Happy Diwali and Naya Saal Mubarak! What a week to light the diyas: a blue wave, the first national referendum on AI data centers, the worst day for chips since the Covid crash, and market message to Mag7 - cash flow matters!
In my August post, The Reckoning, I argued the market had moved from blind faith to "show me the ROI" and that hyperscalers would soon adopt a "no capex increase" tune. Now voters have joined the CFOs in asking the same question. Here's my take...
1. Shubh Deepavali: The Original Year-End Review
The Significance: Light over darkness, and a fresh set of books.
Diwali falls on Sunday, November 8 this year, on the new-moon night of Kartik Amavasya. Whether it's Rama's return to Ayodhya, Krishna's victory over Narakasura, or Lakshmi visiting homes that are clean and lit, the message is the same: knowledge over ignorance, hope over fear, light over darkness.
What I have always loved as student of economics, technology and markets is that Diwali is also the original fiscal year-end. For generations of Indian traders (I come from trader family myself), Lakshmi Puja meant Chopda Pujan, closing the old ledgers and blessing new ones. Vikram Samvat 2083 starts with a clean slate, and Muhurat trading remains Dalal Street's symbolic first trade. Indian merchants were doing annual reviews long before earnings calls.
So let's close the books on a very eventful week. Given what chips did on Thursday, some corners of the market could use a few extra diyas!
2. The Blue Wave and the Data Center Referendum
The Shift: From "AI is inevitable" to "not in my backyard"
Democrats won both chambers on November 3. The Senate flipped with wins that sounded like fantasy a year ago: Talarico in Texas (first Democrat since 1988) and Andrews in South Carolina (first since 1998). The Iran war and affordability were the headline issues. The sleeper issue was data centers.
AI showed up on the ballot as a local issue for the first time: power bills, water, land, and campuses that employ only few people once the cranes leave.
- Exit polls: Over half of voters called data centers a crucial or major factor (NBC).
- Texas: Gina Hinojosa unseated Gov. Abbott on a data center moratorium, in the AI buildout's own backyard.
- The pre-game: New York's moratorium in July, Abbott's own pause in August, and Monterey Park's permanent ban (86% support).
- The irony: Hundreds of millions in pro-AI super PAC money may be the worst ROI of any AI spend this year!
I'm calling this the birth of NIMBYtes: Not In My Backyard, applied to bytes, and to who pays for the grid behind someone else's chatbot.
The market's verdict came in two acts. Wednesday brought the usual "uncertainty is over" rally (S&P +1.6%). Thursday read the fine print: S&P -4.4% to 7,354, Nasdaq -5.5% into correction, VIX up 70% to 28. It was the worst second session after a midterm since at least 1950, concentrated in chips, AI hardware and power equipment.
Don't over-read Washington: a national moratorium needs legislation, a seated Congress and a President who won't veto it. The real bite is at the state and county level: permits, interconnects, water and zoning. Slower than a bill, but harder to lobby away...
3. Mag7 Earnings: Beat, Raise... and Get Spanked
The Shift: From "how much are you spending?" to "how much cash is left after you spend it?"
On paper, Q3 was great: Azure +48%, Google Cloud doubled (+103%), AWS +39%. Yet all six reporters fell the next session. The market is grading free cash flow and capex discipline, not revenue.
The pattern is exactly the one I flagged in August: reward restraint, punish the burn.
Company | 2026 capex plan | Capex signal for 2027 | Free cash flow |
|---|---|---|---|
Amazon | ~$220B (kept) | "Meaningfully more" | -$42.4B (trailing 12 months) |
Alphabet | $195-205B (kept) | "Increase significantly" | -$9.1B (Q3) |
Meta | $135-145B (narrowed) | Not guided yet | -$2.1B (Q3) |
Microsoft | ~$170B (cut from ~$175B) | Capacity slips into 1H27 | +$18.6B (Q3) |
Tesla | ~$20B (cut from >$25B) | Restraint | +$1.5B (Q3) |
From the tape:
- The cutters got a pass. Microsoft trimmed capex and stayed FCF positive: flat stock. Tesla cut capex a fifth: a mild -3.1%.
- The spenders got spanked. Alphabet missed on both lines, burned $9B of FCF, and grew long-term debt from ~$22B to $125B in a year. The buildout now runs on bonds, not just cash flow.
- The bill is landing on the consumer. Apple's new CEO called memory pricing "unprecedented" and Qualcomm's handset sales fell 26%. When AI eats the memory supply, your next phone pays for it.
Add it up: the Mag 7 spent about $647B on property and equipment over the last four quarters, up 82%, and the five that guide plan roughly $750B for 2026. That's nearly all of Big Tech's operating cash flow, leaving no cushion if growth, rates or politics turn.
4. Semageddon: Before the SaaSpocalypse Is Even Over
The Shift: From "capex is going up" to "capex is going up more slowly or peak capex is here" — and for chip stocks, that's the whole ballgame.
Software has rallied off its April SaaSpocalypse lows, but the scars aren't healed. And before that story is even finished, the baton has passed to the other side of the AI trade. Welcome to Semageddon.
Here is my hypothesis in one line: semiconductor stocks are priced on the second derivative of hyperscaler capex, not the level. Chip revenue tracks capex spend; chip valuations track capex growth. Capex doesn't need to fall for chips to crack; growth only needs to peak.
