The whole AI trade on one page — distilled every day into one word: Buy or Sell. Below that call: market sentiment, the AI stock tape, the neocloud buildout, the hyperscalers writing the checks, the data-center build stack with a favorability indicator, Michael Burry's shorts, the macro debt picture, regulation, and how today's valuations stack up against past bubbles.
How this page works: the market panels (tape, indexes, VIX, Fear & Greed, debt clock) update live, in place, every 5 minutes. The trackers, news and analysis are the daily edition shown in the masthead. Anything that can't update is labeled EDITION — nothing is filled in.
Stocks are bouncing hard this morning after last week's AI-led selloff, and the fear gauge is easing. But broad-market sentiment is still in fear, at the edge of extreme fear, Nvidia is falling on its own board probe, and tomorrow's CPI print — with markets pricing a ~78% chance of a December Fed hike — decides whether this bounce holds.
Hey — sell today.
Sell into this morning's bounce, don't chase it. The indexes are up, but the leader (Nvidia) is falling on its own, hype is still crowded at 75, and tomorrow's CPI — with a December Fed hike priced at ~78% — can end the bounce in one print. The spending story underneath is still real; today is about price and timing, not the long-term story.
How it works, in three steps: ① Read every section below. ② Score each of the nine checks below as leaning Buy or Sell — no neutrals. ③ Majority wins, and that one word is the day's strategy. If leadership (Nvidia) and the Fed/CPI check ever split the rest, they break the tie — they're what move this trade fastest.
Every check comes straight from a section below. Change the inputs, and tomorrow's word can change — that's the point of doing it daily.
On every load — and again every 5 minutes, in place, without reloading — this page pulls live quotes (Stooq), the CNN Fear & Greed reading and the Treasury debt anchor, then re-scores the call from four live inputs: index momentum, the VIX, Fear & Greed, and breadth across the tracked names. Score of 50 or higher = BUY, below 50 = SELL. If a feed doesn't answer, that input falls back to its edition value and says so.
Checking live feeds…
Analysis only — not financial advice, and no trade is placed. This is the Monitor's synthesis of the sourced reporting on this page. The market panels — the tape, the index band, the VIX, Fear & Greed and the debt clock — now pull live data in your browser on every load (Stooq quotes can lag the tape by ~15 minutes), and the call above is re-scored from those live numbers on each 5-minute cycle; the live re-score panel shows exactly which inputs were live. News, trackers and the other editorial sections remain the Nov 9 edition (data captured 9:40 AM ET). Any panel whose live feed didn't answer is labeled as edition data — nothing is filled in. AI Pulse Monitor is a publication: it places no trades, manages no accounts, and gives no personal advice — see the disclaimer at the bottom of this page.
CNN Fear & Greed Index, Nov 9, 2026. Seven indicators: momentum, breadth, options, volatility, safe-haven demand.
Edition reading — CNN Fear & Greed, Nov 9, 2026.
The Monitor's AI-specific gauge — how hot the AI trade itself is running, separate from the broad market.
Wall Street's fear gauge. Spiked ~70% last Thursday, now retreating — but still well above calm territory (under 20).
Edition reading — Nov 9, 9:40 AM ET.
Nvidia's board opened an independent investigation Sunday; the company says Jensen Huang continues as CEO. Stock indicated near $180 vs Friday's $186.64 close — leadership risk is back at the center of the AI trade, eight days before earnings (Nov 17).
Wall Street opened sharply higher as oil fell on Iran peace hopes (Brent −5.0% to $93.38, WTI −4.4% to $86.87) and the VIX pulled back. Dow +1.25% to 51,925.59.
While the indexes bounce, NVDA is the drag — down from its $239.24 record last month by ~24–26%. Key levels today: $172 support below, $194 resistance above.
Up from 22 at Friday's close and 19 a week ago, but a month ago it was 43. Put/call options score 8.2 and S&P momentum 8 — both extreme fear. The bounce is real; conviction isn't back yet.
October payrolls surprised hot (+165K vs ~60K expected, unemployment 4.1%), which is why a December hike is priced at ~78%. A hot CPI tomorrow likely kills this bounce; a soft one extends it. Fed speakers Bowman, Goolsbee and Barkin are also on today.
until CPI — Tue, Nov 10, 8:30 AM ET
| Ticker | Price | Today |
|---|
ARM, Salesforce and ServiceNow returned nothing from the finance feed this cycle — shown as “not in feed” rather than filled in with a made-up price, and retried on every 5-minute refresh. Nvidia earnings land Tuesday, Nov 17 — the next big test for the whole AI trade.
