Skip to main content
AI Bottlenecks Put to the Test in Quarterly Reports: The Fall 2026 Review

AI Bottlenecks Put to the Test in Quarterly Reports: The Fall 2026 Review

Back in August, we had identified capital expenditures, memory, and edge computing as the key areas to watch. The October quarterly reports confirm these trends in the financial statements, but the stock market has begun to demand answers: what has held up, what hasn’t, market sentiment, and the schedule of upcoming events.

by Francesco Punta 8 min read

In August, across three articles, we tried to understand where the race for artificial intelligence was being decided: in the budgets of those spending the money, in the scarcity of memory, and in the silicon that’s taking AI outside of data centers. Here is the promised follow-up: a review, complete with financials and market reactions as of October 31, 2026. At the bottom, you’ll find a calendar of upcoming quarterly earnings reports—just one click to add them to your calendar.

The three August theses, in brief

  • Capex isn’t slowing down. In the article on bottlenecks, the combined investments of Microsoft, Amazon, Alphabet, and Meta for 2026 were estimated at nearly $725 billion—77% more than in 2025. Even then, Alphabet’s post-earnings decline was “the first crack.”
  • Memory is the bottleneck, but it’s cyclical. HBM is in short supply, prices are rising, and a word of caution: memory cycles will reverse sooner or later.
  • Edge silicon is shared infrastructure. In our article on local AI and robotics, we highlighted Qualcomm and CEVA, with a specific catalyst: the Qualcomm–ARM deal on October 5.

1. Capex: Companies are accelerating, and the market is demanding a reckoning

Looking at the numbers, the thesis holds up. Meta invested $34.9 billion in the third quarter—nearly 80% more than a year earlier—and narrowed its 2026 forecast to $135–145 billion. TSMC, which manufactures those chips, closed the quarter with $46.7 billion in revenue (+41%), exceeding its own forecast.

Meta: Third-Quarter Capex Nearly Doubled

Capital expenditures, including finance leases ($ billions)

$19.4 billion
Q3 2025
+80%
$34.9 billion
Q3 2026

Source: Meta, Q3 2026 earnings report (October 21, 2026)

The market’s reaction, however, has changed. During the week of October 28–29, investors focused primarily on one figure: how much cash remains after investments—free cash flow.

Same Investments, Opposite Reactions

The figure that moved the stock, the week of October 28–29, 2026

Alphabet

Free cash flow: −$9.07 billion (expected: +$2.7 billion). 2026 capex guidance confirmed at $195–205 billion.

−13% in a single trading session: the worst day since its 2004 IPO

Amazon

AWS +39% to approximately $46 billion. Negative free cash flow; 2026 capex around $220 billion.

Down in after-hours trading despite the cloud business

Microsoft

Capex of $45.2 billion versus an expected $50.7 billion. Free cash flow down but still positive ($18.6 billion).

Stock stable: spending less than expected was rewarded

SK Hynix

Operating profit of 83.3 trillion won (consensus around 77), margin 80%.

Up after earnings report

Sources: CNBC (October 29, 2026), company press releases and earnings calls, Motley Fool, Investing.com

The clearest example is Alphabet: results described as “messy” by analysts, cash burned rather than generated, and the stock losing about 500 billion in value in a single day. The “first crack” in August has turned into a fracture: the market doesn’t dispute that demand for AI exists; it disputes who pays for it and when it pays off.

Google’s core issue: AI is less profitable than traditional search

Behind the collapse lies a business model issue, which Alphabet itself has acknowledged. Search advertising, its most profitable business, is worth $63.3 billion per quarter, but growth has dropped to 12%, down from 17% in the second quarter and 19% in the first. During an earnings call, Chief Commercial Officer Philipp Schindler attributed this in part to AI: searches in AI Mode are growing significantly, but currently generate less revenue than traditional searches. At least, Google isn’t saying so.

The announced countermeasures are concrete: more ads within AI Overviews and AI Mode—for the first time, even outside the United States (the United Kingdom, Brazil, and Mexico by the end of the year)—and ads in follow-up questions during a conversation. According to Google, users find these ads relevant, and clicks on product ads are already comparable to those on organic results. However, no timeline has been provided for closing the gap. In my opinion, for now, AI in search is primarily a defensive measure—a price paid to keep users from switching to competing assistants—while the AI that’s truly paying off is in the cloud, which has grown by 103%. It’s the classic dilemma of having to cannibalize your best business before someone else does: we’ll explore this in depth in a dedicated article, including what it means for those who own a website and rely on traffic from Google.

