Week of 2026-10-05
A record-filled but whippy week ended higher: the S&P 500 and Nasdaq both printed fresh records early, then a midweek AI shock knocked them back before a Friday rebound (SPYSPY$759.97close of 2026-10-23 778.57, QQQQQQ$704.40close of 2026-10-23 a marginal record 751.27; the Dow and S&P led, the Russell 2000 lagged at 278.94 under the weight of a long end at multi-decade highs). The shock was a Financial Times report that OpenAI's annualized revenue is running near $50B - well below the ~$70B widely assumed - which lopped ~$169B off NvidiaNVDA$210.96▲ +0.00% past 30 sessionsclose of 2026-10-23 in a session and sent OracleORCL$126.16▼ -12.87% past 30 sessionsclose of 2026-10-23 down more than 5%, reviving 'circular-financing' worries about the AI build-out. Away from tech the week broadened: nine of eleven sectors rose, led by utilities (GoogleGOOGL$320.21▼ -8.35% past 30 sessionsclose of 2026-10-23 signed a 20-year nuclear deal with Constellation) and energy, gold pushed to new records (GLDGLD$366.61▼ -6.68% past 30 sessionsclose of 2026-10-23 384.58) with silver surging, and the equal-weight S&P broke a seven-week losing streak. The 10Y metric prints 5.26% after touching a 24-year high of 5.35% midweek and easing on a strong 30-year auction; Bitcoin slipped to $83,386 as spot-ETF flows turned to ~$679M of weekly outflows. (The WTI print of $91.18 is unchanged from last week and looks stale - crude actually swung on US-Iran headlines; treat it with caution.)
Macro & Markets
It was a record-filled week that didn't feel like one. Both the S&P 500 and the Nasdaq printed fresh highs early before a midweek AI shock dragged them back, and a Friday rebound carried the major averages to modest weekly gains - the S&P up ~1.2% (SPYSPY$759.97close of 2026-10-23 778.57), the Nasdaq ~0.6% (QQQQQQ$704.40close of 2026-10-23 a marginal record 751.27), and the Dow ~0.9%, while the rate-sensitive Russell 2000 lagged and slipped to an IWMIWM$275.75▼ -4.22% past 30 sessionsclose of 2026-10-23 print of 278.94. The jolt came Wednesday-into-Thursday, when a report that OpenAI's annualized revenue is running near $50B - far below the ~$70B widely assumed - sent chip and AI-infrastructure names sharply lower, the Nasdaq's worst session since mid-August. The real macro story sat in the bond market: the 10-year yield touched a 24-year high around 5.35% midweek before a well-received 30-year auction pulled it back into the 5.2s (the metric prints 5.26%), even as traders trimmed the odds of an October Fed hike to under ~20% ahead of the Oct. 28 meeting. Crucially, the week broadened: nine of eleven sectors advanced, led by utilities and energy rather than semis, and the equal-weight S&P 500 rose ~1.6% to snap a seven-week losing streak - the kind of participation that has been missing all autumn. Oil swung on US-Iran headlines (the WTI metric of $91.18 is unchanged from last week and looks stale), and gold pushed to new records with the GLDGLD$366.61▼ -6.68% past 30 sessionsclose of 2026-10-23 metric at 384.58.
Sources: Wall Street finishes its record-setting week with gains · Nasdaq, SPX Look to End Record-Filled Week With Gains · US 10-Year Treasury Yield (24-year high; 30Y auction) - Trading Economics · Dispersion Brief: Week in Review, Oct 5-9 (sectors, equal-weight)
Emerging Tech Radar
The week's defining emerging-tech story was not a chip but a power contract. Google (GOOGL)GOOGL$320.21▼ -8.35% past 30 sessionsclose of 2026-10-23 signed a 20-year deal to buy 890 MW of newly uprated nuclear output from Constellation Energy (CEG)CEG$253.45▼ -4.20% past 30 sessionsclose of 2026-10-23, alongside a separate 15-year agreement for 2,700 MW from Constellation's existing fleet, with Constellation committing more than $4.3B to modernize 11 reactors across six sites in Illinois, Pennsylvania and New Jersey - the largest tech-nuclear 'uprate' pact on record, feeding the PJM grid, with the first upgraded unit expected online in 2028. The signal matters beyond one deal: with AI data centers now the marginal buyer of electricity, the frontier technology of the moment is as much about firm, carbon-free megawatts as it is about silicon, and hyperscalers are willing to underwrite reactor upgrades to secure them. That pulls a second wave into view - the advanced small-modular-reactor (SMRLoading SMR…) developers building the next generation of supply, a still-pre-revenue cohort that the week's economics made newly credible. The same demand is why the physical layer of AI - power, cooling and grid interconnection - is drawing capital as fast as the models themselves.
