AI Capex Backlash

That was some savage selling on Friday.  Disappointing tech earnings met month end selling, vol control de-risking, and leveraged ETF rebalance selling.  These leveraged ETFs in volatile sectors like SOXL are absolute poison in choppy markets as they have to sell weakness and buy strength.  

Friday looks worse than it really is because the market rallied from Monday to Wednesday, right into all the tech earnings.  If you look at the daily chart, the Nasdaq chart looks bad, but the SPX is just chopping around in a narrow range, mostly between 7600 and 7800.  

The market has sold off into the FOMC meetings in July and September, and rallied strongly afterwards.  So you probably had some eager buyers jumping the gun, thinking the worst of the selling was over.  But tech is just too weak.  It trades like investors are overloaded in the sector and suffering badly, not wanting to triple down.  You have now seen 3 straight weeks of rallies from Monday to Wednesday being met with vicious selling on Thursday and Friday.  It feels like a combination of pre-weekend selling (ongoing Iran war fears) as well as weekly opex forces in a weak market.   

The earnings reactions in tech were bad.  The market is now screaming at the hyperscalers to stop going crazy over AI capex.  The stark difference between the earnings reactions for MSFT and GOOG tells you everything.  Investors are now punishing big AI capex plans and any disappointments from AI adoption (weaker search growth revenue).  GOOG went down hard on this, when it announced negative free cash flow (worse than expected) and huge AI capex spending plans for 2027.  MSFT is keeping its AI capex growth under control to maintain positive free cash flow, taking away dilution fears.  What is eye opening is that GOOG was already trading weak going into earnings, and MSFT was trading strong.  Even with the low expectations going into GOOG, it still cratered.  

GOOGL

MSFT

Then you had the negative earnings reactions in INTC and SNDK after their earnings.  There wasn't a big blow up in the earnings numbers for either company, just a little lighter guidance for INTC.  But the expectations for the semiconductor and AI hardware names are sky high.  Anything but monster  earnings beats are being met with disappointment.  

INTC

SNDK

A continuing source of pain will be tech equity supply for the next few months.  The float in SPCX continues to expand as lock up expirations gradually expire.  META dropped a dilution bomb right after its earnings, saying they will issue up to $95B in stock over the next few months, trying to get ahead of other hyperscalers trying to raise capital.  Anthropic filed its S-1 and likely to do its IPO in mid to late November.  That IPO could be for up to $100B.  With all of this tech equity supply coming down the pike, fund managers will be hesitant to add more tech to their portfolios.  

Retail trading activity has been down this year, as is normally the case when speculative breadth weakens.  Its been ongoing since the speculative small cap bubble hit its peak in October 2025.  There are lots of retail bagholders in speculative tech, waiting to sell at break even or until the pain becomes so great that they capitulate.  It shows you how crazy the retail speculation got in 2025.

Mag7 exposure is very high among hedge funds.  

QQQ short interest has been increasing why SPY short interest has been declining.  

Vol control funds are still holding high equity exposures because of the low realized vol in SPX over past few months.  

Aggregated equity positioning (still overweight) and across fund categories.

Cyclical - Defensive ETF flows have reached extreme negative levels.

5 year CDS for various hyperscalers and AI semiconductor companies

Funding spreads are a concurrent indicator of speculative index futures activity.  It usually peaks before the market does.  They are coming down after spiking in the summer.  Still at high end of range historically.  Not a bullish sign.  Similar drops in funding spreads seen in mid 2018, early 2025, and early 2026 led to serious weakness 2 to 4 months out.  

Not much going on in bonds, slowly grinding lower in yields as the AI trade weakens, even as crude oil keeps going higher ahead of Iran re-strike fears post midterm elections.  Treasuries are no longer an inflation trade, but the anti-AI trade.  

Covered the small short position on Friday, and entered into a small long position in SPX.  Not looking to get aggressively long, just looking for a technical bounce post-election, perhaps a 1-2 week rally.  Its been nearly 3 weeks since the hot CPI number launched the start of the selling in SPX.  We've had 3 waves of selling, each hitting a short term low on 3 consecutive Fridays.  This last wave was the swiftest and strongest.  It felt like some funds were throwing in the tech towel on Friday.   