Verdict: a thesis confirmed in financial statements but contested on the stock market. We’ll explain why a stock can plummet even with rising revenues in our next article, which will focus on how to read a quarterly report.

2. Memory: Record Profits, Falling Stock Prices

SK Hynix closed the strongest quarter in its history: revenue of 104.6 trillion won—surpassing the 100 trillion mark for the first time—and operating profit of 83.3 trillion won, with a margin of 80%, exceeding analysts’ consensus (approximately 77%). Management says that customer demand is more than double what it can produce, that long-term contracts cover more than half of its capacity for 2027–2028, and that 2027 capital expenditures will rise significantly. On October 8, Samsung had forecast a record operating profit of approximately 107.4 trillion won.

Yet memory stocks had a rough October: over the five trading sessions through October 15, SK Hynix lost 26%, Samsung 22%, and Micron about 11%. Two reasons have been cited. China’s CXMT announced a new, higher-density server memory chip, which could take market share away from the three industry leaders. And one forecast—so far an isolated one—points to a wave of new supply in 2029. Against this backdrop are rising interest rates and doubts about how AI spending will be financed.

This is where it’s important to distinguish between causes and symptoms. Memory prices are still rising: TrendForce estimates DRAM prices will increase by 10–15% in the fourth quarter. The cyclical risk we discussed in August, therefore, is not yet reflected in chip prices, but it is already evident in stock prices, which are anticipating the years 2028–2029. Even the most cautious analysts view this as a return to normalcy rather than a collapse: Counterpoint estimates DRAM prices will fall by 23% in 2028, but will still be more than triple the 2025 level.

Verdict: scarcity confirmed, cycle anticipated by the stock market. The memory cycle and the Chinese challenge deserve a separate in-depth analysis, which will be covered in upcoming articles.

3. Edge AI: The Qualcomm–ARM Case Ends with a $2 Award

On October 13, a Delaware jury ruled in Qualcomm’s favor regarding ARM’s breach of the licensing agreement, but awarded only $2 in total damages and ruled in ARM’s favor on four other claims. The risk we had highlighted (Qualcomm accounts for about 9% of ARM’s revenue) has been mitigated; ARM’s appeal regarding the previous case—the one involving Nuvia cores—remains pending. The real tests will be the earnings reports from Qualcomm and ARM on November 4 and from CEVA on November 9, as CEVA must demonstrate its ability to convert NPU licenses into royalties.

Market sentiment: fear, not panic

CNN’s Fear & Greed Index summarizes seven U.S. market indicators on a scale from 0 to 100, each weighted equally: the S&P 500’s momentum relative to its 125-day moving average, the number of stocks hitting annual highs or lows, the volume of rising stocks versus falling stocks, the ratio of put options to call options, expected volatility (the VIX index), the yield spread between risky and safe bonds, and the extent to which government bonds outperform stocks (demand for safe-haven assets).

Fear & Greed Index: in the fear zone

From 0 (extreme fear) to 100 (extreme greed). Below 45: fear; below 25: extreme fear

2,630 Oct
3,429 Oct
161 weeks ago
301 months ago
371 years ago

Source: CNN Business, Fear & Greed Index, updated October 30, 2026 (7:59 p.m. ET)

At the close on October 30, the index stood at 26, indicating “fear”; it had been at 34 the day before and at 16—indicating “extreme fear”—a week earlier. All seven indicators are in the “fear” zone, and two are in “extreme fear”: S&P 500 momentum and the flight to safe-haven assets. The picture is one of fear that had subsided but was reignited by Alphabet’s and Amazon’s earnings reports.

Two caveats. This is a sentiment indicator, not a value indicator: it reflects how investors feel, not how much companies are worth. And historically, it has often been more useful when interpreted in reverse, because widespread fear tends to appear near market lows, not highs. It’s the same emotional trap we discussed in the article on retirement and ETFs: reacting to the “stock of the day” rather than to the investment thesis. You can also find the technical terms in this paragraph in the investor’s glossary.