Sources: Google signs 20-year nuclear deal with Constellation Energy to power AI build-out · Google goes nuclear: 890 MW from Constellation (deal terms) · Google Secures 890 MW Of Nuclear Power from Constellation Energy
Startup Watch
Private capital kept pouring into fintech and AI even as public markets wobbled. Ramp, the corporate-spend startup, reached roughly a $60B valuation in a ~$1.85B round led by Dragoneer and Thrive Capital - a striking mark for a company that was valued a fraction of that two years ago, and a sign that the private market's appetite for AI-adjacent fintech hasn't cooled. In China, Manus (the AI-agent company whose parent is Butterfly Effect) raised more than $500M in its first round since splitting from MetaMETA$642.62▼ -3.45% past 30 sessionsclose of 2026-10-23, led by Boyu Capital and IDG with Tencent participating, underscoring that the agent race is well funded on both sides of the Pacific. Crunchbase's tally of the week's biggest rounds skewed heavily toward AI infrastructure, space and cybersecurity - including AI-security startup Armadin's ~$255.5M Series B led by Andreessen Horowitz and Accel. The throughline: the funding is concentrating in the picks-and-shovels of AI (compute, security, agents) rather than in consumer apps, the same layer the public market is now pressure-testing on revenue.
Sources: Ramp Hits $60 Billion Valuation in $1.85 Billion Funding Round · China's Manus raises over $500M in first funding round since split with Meta · The Week's 10 Biggest Funding Rounds: AI infrastructure, space tech lead
Sector Tailwinds & Headwinds
Leadership flipped from growth to the real economy this week. Utilities were the standout - XLU jumped roughly 4%, powered by the GoogleGOOGL$320.21▼ -8.35% past 30 sessionsclose of 2026-10-23-Constellation nuclear pact and a parallel wave of power-demand deals, as the group most levered to AI's electricity appetite finally got paid for it. Energy was close behind (XLEXLE$69.33▲ +7.44% past 30 sessionsclose of 2026-10-23 up ~3.6% and the top sector on Thursday) as oil firmed on US-Iran headlines and majors eyed Middle East supply. Materials and precious metals had the wind at their back too: gold set records (December futures near $4,216/oz) and silver surged ~2.7% in a single session toward $61, lifting miners. The clear headwind was technology, the week's weakest sector (down ~0.5%), as the OpenAI revenue report knocked the AI-infrastructure complex. A second headwind blew through anything rate-sensitive: with the 10-year at a 24-year high intraweek, small caps (the Russell 2000) and other borrowing-dependent groups lagged even as the broad market rose. The tell for the week was the equal-weight S&P breaking a seven-week losing streak - breadth finally doing the heavy lifting.
Sources: Dispersion Brief: Week in Review, Oct 5-9 (sector winners and losers) · Market Review: October 09, 2026 (gold/silver settlements) · Google goes nuclear: 890 MW from Constellation (utilities catalyst)
Disruption Watch
The disruption this week was aimed inward, at the AI build-out's own financing. A Financial Times report that OpenAI's annualized revenue is nearing ~$50B - versus the ~$70B figure that had circulated (a gap that appears largely definitional, gross versus net) - cracked the assumption underpinning a trillion dollars of infrastructure commitments. Nvidia (NVDA)NVDA$210.96▲ +0.00% past 30 sessionsclose of 2026-10-23 shed roughly $169B of market value in a session, Micron (MU)MU$877.55▼ -5.03% past 30 sessionsclose of 2026-10-23 fell with it, and Oracle (ORCL)ORCL$126.16▼ -12.87% past 30 sessionsclose of 2026-10-23 dropped more than 5% because its enormous cloud backlog is concentrated in that single customer. The episode revived the 'circular-financing' critique - that the same handful of companies are each other's suppliers, customers and investors, so demand can look larger than the end revenue justifies. Nothing was actually canceled, and NvidiaNVDA$210.96▲ +0.00% past 30 sessionsclose of 2026-10-23's next earnings print is the real test of whether this was a scare or a signal. A quieter, more durable disruption ran through power markets: when hyperscalers start funding nuclear reactor upgrades directly, the traditional utility-ratepayer model of who pays to build generation is being rewritten in real time.