Neoclouds are the GPU-rental firms underneath the AI boom: they borrow heavily, buy Nvidia chips, and rent the computing time to Microsoft, OpenAI, Anthropic and Meta — while a second group of ex-bitcoin miners leases them the powered buildings. They are the most leveraged link in the AI chain, so this tracker follows both the stocks and the commitments behind them: contracts signed, expansions passed on, rented capacity switched off, and fleets being decommissioned.
| Ticker | Price | Today |
|---|
The group fell 18–50% in the three weeks into Nov 5 — Nebius −17.6% and CoreWeave −14.3% on Nov 5 alone — before bouncing with the rest of the tape this morning. Every listed name runs annualized volatility above 90%. Private names tracked below without prices: Nscale, Lambda, Crusoe, Fluidstack.
Microsoft declined an option on ~700 MW more at Nscale's Barstow, Texas campus (it keeps the base contract and first refusal on the unused capacity), walked away from anchoring Nscale's ~$3B Loughton campus in England after the grid operator couldn't guarantee power until the early-2030s, and let a 1.35 GW West Virginia letter of intent lapse. Bloomberg reports it is also debating whether to sign new third-party cloud deals at all — while adding $200B+ of leases for facilities it runs itself.
After cabling faults, an April generator fire and August hardware delays, Microsoft agreed a revised timeline with IREN that could push the Childress, Texas campus back weeks or months, and stepped up audits of IREN, Nebius and Nscale. The suppliers paid for the caution in one session: Cipher Digital −21%, IREN −18%, CleanSpark −15%, Nebius −13%, CoreWeave −9%.
The 2023 deal that rented Oracle cloud servers for Bing's chat search was not renewed, and the machines — older-generation Nvidia chips that no longer meet Microsoft's specs — have been physically switched off over recent months. A direct read-across for Oracle (ORCL), already the tape's weakest name and Burry's best short.
IREN spent the year retiring mining hardware and redirecting power and data-center capacity to AI cloud — AI revenue passed mining revenue for the first time in the June quarter — and targets substantially completing the switch by Dec 31, 2026. The June quarter carried $450M of impairments on the way out, against $1.98B of construction spending in that quarter alone. Its $9.7B Microsoft contract (20% prepaid) anchors the Childress campus.
Microsoft is up to ~$43.8B of Nscale's $103.4B total contract value, ~$17.4B at Nebius, $9.7B at IREN, and was ~67% of CoreWeave's 2025 revenue. Anthropic, Meta and OpenAI are still adding to backlogs — CoreWeave's stands near $104B and Applied Digital counts ~$36B of contracted lease revenue — but one tenant's change of mind reprices the whole sector, as last week proved.
CoreWeave reports Tuesday, Nov 10 (5:00 PM ET) — the question is whether billed megawatts turn into collected cash against that $51.6B debt stack. Nebius follows Thursday, Nov 12 before the open. IREN already reported: revenue timing pushed into early next year on network redesigns at Childress.
The five giants writing the checks the whole AI buildout depends on. Spending is still accelerating — but the pattern changed this month: the hyperscalers increasingly want to own the halls instead of renting them from neoclouds, and voters, grids and lenders are all starting to set terms. This tracker follows every dollar of guidance and every deal walked away from.
| Company | 2026 capex plan |
|---|---|
| AMZNAmazon / AWS | ~$220B |
| GOOGLAlphabet / Google | $195–205B |
| MSFTMicrosoft | ~$175B |
| METAMeta Platforms | $135–145B |
| ORCLOracle | $90–95B gross* |
Capex definitions differ by company (Microsoft and Meta include finance leases; Amazon reports purchases of property and equipment; Alphabet is cash capex) — the rows are close cousins, not identical measures.
On top of the neocloud moves tracked above — the declined ~700 MW Nscale Texas option, the canceled ~$3B Loughton (UK) anchor, the lapsed 1.35 GW West Virginia LOI and the Oracle wind-down — Microsoft cut its 2026 capex forecast by $5B, blaming third-party hardware delays, and offered voluntary buyouts to ~60,000 employees (Oct 26) while guiding December-quarter capex above $50B. The direction is unmistakable: own the halls, rent less, and hold outside builders to a harder quality bar.