Seasonally we are exiting the stock buyback period and entering a strong part of the year from end of October into year end.  Coinciding with mid term elections, a post event relief rally is the most likely scenario.  I don't expect a strong rally because there hasn't been a strong selloff (outside of tech), and tech has led the selling, which is a bad sign for the intermediate term.  The price action in October is a foreshadow of things to come, so rallies in SPX will be shortable if it gets above the top of the range (7800).  Its rare to see heavy selling in November and December in a bull market when you've already had a few weeks of selling in October.  So expect more chop until year end, and then 2027 should come out with some real fireworks to the downside.  

AI Bagholders

This week has been a continuation of last week, as the tech unwind continues, while the defensive and less crowded sectors stay strong.  Breadth has been relatively strong, yet the markets trade weak.  Another week that debunks the old-fashioned technician's belief that bad breadth is bad for the market, and good breadth is good for the market.  

The AI hype, which was at a fever pitch in May and June, has resulted in a monster bull trap, leading to big losses for those late to the AI trade.  I keep hearing the same thing from retail.  How can the SPX be only down 2% from all time highs while all these AI stocks are down anywhere from 30 to 60% from all time highs?  Its because the inflows keep flowing into equity index ETFs, keeping a bid for SPX, while retail and hedge fund flows into tech are drying up as AI equity supply ramps up, putting downward pressure on AI stocks.  

Some similarities to 2021 over the past few months.  The scale of the pain this time is much greater, as retail investing is now much more pervasive.  Just another warning sign of pain to come for 2027.

Here is a picture of how AI has done versus the SPX.  High beta momentum is basically the speculative names in the AI trade.  Down 41% from the late June highs, while SPX is down just 2% from its more recent October highs, and above the levels seen in late June.  

We have not seen this kind of underperformance among a group of high beta momentum names vs the SPX since 2000.  Even more perverse, has the been the strength in the defensive and less popular sectors of the market, while tech is getting killed.  

On Thursday, Oct. 22, the worst day of the week for the SPX, down 95 points, here were the sector breakdowns:

This kind of extreme dispersion among sectors on a big down day is rare, with SPX equal weight up 0.5% or more and NDX down 2.5% or more.  Precedents were mostly in 2000 and 2001.  

The market is giving you some very obvious hints that the peak of the AI bubble is behind us, and the consequences of the AI bust are in front of us.   The market internals for the AI bubble in 2026 is closely following the pattern of the dotcom bubble in 2000.

The hyperscalers and OpenAI/Anthropic continue to try to raise more capital despite the challenging environment for AI right now.  META dropped a little dilution bomb in their earnings filing, warning of possible huge amount of capital to be raised.  Zuck went all in on the metaverse and failed.  Is this his AI version of the metaverse gambit?  

OpenAI is now targeting a lower valuation to raise more private equity capital.  

We are getting close to the point where the capital markets will start closing off for giants in the AI space like OpenAI and Anthropic.  They have shut off the IPO window for the neoclouds.  Its only a matter of time before they shut it off for the bigger players when it becomes more clear that the economics for the AI business model is worse than advertised.  

There is growing AI fatigue as lots of subpar content, even things as short as tweets are being AI generated.  This is creating lots of internet pollution, as content quality degrades and AI generated information crowds out higher value human generated information.  Soon, these LLMs will be training on mostly AI generated data to create more AI generated data.  Its akin to a snake eating its tail.  

There is a lot of hype about AI making employees so productive, but a lot of that productivity is going towards bullshit production.  AI generated images, cookie cutter analysis, mediocre articles and papers, etc.  Most of this is low value, which means AI compute will need to be priced low (low to negative margins for OpenAI and Anthropic) to attract sufficient demand.  