Sources: OpenAI's Revenue Is Reportedly $20 Billion Lower Than Thought. Nvidia Just Lost $170 Billion · Oracle Stock Crashes After OpenAI's Revenue Gap · Nvidia drops nearly 5%, leading chip stocks lower amid renewed worries of circular financing · Micron, Nvidia shares fall after OpenAI revenue report raises concerns
Crypto
The 'uptober' rally stalled, and the tape was set by fund flows rather than fundamentals. Bitcoin began the week above $86K - with traders pricing out an October Fed hike - but slid to the $83,386 metric print as US spot-Bitcoin ETFs swung to roughly $679M of net outflows, a three-month high; the worst day was Wednesday (~$485M out, led by Fidelity's FBTC), with only a small inflow returning Friday. Ether ETFs fared worse in proportion, shedding about $542M, nearly all from BlackRockLoading BLK…'s ETHA. The reversal is stark against late September's ~$2.4B inflow week, and it maps onto the same macro backdrop pressuring small caps: a long end at 24-year highs pulling risk appetite out of the most speculative corners. With crypto legislation quiet and no major institutional catalyst this week, price is being set almost entirely by the ETF bid - and this week that bid went the wrong way. Sentiment readings turned mixed-to-fragile, with at least one tracker flagging that the risk-on rally is 'splintering' beneath a still-elevated greed reading.
Sources: Bitcoin ETFs Record $679M in Weekly Outflows Amid Uptober Slump · Bitcoin ETF Weekly Outflows Hit 3-Month High, Ethereum Funds Extend Losing Streak · Bitcoin above $86,000 as traders price out an October Fed hike · Cracks Beneath the Greed: Why Crypto's Risk-On Rally Is Splintering
Hype Watch
Start with the mood. Risk appetite improved over the week but stayed cautious: CNN's Fear & Greed Index climbed off the low-40s to around 45 - the border of neutral - yet its market-breadth component sat in 'extreme fear,' a split that says the index highs are masking a nervous tape underneath. A separate stock-only gauge read closer to 66 ('greed'), so the signal depends on what you measure. The crowded trade that got tested was, predictably, the AI-infrastructure complex: NvidiaNVDA$210.96▲ +0.00% past 30 sessionsclose of 2026-10-23, OracleORCL$126.16▼ -12.87% past 30 sessionsclose of 2026-10-23 and the OpenAI orbit, where a single revenue headline erased hundreds of billions in minutes - a reminder that the most-owned story is also the most fragile to any crack in its assumptions. The week's fresh hype gathered around nuclear-and-power names, which ripped on the GoogleGOOGL$320.21▼ -8.35% past 30 sessionsclose of 2026-10-23-Constellation deal; here the fundamental substance is real (the demand and the contracts are tangible), but the move has already priced a lot of future build-out, and the pre-revenue SMRLoading SMR… developers riding the theme are running on narrative more than numbers. Net read: sentiment is late-cycle - resilient earnings and secular AI spending propping up valuations against restrictive financial conditions, with the crowd leaning on a handful of the same names.
