volatility is high, which makes put spreads or protective structures more sensible than an outright long position until the PC market stabilises and Foundry shows progress. Asymmetry can be sought through long-dated call options if Data Center and AI keeps delivering, but position size must allow for the risk of further disappointments in guidance. Capital preservation is paramount: the company is in transition, and an execution misstep could prolong the period of weak cash flow. Intel has less flexibility to cut capital spending if demand slows, because fabs take a long time to build and carry high fixed costs, and less ability to avoid another round of financing if targets are missed. This is not investment advice but a framework for your own judgement.
For both companies, free cash flow and the way capital spending is financed will matter more than revenue and operating profit over the next four to six quarters. Amazon is financing its capital spending with debt on top of an already profitable business with record margins. Intel is doing so by issuing shares, with Foundry still loss-making and the PC market shrinking. The difference lies in the quality of the balance sheet, the quality of earnings and the return on new capital.
Amazon is not the only company with record capital spending. Alphabet, Microsoft and Meta are also raising theirs by billions of dollars. If everyone builds at once, the risk is not only whether there will be enough demand for Amazon’s services, but whether the price of computing power falls before the investment pays off. If oversupply emerges after 2028, Amazon has time to repay part of its debt. If it emerges as early as 2027, while free cash flow is still negative, the pressure on AWS profitability will arrive just as interest costs begin to weigh on results. Amazon has a diversified revenue base - retail, advertising and AWS. Intel depends more on pricing in Data Center and AI, because the PC market is shrinking. If server processor prices fell by 15-20% because of oversupply or large customers’ stronger bargaining power, the blow to Intel would be harder and faster, because the division’s high margin leaves little room to offset price pressure with volume.
When Rome’s public finances were sound, it could maintain its aqueducts even through temporary dips in revenue. When its finances deteriorated, even well-engineered channels became impossible to keep up. The same logic applies today. Amazon can still withstand a spell of negative free cash flow, because its foundations are sound. Intel cannot afford the same, because its foundations are shaking and the issuance of new shares is already a fact. The question is not only whether the foundations will hold and whether returns will arrive fast enough to help fund the investment. It is also whether information on the quality of the debt will be complete and timely enough. It needs to allow the AWS growth hurdles to be adjusted before the window in which free cash flow must start improving closes. And, finally, whether the company will respond to the first warning signs by cutting capital spending before the market re-rates the stock.
Frequently asked questions
How will we know whether the Amazon thesis is being confirmed or refuted?
Watch three indicators: AWS growth, free cash flow and advertising revenue. Confirmation: AWS growth above 30% (above 35% with a high share of floating-rate debt) and improving free cash flow by the end of 2027. Refutation: AWS growth below 25% while debt keeps rising. Early warning: AWS below 35% in the first quarter of 2027 or another miss in advertising. What matters then is whether the company cuts 2027 capital spending by 10-15%.
How will we know whether the Intel thesis is being confirmed or refuted?
Watch Data Center and AI growth and external Foundry revenue. Confirmation: growth above 50% over the next two quarters and external revenue above $1bn a quarter by the end of 2027. Refutation: growth below 30% and external revenue below $500m, which raises the risk of another round of financing.
Why is the quality of Amazon’s debt part of the test rather than a footnote?
Because interest costs are already rising. If the floating-rate share is large, the hurdle for AWS growth goes up. If information on the type of borrowing is disclosed too late or incompletely, it will be hard to judge in time whether the AWS hurdle needs raising. If growth falls below 25% and debt keeps rising, whatever its quality, the pressure on the company becomes tangible and changes our assessment.
Why is Intel’s dependence on internal orders a risk?
Because the lack of external contracts increases dependence on internal demand. If 14A capital spending is cut or delayed, the roadmap slips, and that deters future external customers too. The conditions for success become harder to meet. The risk is confirmed if Data Center and AI growth falls below 30% and external revenue stays below $500m.
How do the two companies differ in their ability to avoid new financing?
Amazon can fund part of its 2027 needs with debt it has already raised, provided it responds to the first warning signs - AWS growth slowing below 35% or continued weakness in advertising revenue - by cutting capital spending. Without such a response, that ability diminishes. Intel has less ability to avoid another round of financing if it misses its targets for Data Center and AI growth and external Foundry revenue, whatever its response, because its foundations are already shaking.
Disclaimer: This article is for educational and informational purposes only and does not constitute investment advice or a recommendation to buy or sell securities or other financial instruments. All investments carry a risk of capital loss. Markets are volatile and can change sharply under the influence of geopolitical, economic and other factors. Do your own research, consult a licensed financial adviser and consider your personal financial situation, risk tolerance and investment objectives before making any investment decision. The author and the platform accept no liability for losses arising from actions or omissions based on the information in this article. Invest responsibly.