There are gigantic compute commitments for 2027 and 2028 for the two LLM giants as data centers come on line.  You are looking at a potential credit disaster if the demand for the compute isn't sufficient, as I suspect will likely be the case.  That is a contrarian take because almost everything I hear is about the insatiable demand for compute from 15-minute AI experts like CNBC's Josh Brown.  

This bull market started in October 2022, has taken the SPX from 3491 on October 13, 2022, to 7844 on October 6, 2026.  That's a 124% return plus dividends in 4 years.  That's a huge move for an economy that's growing this slowly.  In the next bear market, investors may take comfort in seeing Treasury bonds rally while stocks fall, thinking that there won't be a repeat of 2022.  That's right, the next bear market won't be like 2022.  It will be a stock focused bear market, not a bond focused bear market.  There won't be much comfort in seeing their much smaller bond portfolio rally a few percent while their much bigger equity portfolio goes down 30-50 percent.   

BofA client flows for the weekend ending October 16.  Institutions have been heavy sellers in October.  

BofA clients were heavy sellers of tech for the week ending October 16.  Need to see more of this to get a short term bottom.

Retail is still buying the dip in semiconductors, as of Oct. 21.

Hedge funds have been driving the tech led selloff.  Can't tell anymore if hedge funds are the smart money or the dumb money.  They used to be the dumb money when retail only had a small footprint in the market.  Now that retail has gotten so big in the stock market, taking the mantle of dumb money, hedge funds are now a trail mix of both smart and dumb money.  Not as much signal to be gained from hedge fund positioning as in the past.  

GS fundamental long/short hedge funds at very low net leverage levels.   

Bonds continue to squeeze higher as CTAs cover shorts, and long tech/short Treasury bond pair trades get unwound.  The bond market has finally found a bull catalyst(the AI bust) which is actually a bear catalyst for stocks.  AI is the elephant in the room, sucking out all the oxygen, putting sticky inflation and the Fed hiking cycle into the closet where no one is paying attention.  I believe the past 2 weeks are a preview of next year.  AI will be the dominant driver for both stocks AND bonds in 2027.    

You still have not seen any signs of capitulation in the dark pool data. The 2 significant bottoms in 2026, in late March and late July had big drops in the DIX, getting down towards 40%.  Aside from one day (Oct. 15), DIX has been fairly high during this tech selloff.  

Nothing notable in the COT data as of Oct. 20.  Speculators remain heavily long NDX futures.  

There is speculation that there will be a restart of US attacks in Iran after the midterm elections, providing a strong bid to crude.  I continue to believe that the ramifications of the war are being overstated, as the Fed is just not that important anymore, and higher inflation is only a negative if the Fed tightens financial conditions because of it.  This Fed board has clearly shown that it doesn't have the balls, or the will to hurt the stock market to try to lower inflation.  So any equity selling resulting from a further rise in crude is probably a buying opportunity.  

The market had quite a strong bounce (stronger than expected) from Monday to Wednesday, shaking the convictions of some bears.  Having a smaller position makes it much easier to hang on to positions and withstand some heat.  Wednesday turned out to be a huge head fake.  The sudden enthusiasm over SPCX's revenue numbers and a possible SPCX/TSLA merger was enough to set the bull trap.   Thursday took back 3 days of gains in just a few hours.  Now that the market is closer to SPX 7600, there isn't as much downside to play for.  Covered half on Friday, and will the rest ride until Monday/Tuesday.  

Do not want to be short after FOMC on Wednesday, and the slew of tech earnings coming out Wednesday/Thursday of next week.  The bearish news flow for AI is getting heavy and the seasonal calendar is turning more bullish.  Base case is that the we make a bottom sometime next week, and then bounce.

Sneak Preview of 2027

Its getting real.  They hit the heart of the US stock market this week.  AI is the heart of both the stock market and the economy.  The financialization of the US has made the stock market the de facto economy.  That is why you saw the bid in Treasuries, despite the hot CPI and PPI numbers.  When investors sense weakness in the AI boom, it doesn't take a genius to connect it to the economy, and thus bond yields.  Going forward, expect a reversal of the positive correlation between stocks and bonds.  The 2022-2023 playbook is useless for this market.  It will lead investors astray during the AI bust.  We're in the 2000-2002 playbook, where tech stock weakness leads to bond market strength.  