Sources: Fear & Greed Index moves into neutral zone as sentiment improves · CNN Fear & Greed climbs to 43 but 'fear' still dominates breadth · Week #41 Market Update for October 5-9, 2026 (late-cycle setup) · Nvidia drops nearly 5% amid renewed worries of circular financing
Fibonacci Trend Check
Technical observation, not investment advice. All three majors are testing very different parts of their ranges. The S&P 500 (SPYSPY$759.97close of 2026-10-23) printed a fresh record near $781 before closing the week at 778.57, so it sits above every retracement of its spring-low-to-record swing; the 23.6% line at $745 is the first floor that matters, and only a break of it would open the 38.2% at $723 - i.e. the AI-capex scare would have to deepen into a trend before the chart turns. The Nasdaq 100 (QQQQQQ$704.40close of 2026-10-23) looks the same on the surface - a marginal record at 751.27 - but it took the midweek AI hit hardest, so its 23.6% support at $706 is the level to watch; losing it would separate a pause in the AI trade from a genuine de-rating toward the 38.2% at $677. Bitcoin is the outlier: at $83,386 it is sitting right on the 61.8% retracement (~$82,832) of its entire $59K-to-$121K advance - the last Fibonacci floor of the bull structure. Holding it and reclaiming the 50% line near $90,120 would repair the chart; a weekly close below $82,832 would turn the pullback into a round-trip toward the spring low. Levels shown are drawn from each symbol's own swing and are where support and resistance now sit, not forecasts.
▲ Bull case Holds the $781 record and closes above it → nothing above but blue sky - the measured move extends and the record-week momentum carries, with the 23.6% line at $745 as the trailing floor. The AI-capex scare reads as a one-day wobble rather than a top.
▼ Bear case Loses $745 (23.6%) → the first real pullback since the spring, with the 38.2% line at $723 the next test - the level that would mark the AI-capex-sustainability worry graduating from a scare into a trend.
▪ Base case Chops between $745 and $781 → a healthy digestion of the record run while leadership rotates from tech into utilities, energy and metals - breadth doing the work the index highs no longer need from semis.
▲ Bull case Reclaims and holds the $752 record → the midweek AI shock is fully unwound and the index presses to new highs, with $706 (23.6%) as the shallow floor. The OpenAI-revenue scare confirmed as noise once Nvidia's print lands in range.
▼ Bear case Loses $706 (23.6%) → the deepest tech-led pullback of the autumn opens toward the 38.2% line at $677 - the level that separates a pause in the AI trade from a genuine de-rating of the capex names.
▪ Base case Holds $706 on a retest → the record high stands as resistance while the complex consolidates; a higher low above the 23.6% line keeps the uptrend structurally intact even if momentum cools.
▲ Bull case Holds $82,832 (61.8%) and reclaims $90,120 (50%) → the deepest retracement of the entire up-move holds as support and the 50% line comes back into play - ETF outflows exhausting and the 'uptober' seasonal reasserting.
▼ Bear case Loses $82,832 (61.8%) → the last Fibonacci floor of the bull structure gives way and the move becomes a full round-trip toward the $59K spring low - flow-driven selling overwhelming the seasonal bid.
▪ Base case Chops around the 61.8% line → price pins to the $83K area while spot-ETF flows decide the next leg; a close back above $90K would repair the chart, a weekly close below $82.8K would break it.
Sources: Nasdaq, SPX Look to End Record-Filled Week With Gains (index levels) · Bitcoin ETFs Lost $731 Million in Two Days, Then Turned Positive (BTC price path)
Stocks to Watch
Conviction book derived from this week's report - not investment advice. This week's developments put four names in play across four sections, and the book leans into the shift away from the crowded AI-infrastructure trade. The highest-conviction long is Constellation Energy (CEG)CEG$253.45▼ -4.20% past 30 sessionsclose of 2026-10-23, from Sector Tailwinds & Headwinds and Emerging Tech Radar: GoogleGOOGL$320.21▼ -8.35% past 30 sessionsclose of 2026-10-23's 20-year, 890 MW nuclear-uprate deal (plus a 2,700 MW supply pact and $4.3B of reactor modernization) makes Constellation the clearest way to own AI's new binding constraint - power - through existing assets. The second long, Pan American Silver (PAAS)PAAS$40.15▼ -17.45% past 30 sessionsclose of 2026-10-23, comes from Macro & Markets: with gold at records and silver surging ~2.7% toward $61, a mid-cap primary silver miner gives levered exposure to the debasement bid that the GLDGLD$366.61▼ -6.68% past 30 sessionsclose of 2026-10-23 metric only partly captures, and it is the book's non-mega-cap leg. A smaller, speculative long in Oklo (OKLO)OKLO$26.35▼ -27.23% past 30 sessionsclose of 2026-10-23 - flagged as a call option, not a core holding - extends the Emerging Tech read-through to the advanced-reactor developers the week's economics made more credible. On the short side, Oracle (ORCL)ORCL$126.16▼ -12.87% past 30 sessionsclose of 2026-10-23, from Disruption Watch and Hype Watch: the OpenAI revenue report hit OracleORCL$126.16▼ -12.87% past 30 sessionsclose of 2026-10-23 hardest of the mega-caps because its cloud backlog is so concentrated in that one customer, making it the cleanest expression of the AI-capex-sustainability question. The startup-derived idea worth watching but not sized here is the private AI-security wave (Armadin's ~$255.5M round), which has no clean mid-cap public proxy this week.