Oracle cancelled a $7B chip lease to Tencent after U.S. officials objected, losing contracted income, and lost the Microsoft rental arrangement (see Neoclouds). To fund the build it raised $20B — $10B of new shares sold at an 11% discount and $10B of bonds at ~7.8% — diluting holders and raising the profit bar every new hall has to clear.
Trenton, Ohio voters banned data centers above 25 MW with roughly 80% of the vote on Nov 3, closing the annexation path for an ~1,000 MW, 18-building AWS campus on ~600 acres — the city had already withdrawn the annexation petition in September. One of 41 local measures decided Nov 3; 40 went against the industry (see Regulation).
About 67% of Piqua voters backed a charter amendment aimed at Meta's planned 1.4 GW campus, and recalled the mayor and two commissioners over it (Cleanview tally of unofficial returns). Meta's biggest political fight is no longer in Washington — it's at city hall.
Google announced its first Wyoming data center on Nov 5 ($17B, Laramie County) and the next day acquired Crusoe's southeast Wyoming site outright (price undisclosed). Crusoe had been pushed aside at the project in June after Google raised cost and timetable concerns; it keeps Google's Armstrong County, Texas build. Even the “good” hyperscaler news now reads: we'll take the site, we'll run it ourselves.
ING plans to sell roughly $500M of data-center loans, including a ~$25M slice tied to the Stargate site in New Mexico. Small money next to $800B of capex — but the first sign that bank appetite for AI data-center debt is no longer unlimited, which is exactly the channel the neoclouds borrow through.
The build, in layers: a data center is not one building — it's seven systems that all have to land at the same time. Miss one — power, water, fiber, a transformer — and the GPUs sit dark. This section walks each layer in build order, names the companies that supply it, puts a price on the whole stack, and rolls it into one Build Favorability Indicator: is this market favorable, for whom, and who is getting hit?
Favorable for the suppliers who sell the scarce layers — power gear, cooling, fiber, turbines, powered land. Not favorable for leveraged builders, late projects without power, or anyone paying today's prices without a signed tenant. Demand is not the constraint; electrons, permits and the cost of money are. That split — thriving picks-and-shovels vs. squeezed builders — is the indicator's whole point.
Score = demand and pricing power for suppliers in that layer, minus lead-time, cost, permitting and financing friction. Low is not “no money” — it means the layer blocks the whole build.
Blended 44 is an average of unlike things — which is why the layer scores matter more than the headline. A project scoring 28 on power does not get rescued by a 72 on fiber.
Site near transmission, fiber and water, zoned and permitted. Powered land in primary U.S. markets averages ~$584K per MW, up ~51% in a year (Cushman) — and voters are now part of the permit: 40 of 41 local data-center measures on Nov 3 went against the industry (see Regulation). Land itself is only ~7% of the all-in facility stack; a site without a power path is worth close to zero.
Everything starts here. Grid connection at high voltage, a private substation, transformers stepping power down, switchgear, UPS/batteries and backup generators — plus the generation behind it (gas, nuclear restarts/SMRs, renewables + storage). Electrical systems are 40–45% of construction cost. A 1 GW campus draws roughly as much as ~800,000 homes, around the clock. Hyperscalers now sign directly with generators — Microsoft's Three Mile Island restart deal is the template.
A 120 kW AI rack is a very expensive heater. Air cooling no longer suffices at 80–150 kW racks, so liquid goes straight to the chip: cold plates, coolant distribution units (CDUs), heat exchangers, then cooling towers or dry coolers outside. Liquid cooling runs ~$4.5–5.2M per MW vs ~$1.8M for air. Evaporative towers consume water daily and draw local opposition; closed-loop / zero-evaporation designs recirculate a one-time fill (Microsoft cites ~125M litres saved per site per year) but cost more up front.
Two diverse fiber entry points minimum, long-haul routes to other regions, and inside the hall a fabric that makes thousands of GPUs act as one machine: NVLink inside the rack, InfiniBand / Ethernet between racks, optical transceivers turning data into light and back. This is the least blocked layer — fiber can be trenched and lit faster than power can be generated — so suppliers grow without carrying the grid's lead times.
Sitework, structure, white space, busway/PDUs, fire suppression, security and building systems. Counter-intuitively the building is the smaller bill: shell + site is ~15–18% of construction cost; labor is 40–50% of total construction cost across trades, and skilled electricians are the scarce trade. General and electrical contractors with data-center crews are booked out — good for them, a cost and schedule risk for owners.