That's what Thursday started pricing:
- The peak moved forward. Dan Ives, usually the loudest bull, now sees capex growth peaking in 2027, not 2028-29, with 10-20% of 800-1,000 planned US data centers cancelled or delayed.
- The carnage: SOX fell 12% on Nov 5, its worst day since March 2020, and sits ~32% below last month's peak. Micron and AMD fell ~17% each, despite AMD's data center revenue nearly doubling.
- Suppliers executed, spenders acquitted: From Oct 28 to Nov 5, twelve AI hardware suppliers fell ~32% on average while Microsoft, Amazon and Meta were flat. Fewer new data centers make the existing capacity more valuable: the customer's relief is the supplier's pain.
- It went global: TSMC -10%, Advantest -12%, Kospi circuit breakers, Nikkei in a bear market.
The history rhymes, loudly. In 2000, carriers were still raising capex when Cisco peaked; the SOX went on to lose ~80% by 2002. The fiber got used, but its sellers never got their multiples back. Shovel stocks discount the end of the gold rush before the miners stop digging.
Now, the case against Semageddon. It is fair to note that this is, so far, a valuation event and not an earnings event. Nvidia guides ~$108B, memory is scarce, and 2027 visibility is intact. Nobody is cancelling orders; markets are just removing 2028-29 growth. And Tokenomics cuts both ways: fewer GPUs per token, but far more tokens.
My take: we are probably closer to peak capex growth than peak capex. That means a regime change in semis from "buy every dip" to "pick the bottleneck": memory, power, optics and custom silicon over pure capacity plays. Market is fearing "cuts" in capex in later part of 2027 and pricing the possibility now...
5. What to Expect for the Rest of 2026
The Shift: From a narrative market to a calendar market. The next eight weeks will be decided by four dates, not by headlines.
My New Year's Day range for the S&P was 6,000 to 8,000. We touched 7,819 on October 6 and closed Thursday at 7,354, about 6% off the top. Friday's bounce shows dip-buyers are still here, but they're waiting on these dates:
Date | Event | Why it matters |
|---|---|---|
Nov 10 | October CPI | Sets the tone for the Fed; oil near $97 isn't helping |
Nov 13 | October PPI | Pipeline inflation after the Iran-war oil spike |
Nov 17 | Nvidia Q3 FY27 | The Semageddon referendum; guide is ~$108B |
Nov-Dec | Lame-duck Congress | Spending deadlines; any AI or energy bills |
Dec 8-9 | FOMC | Futures price ~70% odds of a rate hike |
And then there is the real test of AI - Anthropic IPO. If that goes well, it may give short-term respite till the first earnings report.
Note the Fed line: under Warsh, the market debates not how many cuts but whether we get a hike. Warshnomics, meet NIMBYtes: costly capital plus local resistance is tough on a debt-funded buildout.
My scenarios for year-end:
- Range trade into Santa (55%): S&P chops between the 200-day (~7,320) and the election-day close (~7,570). Nvidia beats, but the market wants 2027 visibility. Leadership rotates to AI spenders, software, healthcare and financials.
- Relief rally (25%): Strong Nvidia guide, cooler CPI, fading hike odds. S&P retests 7,800; semis recover half the drawdown.
- Semageddon spreads (20%): A sustained break of the 200-day, more Qualcomm-style guide cuts, and moving state bills take the S&P to ~7,040, a 10% correction. The 2018 midterm path is the closest precedent.
Longer term, this week fits my August call of a 1929-style reckoning around 2028-29. Bubbles don't end when spending stops; they end when the growth stops, and shovel sellers get repriced first. This may have been the opening act.
The Bottom Line
The AI trade has a new stakeholder. On November 3, voters and utility customers joined the CEOs and CFOs, asking the same question: show me the return, and show me who pays for it.
That doesn't kill AI. It slows the pace of the buildout, and for its suppliers pace is everything. The SaaSpocalypse hit incumbents first; Semageddon hits suppliers first.
Investor takeaways:
- Watch the second derivative. Capex growth drives chip multiples; watch Nvidia on Nov 17 and January's 2027 capex guides.
- Own the bottlenecks, not the capacity. Pricing power over pure volume.
- Spenders over suppliers, for now. A slower buildout eases the hyperscalers' FCF squeeze and helps software.
- Map the politics locally. States and permits, not just Congress.
- Respect the Fed. A December hike is live. Keep dry powder; no margin for this dip.
- Keep the long view. Close the ledgers like a Diwali merchant and start Samvat 2083 clean.
Wishing you and your family a very Happy Diwali and a prosperous New Year. May the year bring more light to your portfolio and fewer diyas needed for the SOX!
/Shyam
Not investment advice; just my views. Do your own research.
Claude was used to do research and organize
Sources
- ABC News: midterm live results
- Futurism: exit polls on data centers
- Crypto Briefing: NY and Texas data center pauses
- Inquirer: Monterey Park data center ban
- SimianX: Mag 7 Q3 2026 earnings tracker
- SimianX: the November 5 stock market drop
- Schwab: SOX 32% below its peak
- 24/7 Wall St.: SOXX and SMH weekly losses
- Vedantu: Diwali 2026 date