How I'd invest $1,000 this week
Conviction-weighted, derived from this week's report — not investment advice. Long and short books each draw from a $1,000 budget. Every book is marked to market on the portfolio page.
From Sector Tailwinds & Headwinds and Emerging Tech Radar: the week's clearest catalyst was GoogleGOOGL$320.21▼ -8.35% past 30 sessionsclose of 2026-10-23's 20-year deal to buy 890 MW of uprated nuclear power from Constellation plus a 15-year, 2,700 MW supply pact, with Constellation investing $4.3B to modernize 11 reactors - the largest tech-nuclear uprate yet. Utilities led all sectors on it. Highest-conviction long: it monetizes AI's new binding constraint (power) through existing assets rather than speculative new build.
From Macro & Markets: gold set fresh records and silver jumped ~2.7% in a day toward $61 as the debasement bid broadened into the whole precious-metals complex. A mid-cap primary silver miner gives levered exposure to that move that the GLDGLD$366.61▼ -6.68% past 30 sessionsclose of 2026-10-23 metric itself doesn't - the non-mega-cap leg of the book, and the one that benefits most if silver keeps outrunning gold.
From Emerging Tech Radar: the GoogleGOOGL$320.21▼ -8.35% past 30 sessionsclose of 2026-10-23-Constellation deal confirmed that AI data centers will pay up for firm, carbon-free power, and the read-through runs down to the advanced-reactor developers building the next generation of supply. A speculative small-cap with no revenue yet - sized small and flagged as speculation - held as the book's call option on the SMRLoading SMR… theme the week surfaced.
From Disruption Watch and Hype Watch: the OpenAI revenue report (~$50B annualized vs. the ~$70B assumed) hit OracleORCL$126.16▼ -12.87% past 30 sessionsclose of 2026-10-23 hardest of the mega-caps because its multi-hundred-billion-dollar cloud backlog leans so heavily on one customer. The paired short against the AI-capex-sustainability question - if the revenue underneath the build-out is smaller than priced, the name most exposed to that single contract has the most multiple to give back.
Sources: Google signs 20-year nuclear deal with Constellation Energy · Market Review: October 09, 2026 (gold/silver settlements) · Oracle Stock Crashes After OpenAI's Revenue Gap · The Week's 10 Biggest Funding Rounds: AI cyber (Armadin) and more
Something Few People Know
When GoogleGOOGL$320.21▼ -8.35% past 30 sessionsclose of 2026-10-23 agreed this week to buy 890 MW of 'uprated' nuclear power from Constellation, it was tapping one of the quietest success stories in US energy - a way the country has added reactors' worth of electricity without building a single new reactor. Through a regulatory process called a power uprate, the Nuclear Regulatory Commission lets operators squeeze more output from existing plants - by replacing steam turbines, improving fuel, or simply measuring coolant flow more precisely (a 'measurement-uncertainty recapture' uprate can unlock ~1-2% more power just from better instruments). As of January 2022 the NRC had approved 171 uprates totaling roughly 8,030 megawatts-electric - the agency's own estimate is that this is equivalent to adding about eight new reactors to the fleet, achieved incrementally and almost invisibly over four decades. That is why the GoogleGOOGL$320.21▼ -8.35% past 30 sessionsclose of 2026-10-23-Constellation deal is structured around modernizing 11 existing reactors rather than breaking ground on new ones: the fastest firm, carbon-free megawatts available to power the AI build-out aren't new plants at all - they're hiding inside the ones already running.
Sources: NRC Backgrounder on Power Uprates for Nuclear Plants (171 uprates, ~8,030 MWe, ~8 reactors) · Google goes nuclear: 890 MW from Constellation (11-reactor modernization)