Credit spreads for AI data center stocks are blowing out and investors are starting to question the sustainability of the AI capex boom.  Last month, Anthropic delayed its IPO, which was initially planned for October into November/December.  Firmus, an AI data center company, scrapped its $5B IPO plans this month, when there wasn't demand at a $30B valuation.  

The hot topic of the weak, CRWV credit spreads, are blowing out.  

AI data center bonds steadily weakening since July.

The hyperscalers have gone into negative free cash flow at current spend levels.  In order to have further growth from current capex, additional capital will need to come from equity/bond investors.  It appears that even at these higher yields, there just isn't as much demand as most expected from bond investors.  As for equity capital, if you push valuations to nosebleed levels, IPO buyers think twice about whether they can even flip  for a profit.  That's why you are seeing AI IPOs getting delayed.  After seeing how the SPCX IPO played out, people are realizing the upside is limited when you IPO at ridiculous levels.  There aren't enough speculators out there to chase these AI stocks at crazy valuations.  Those gamblers have already gorged themselves on AI names, and have little room to add more bags to their portfolio.  

Musk has front run the crowd brilliantly by taking up a huge chunk of equity capital before the other big AI players were ready to come to the money trough.  Now, Anthropic and Open AI will probably have to IPO at lower valuations than they want to, in order to attract enough initial bids.  

The parabolic phase of the AI bubble has ended.  The NDX is now 4 months into the choppy, messy, and sometimes violent sideways range that form a long term top.  For the SPX, it has been 2 months since the initial top just above SPX 7800.  The length of the choppy, range bound phase varies for each bull market.  The following are some bull market top charts for reference.  

2000:  Initial top in late March.  Range bound phase (5 months) until final top in early September. 

SPX 2000 Top

2007:  Initial top in mid July.  Range bound phase (3 months) until final top in mid October.  

SPX 2007 Top

2015:  Initial top in late November 2014.  Range bound phase with only marginal higher highs until final top in late July (8 months)

SPX 2015 Top

2018:  Initial top in late January.  Range bound phase with margin higher highs until final top in early October (8 months)

SPX 2018 Top

2022:  Initial top in early November 2021.  Range bound phase (5 months) with marginal higher highs until final lower high top in late March.

SPX 2022 Top

2026:  Initial top in August 2026? Range bound phase (2+months) until final top in ?

SPX 2026 YTD

YTD inflows into tech funds $142B, with 2.5 months left in the year.  

The market has been very complacent about the AI boom.  Huge growth for the next few years is viewed as a sure thing.  It shows up in the money flows, as you have seen enormous inflows into tech sector funds.  It wasn't like this in 2025.  In 2025, there were still lots of doubt about AI end demand and monetization.  The memory mania was just getting started.  Now, every fast money investor and speculator is bulled up on semiconductors and DRAM.   It is the worst possible setup for bulls, as a heavily saturated market with high expectations faces huge disappointments.  

This past week is a sneak preview for the main event in 2027.  2027 should finally usher in an extended bear market coming from the AI bust, as OpenAI and Anthropic have trouble paying off their debt obligations as it grows with each AI data center launch (many more scheduled in 2027 and 2028 compared to 2026).  That should percolate down the chain to the semiconductors, AI hardware and power providers.

GS Prime Book net exposure to semiconductors is still very high.  Net exposure is still more than double the levels seen for 2024-2025.  

Previous scares such as the yen carry trade in August 2024, the tariff fears in February-April 2025, and the Iran war in March 2026 were tangential to the main theme.  The main theme being AI.  This is the first time since the start of the bull market where the scare is coming from the heart of the market.  Open AI expected revenues lowered, CXMT tech breakthrough news in the HBM market, and Coreweave and AI data center credit spread blowing out.  This is not something that can be just shaken off and ignored.  Too many AI speculators are losing money.  They've been bleeding heavily since June, and its just gotten worse.  There is a ton of overhead supply waiting, as AI bagholders will look to unload on strength.