The Report Card as of October 31

  • Capex accelerating—confirmed in financial statements, disputed on the stock market. The market rewards those who spend less than expected and punishes those who burn through cash.
  • Memory shortage — confirmed. Record profits and prices still on the rise.
  • Memory cycle — anticipated by the stock market, but not yet by prices. Watch: CXMT and smartphone demand.
  • Qualcomm–ARM legal risk — downplayed. Symbolic damages; Nuvia’s appeal still pending.
  • CEVA, from licenses to royalties—pending. Response on November 9.

The calendar: quarterly earnings reports that round out the picture

Past events remain visible in gray. You can add upcoming events to Google Calendar, Outlook, or Apple Calendar, or subscribe to the full calendar: if a date changes, I’ll update it here, and your calendar will update automatically. After NVIDIA and Micron report their earnings, I’ll update this article with the final picture.

Times in US Eastern Time. Your calendar converts them to your own time zone. Google Calendar subscriptions can take a few hours to refresh.

  1. SK Hynix: Third-Quarter 2026 Results

    08:00 PM EDT

    The most direct test of HBM memory. Consensus estimates put operating profit at around 77 trillion won (Q2: 60.5). Key points to watch: HBM4, long-term contracts, margins.

    Source
    Google CalendarOutlook
  2. Alphabet: Third-Quarter 2026 Results

    04:30 PM EDT

    Capex 2026 and Google Cloud's Growth. Following the stock's decline this summer, the focus is on the return on investment.

    Source
    Google CalendarOutlook
  3. Microsoft: First-Quarter Results for Fiscal Year 2027

    05:30 PM EDT

    Data Center Spending and Azure's Growth: Is Capex Continuing to Accelerate?

    Source
    Google CalendarOutlook
  4. Samsung Electronics: Final Results for the Third Quarter of 2026

    09:00 PM EDT

    Final figures following the October 8 estimate (operating profit of approximately 107.4 trillion won): how much does this figure matter, and what does it tell us about the fourth quarter?

    Source
    Google CalendarOutlook
  5. Amazon: Third-Quarter 2026 Results

    05:00 PM EDT

    Capex 2026 and AWS Growth.

    Source
    Google CalendarOutlook
  6. Qualcomm: Fourth-Quarter Results for Fiscal Year 2026

    04:45 PM EST

    First-quarter results following the ruling against ARM: automotive, AI PCs, and the new robotics division.

    Source
    Google CalendarOutlook
  7. ARM Holdings: Second-Quarter Results for Fiscal Year 2027

    05:00 PM EST

    Initial figures following the lawsuit with Qualcomm, which accounts for about 9% of ARM's revenue.

    Source
    Google CalendarOutlook
  8. CEVA: Third-Quarter 2026 Results

    08:30 AM EST

    The Litmus Test for the "Edge" Theory: Are NPU Licenses Starting to Turn into Royalties?

    Source
    Google CalendarOutlook
  9. NVIDIA: Third-Quarter Results for Fiscal Year 2027

    All day · time to be confirmed

    Data center revenue and comments on CoWoS and HBM packaging. Date announced by the company; time to be confirmed (usually after the close of Wall Street).

    Source
    Google CalendarOutlook
  10. Micron: First-Quarter Results for Fiscal Year 2027Estimated date

    All day · time to be confirmed

    Season Wrap-Up: DRAM and HBM Prices, and Whether October’s Drop in Memory Stock Prices Is Reflected in the Numbers. Estimated date; not yet announced by the company.

    Source
    Google CalendarOutlook

Published on

Share:
Francesco Punta

Written by

Francesco Punta

Full stack web developer and data analyst

I build websites and web applications and work with data. On this blog I write about technology, artificial intelligence and personal finance, starting from numbers and verifiable sources to understand what is really happening.

Articles on the same topics

AI Bottlenecks in 2026: Chips, HBM Memory, Packaging, and Power

Artificial Intelligence · Investments

AI Bottlenecks in 2026: Chips, HBM Memory, Packaging, and Power

The quarterly financial reports from Big Tech and chipmakers show a race that’s accelerating, not slowing down. But the real bottleneck is no longer silicon: it’s packaging, memory, and electricity. Real data on the AI hardware supply chain in 2026.

12 min read