The reason the building exists — and the biggest check. One Nvidia GB200 NVL72 rack: 72 GPUs, ~120 kW, roughly $3M a rack. Chips, servers and network add ~$29–42M per MW on top of the facility — more than the building itself — and refresh on a 3–5 year cycle, not the building's 20–30 years. That mismatch is Burry's depreciation argument in physical form (see Burry / Scion).
Commissioning, 24/7 operations, physical + cyber security, maintenance, and the power bill forever: electricity is 50–70% of operating cost, so PUE (power usage effectiveness — overhead per watt of compute, ~1.2–1.4) flows straight to margin. A finished hall without cleared operators and a power contract is still not a business. Staffing experienced data-center technicians is a genuine constraint in new markets.
Construction cost split, shell + core — before GPUs. Power + cooling together are 60–70%: the building is almost an afterthought.
Splits synthesized from JLL / Turner & Townsend / Cushman & Wakefield 2024–26 analyses; Cushman's all-in guide puts power infrastructure alone at ~21% (~$3.7M/MW) before in-building electrical. Scope differs by source — never compare per-MW figures without the scope line.
Facility = Cushman all-in average $17.6M/MW ex-GPUs. All-in adds chips/servers/network at a ~$35M/MW midpoint of the $29–42M range. Ranges, not quotes — market, redundancy and cooling design move the facility figure from $8.9M to $23.3M per MW before a single GPU.
| Campus size | Facility only (ex-GPUs) | All-in incl. compute |
|---|---|---|
| 50 MWSingle large hall cluster | ~$0.9B | ~$2.6B |
| 100 MWStandard hyperscale campus | ~$1.8B | ~$5.3B |
| 500 MWMulti-hall AI campus | ~$8.8B | ~$26B |
| 1,000 MW (1 GW)Stargate-class | ~$17.6B | ~$50–60B |
Estimator uses the published benchmarks on this page only: standard facility $11.3M/MW shell + core ($17.6M all-in ex-GPUs); AI facility $20–37M/MW fitted; compute +$29–42M/MW. Operating cost is excluded — at 100 MW, power alone is roughly $100M+ a year at typical industrial rates and PUE ~1.3.
The public page stays free — the Daily Call verdict, the sentiment scores, the tape snapshot. Subscribers get the reasoning, the full trackers, the archive and the tools behind it. Paid checkout opens as soon as our pre-launch legal review is complete; join the list below and you'll be first in when it does. No card is taken today.
The public page — no card, ever
Or $240/year — two months free
Or $490/year — two months free
One email each market morning: the call, the score, and what would flip it. Paid plans launch to this list first. Checkout is not open yet — no payment is taken or requested today.
Everything in the two trackers above points the same way: demand and hyperscaler spending are not the problem ($237B spent in one quarter, guidance raised, backlogs near $1.9T). The problem is who carries the risk. The anchor tenant is walking away from rented capacity, lenders are testing the exit, ballot boxes are killing campuses, and the neoclouds fund it all with double-digit debt. Sentiment for the space overall: strained — bullish the spend, bearish the middlemen.
Watchlist notes in the Monitor's house format: candidate, call, signals, entry logic, stop, target. Levels are anchored to prices reported in the trackers above. Event risk is flagged where earnings sit inside the trade window.
Signals: $51.6B of debt incl. leases against $5.5B of cash; 2032 bonds yielding >16% say credit markets are already pricing distress; Bernstein estimates ~$85B more needed to deliver the $104B backlog; anchor customer (Microsoft, ~67% of 2025 revenue) is publicly stepping back from renters; analysts cut targets on Nov 9 (Rosenblatt $250→$160, Wells Fargo $160→$115). Stock is −61% from its Oct 2025 high but bouncing +4.7% today into earnings.
Entry logic: No position ahead of the print — earnings Tue Nov 10, 5:00 PM ET can gap either way. Act only on a failed post-earnings bounce into the $54–58 breakdown zone, or a daily close back below $46.40 (the Nov 6 close).
Stop: Daily close above $62. Targets: $40.58 (52-week low), then the mid-$30s.
Signals: Construction spending running at ~14× revenue ($1.98B vs $137M in the June quarter) with $450M of impairments taken on the mining exit; Childress timeline pushed back under Microsoft oversight; the $4B ARR target slipped into next year; relative weakness — down −2.8% today while every peer bounces.