Despite the ominous signals coming from the market, Its likely that you will get a bottom sometime in late October.  But don't expect a big, extended bounce like you saw coming after the late March bottom.  I would expect the next rebound to last less than a month, with marginal highs at best for the NDX.  

Just based on the calendar, how late it is in 2026, I don't expect any big moves to the downside until early 2027.  The market is still up over 10% on the year, with 2 months left in 2026.  Money managers don't have urgency to sell off their equity holdings at year end, which would bring foward capital gains taxes by one year.  So its rare to see a market go down hard into year end when its up that much in mid October.  You did see a rare 4th quarter selloff in 2018 which crushed the market, but the market was never up more than 10% on the year, and spent most of the year +/- 5%.  There will likely be a golden opportunity to put on big short positions with high conviction in late December, to front run the delayed capital gains selling, as well as the coming AI bust which should begin to play out in earnest in the first half of 2027.

CBOE SKEW spiking as the market is falling.  Uncommon.  Usually SKEW spikes during rallies, not selloffs.  Last time you saw this was in early December 2024.  Market went into a choppy sideways range for 2 months before a big plunge starting on February 2025 opex.  SKEW is now at levels last seen in early 2025. 

 

VIX move relative to SPX was off the charts for Friday.  Reflected in the high SKEW reading, as demand for tail hedges spiked.  This is no ordinary selloff.  

Big volume coming in for VIX call options expiring on October 21 at the cash open.  Could keep VIX stable until they expire on Wednesday, then could see another wave higher in VIX afterwards if SPX selloffs later in the week.

The market has been quite underhedged since August.  Investors decided that SPX puts were unnecessary, and prices for near term OTM puts were taken down to very cheap levels.  The systematic options sellers got more aggressive, and they helped to contribute to lower realized vol.  It looks like that period of rampant put selling is behind us.  The VIX has finally woken up, and abnormally strong VIX is a short term leading indicator for SPX weakness.  I expect this VIX strength relative to SPX movement to persist until the week of the FOMC meeting and majority of Mag 7 earnings, from Oct. 26 to Oct. 30.  

You saw some interesting moves in the bond market this past week.  It caught me off guard, as I expected a hotter CPI number to keep bond yields higher.  But the ferocious selloff in AI rendered the CPI number as meaningless, and the Fed funds futures are now only pricing in about a 15% chance of an October hike.  It was as high as 60% odds right after the CPI number came out.  Much of the move higher in Treasuries, especially the long end, felt like short covering as CTAs are heavily short.  Its becoming clear that Treasuries (in particular, short to intermediate term) are now likely to be an inverse Nasdaq trade for the next couple of years.  It will catch some bond traders off guard who are still focused on inflation, instead of AI.  Expecting a monster bull steepening move in the yield curve for 2027.

Put on shorts in both stocks and bonds in the middle of the week.  Took the loss on bonds but holding short in SPX.  Should have shorted NDX but was a bit chicken little as I didn't have a premium entry.  Looking for initial downside target of SPX 7600, and then 7550 if downside momentum picks up.  Not looking for a big move lower, so playing smaller size.  The risk/reward for shorts should be much better in 2027, when it will be time to be more aggressive with positions and entries.  

Listless

The word that I hear a lot these days is resilient.  On CNBC, they are in awe of how resilient the stock market is.  They act like its a miracle that stocks are going up as yields continue to go higher.  Many are forgetting that yields have been going up since March along with the SPX.   Stocks can go up for a long time while bond yields are going higher.  Watching the bond market doesn't really help much when predicting the stock market.  

Then you have the breadth boys.  The same old technicians on Twitter complaining about bad breadth.  Those guys still think that strong breadth is the cure all for the world's ills.  Never mind that a broadening market with good breadth in January and February preceded the worst month of the year.  Same thing happened in July 2024 as the market broadened out, right before you had the worst 2 weeks of the year.  In the stock market, they never let facts get in the way of a good story.  