Entry logic: Fade rallies toward $24.90 (the Oct 23 close) that fail to hold, or a daily close below $21.64 (Nov 6 close) confirming the bounce is over.
Stop: Daily close above $27. Target: $17.16 (52-week low).
Signals: The hyperscaler with a neocloud's balance sheet: lost the Microsoft rental deal, cancelled the $7B Tencent lease, funding a $90–95B build with shares sold at an 11% discount and bonds at ~7.8%. Already Burry's best-performing short (−41.5% since his Jan disclosure — see Burry / Scion above), so this is a “press the winner” watch, not a fresh idea.
Entry logic: Only on relief rallies that stall below the $130s; the easy money on this leg is made and squeezes are violent.
Stop: Daily close above $140. Target: $100 round number, then reassess.
MSFT, AMZN, GOOGL and META are spending recklessly by any historical standard, but they fund it from the strongest cash flows in the market, their cloud growth is accelerating into the spend, and the post-midterm market is rewarding them for it. The bearish read in this space expresses through the leveraged renters, not the giants. If that changes — a guidance cut, not a vibe — this panel will say so.
−41.5% since Burry disclosed the short on Jan 9. From the ~$285 level where he likely put it on last October, Oracle is down roughly 60%. Premarket today: $114.64, another −1.26%.
“At this point Wall Street is AI.”
Starts from the Joint Economic Committee's official figure — $40.55 trillion as of Nov 4, 2026 — and ticks at the JEC's measured pace of $4.69 million per minute (~$78,000/sec). Up $2.46 trillion in a year: $118,587 per person · $300,840 per household. At the 3-year average pace, $41 trillion arrives around Jan 15, 2027. Average interest rate on the debt: 3.57%. Edition anchor — checking the live Treasury feed…
Voters decided 41 local data-center measures across 10 states — 40 of 41 went for tighter limits: bans, moratoriums, zoning rules. Ohio led with 21 measures. Counting earlier 2026 votes, 46 of 47 went against the industry. Results still unofficial.
Texas — the state with the most planned AI data-center capacity — elected Gina Hinojosa governor over Greg Abbott on a promise to pause new data-center construction. Georgia voters moved the same direction.
Democrats won both chambers campaigning on AI accountability, data-center moratoriums and Big Tech regulation. Anthropic has circulated a draft federal AI bill (preempting some state laws, harm thresholds of 50 deaths / $1B) drawing fire from safety advocates who back the bipartisan FRONTIER Act instead.
Clarksville, Arkansas votes Dec 1 on data-center bond measures (up to $50B and $5B). Mandan, North Dakota votes Dec 15 on banning data centers of 25 MW or more.
Over 40% of planned U.S. data-center capacity sits in states where officials have pledged or imposed pauses (Cleanview analysis) — a direct read-across for every AI infrastructure name on the tape above.
The odd fact of this selloff: the bubble's poster child, Nvidia, now trades at less than half its own 10-year average multiple — which is why Burry's “Cisco” comparison stings. Cisco in 2000 was a great company whose stock simply priced in a perfect decade.
AI Pulse Monitor is an information and market-data publication published by Vaivin LLC of Frisco, Texas. It is a publisher, not a financial advisor, investment adviser, or broker-dealer: everything here is general, impersonal, and published identically for all readers. It places no trades and manages no accounts.
Contact: [email protected]
The market panels (tape, index band, VIX, Fear & Greed, debt clock) refresh in place every 5 minutes — the page never reloads, so your place and any estimator inputs are kept. Each live panel is stamped with its data time. If a feed doesn't answer, the panel keeps its edition values and is labeled EDITION — nothing is ever filled in, estimated, or carried over silently. All editorial sections (news, trackers, analysis) are the dated edition shown in the masthead.
Each edition scores nine checks drawn from the sections on this page — sentiment, the tape, AI hype, leadership, Fed & macro, hyperscaler buildout, neocloud credit, Burry & valuations, and regulation — each leaning Buy or Sell, no neutrals; the majority is the day's word. Between editions, a live re-score runs in your browser from four live inputs (index momentum, the VIX, Fear & Greed, and breadth across the tracked names): a score of 50 or higher reads BUY, below 50 reads SELL, and the panel shows exactly which inputs were live and which fell back to edition values. The call's "as of" stamp is always the data's time, never your clock.