Rather than resilience, the more apt word to describe the market is listless.  It should feel more exciting than this.  If AI is as transformational as all the pumpers out there are saying, why does this market seem so boring?  Even a lot of the AI names feel waterlogged.  NDX is still in a long term uptrend, but now its going up in drips and drabs, you aren't seeing the parabolic frolic that you saw for most of 2024, second half of 2025, and in Q2 of 2026.  You get a couple of up days, everything looks great, and then you get an out of the blue gap down that takes back all the previous day's gains, just like you saw on Wednesday.  

The market feels saturated.  The chasers don't have to chase anymore to get their allocation of the hottest names.  They can wait a few days and they get their pullback, and can buy cheaper.  Why would anyone in this market want to chase strength when they can wait a few days and buy on weakness at lower prices?    When uptrends flatten out and you get sideways chop, chasers get punished, and regret buying into strength.  Those who are more patient are getting filled quicker, and don't feel the need to chase anymore, because pullbacks, while being shallow, are more frequent.  It reinforces the lack of momentum in this market.

A lot of money has poured into US stocks over the past 3 years.  In particular, the last 12 months.  Expecting these kind of equity inflows to continue is wildly optimistic.  We've had historic levels of money pouring in.  It seems like everyone has a big allocation to US stocks, yet many say that its the most hated bull market ever.  Its like Yogi Berra said: no one goes there anymore, its too crowded.  

We are in that awkward period where stocks are still in an uptrend, but the trend is weakening, and the majority are heavily allocated to the best performing names, which are outperforming, so no one is really worried.  They talk about bond yields and how they could hurt stocks, but I still hear a lot of those CNBC experts who think that the Fed could just hike one more time and be finished with this rate hiking "cycle".  There is still plenty of denial out there.  Many are still skeptical that the Fed will continue hiking rates beyond December into 2027.  The STIRs market is pricing in 4 more hikes for the next 12 months.  I'll pick the market prediction over a CNBC talking head any day.  

All this talk about the bond market is a distraction from the real mover, AI.  All it took was an FT rumor about OpenAI revenue estimates being $20B less than forecast to trigger a quick intraday downdraft.  After the dust has settled, all the NDX outperformance over SPX since September 21 has been erased in a span of 2 days.  That is meaningful.  The NDX strength made me hesitant to short this market.  Now that NDX is no longer outperforming, it makes me want to short the next bounce that comes.  

BofA client stock, bond, and cash allocations, as of October 1.

BofA client flows for the week ending Oct. 02.  Institutions were heavy sellers, while hedge funds were buyers.  

Schwab Trader Activity Index for September.  Remain near highs over the past 3 years.  

SpaceX saw net flows 50% higher than the second-ranked stock, with buying peaking in the final week of the month. The size and timing of the flows suggest enthusiasm remained strong rather than fading as the month progressed.

Post opex Octobers (starting October 16) have been historically weak.  

They continue to buy the dip in the long end.  

Mag 7 net exposure at 5 year highs, MAGS ETF monthly inflows at all time highs.

Heavy inflows into government bonds last 2 weeks.  

Lot of speculative call buying in TLT.  

Rising global bond yields are a symptom of a lack of liquidity at current prices for stocks and bonds.  When there is plenty of liquidity in the system, you get both stocks and bonds rising.  When there is a liquidity constraint, it is difficult for both stocks and bonds to go up together.  In the short to intermediate term, stock-bond correlation is mixed.  In the long term, bond market weakness is often a leading indicator for stock market weakness.  

The big event for the coming week is the CPI on October 14, and opex on October 16.  I expect a grind higher into the CPI number, and then a pullback afterwards as we get to the 2nd half of month.  Still on the sidelines waiting for a bit more of a rally to put on shorts, in both stocks and bonds.  This time, the time frame for the short trade will be less than 2 weeks.  It still feels too early to put on long term shorts, as we likely get a post election relief rally in November, even with a Democrat sweep of Congress.