What Would Stop The AI Sell-Off

Introduction

The AI trade got a reality check this past week, with the VanEck Semiconductor ETF (SMH) dropping close to 10%, as capital rotated out of tech into sectors such as Financials, Energy and more. This is a possibility I predicted could happen in both my macro posts earlier this month.

 

Following the OpenAI revenue controversy, Hock Tan’s cautionary warnings and CXMT’s architectural breakthroughs, sentiment had already soured. Anthropic’s Q3 numbers worsened matters further, as the revenue growth rate slowed sequentially from Q2.

This in turn coincided with a Original Postress-releases/2026-10-14-gartner-says-over-us-dollars-three-and-a-half-trillion-of-artificial-intelligence-infrastructure-spending-could-be-exposed-to-impairment-by-the-end-of-2030" rel="nofollow noopener">devastating industry report from Gartner indicating a potentially significant revenue deficit in AI data centers by 2030, which triggered a massacre in the neocloud complex.

Moreover, the macroeconomic backdrop deteriorated rapidly last week, after inflation numbers came in much hotter than expected, while rising oil prices (CO1:COM) and the US dollar (DXY) worsened liquidity conditions.

At the start of last week, I presented a game plan in my macro post on how I plan on positioning The Portfolio by outlining key support levels for both the S&P 500 (SPY) and SMH.

Last week as fear took over, I executed my plan, drawing down my cash levels from 18% of the portfolio to 13%, as I started expanding exposure across some of my highest-conviction names while also booking multibagger profits in some of my software holdings.

Despite the volatility in the AI space, my portfolio continues to outperform growth benchmarks like the QQQ (QQQ) and AIQ (AIQ), both on an MTD (month-to-date) and YTD (year-to-date) basis, given my disciplined and selective capital allocation strategy and liquidity position in the portfolio.

On a YTD basis, The Portfolio is delivering returns of 27%, compared to 15% for QQQ and 22% for AIQ.

In this post, I will discuss the 4 things that need to happen in order to stop the AI selloff. I have also laid out my forward investing game plan on how I plan to deploy the remainder of my cash should the AI selloff worsen.

1. Oil and Dollar Needs to Move Lower

Up until now, the elevated 10Y US Treasury bond yields were not disrupting the AI trade; however, last week’s hot inflation reports, coupled with rising oil prices above $100/barrel and a strengthening US dollar, affected liquidity conditions.

Meanwhile, the Fed’s hand is tied in all of this. Amid a higher than expected CPI read of 3.7% (compared to 3.4% expectation) and PPI read of 6.3% (vs. the 5.4% expected), the Fed is under pressure to raise interest rates. Except it is not so straightforward.

The Fed is faced with a dilemma, where on one hand, it has little control over the supply-led inflation that we are facing today. On the other hand, if the Fed chooses to raise interest rates, it will significantly raise the borrowing costs of the US government, which in turn will further increase deficits and aggravate the already gargantuan national debt problem.

The good news is that growth in the US remains exceptionally robust, driven both by AI investments and consumer spending. Particularly, AI spending has become one of the biggest drivers of US GDP growth, as can be seen below.

 

In order for the AI trade to revive once again, we first need to see signs of liquidity conditions improving.

The way I see it, it starts with oil prices moving lower below the 200DMA (day moving average) at $83/barrel, driven by a diplomatic breakthrough between the US and Iran. This will almost immediately weaken the US dollar below the 200 DMA to $99.45, which will be positive for liquidity, as the pressure to raise interest rates will ease and long-term bond yields will peak.

2. Hyperscalers Need to Confirm the Spending Cycle

The second biggest catalyst I am watching is what hyperscalers say about their capex plans for the remainder of FY26 and forward guidance for FY27.

In the previous Q2 earnings season, all hyperscalers, with the exception of Microsoft (MSFT), revised their forward capex projections higher. Amazon (AMZN) management even provided one of the cleanest and most credible frameworks by helping investors connect the dots between AI demand, capex, margins, and payback period. They did that by breaking down their AI spending into two categories that included 1) Servers and Networking Equipment and 2) Data Centers, as I explained in this post.

At the moment, current estimates point to hyperscaler capex, including that of Oracle (ORCL), growing to $1.13T, based on research from Apollo. If hyperscalers confirm the spending cycle below, this will be good news for the AI trade in my opinion, as it would mean additional dollars of demand flowing into GPUs, custom AI chips, memory, networking equipment, optical components, power infrastructure, cooling systems, and data center construction.

 

Aside from hyperscalers confirming the spending cycle, we also need to see ongoing acceleration in cloud revenues for AWS, Azure, and GCP (GOOG), along with surging backlog growth to provide forward revenue visibility as well.

In Q2, Azure and GCP accelerated their revenues much faster than expected at 43% and 82%, respectively, while AWS posted their fastest revenue growth in 18 quarters, growing at 37%. Backlog, including Oracle’s, had also surged over 180%+ YoY to $2.3T.

What would also boost investor sentiment is expanding gross margins for their respective Cloud businesses, especially as all hyperscalers are heavily investing in their custom silicon roadmap to build full-stack efficiency gains and improve total cost of ownership for their customers.

Most recently, Microsoft announced that they will scale their Maia family of ASICs to multi-gigawatt deployments, making it the primary inference system for their own products and services, especially after the company redesigned its Copilot, where it bundled Office productivity tools, code generation, and an autonomous AI agent called Autopilot, thus marking a pivotal shift from standalone chatbot assistants to an “AI operating system.“

Separately for Meta (META), all eyes will be on how quickly the company can scale revenue from Muse through a combination of transaction-based and subscription revenues. Based on estimates from Truist research, Meta Muse revenue could reach ~$29B by 2030, with advertising and agentic commerce becoming its largest revenue source at ~44% of Muse revenue.

There is no doubt that the fear in the AI trade is palpable. We can see the fears translate into surging CDS spreads on AI debt, as investors question whether companies can generate enough profits to repay their growing debts. According to this X thread, “bonds financing massive US data centers for the likes of Meta, NVIDIA (NVDA), Oracle, and Microsoft are trading well below face value. Meta’s “Hyperion” bonds dropped to 92 cents on the dollar, Microsoft’s “Project Excalibur” fell below 88, and CoreWeave’s plunged under 80.”

 

However, I believe that hyperscalers are in a much stronger position, given the superior quality of their balance sheets, along with evidence of robust GPU pricing across generation. This tells me that their respective backlogs are converting at higher rental prices than the original contracts, which bodes well for both the top and bottom lines of hyperscalers.

3. The Neocloud Massacre Will Resolve in One of the Two Ways

Unfortunately, neoclouds are in a much more perilous state as compared to hyperscalers.

It all started after news emerged of the potential $20B revenue shortfall for OpenAI, followed by Anthropic’s revenue growth slowdown, which intensified concerns over circular financing.

Per Gartner’s estimates, AI data center owners would need $8.5T in revenue by 2030 to cover costs and earn a decent return. However, spending on AI services that would flow to these providers is projected to reach no more than $3.6T, implying a potential $4.9T shortfall. As a result, Gartner believes that 80% of neoclouds will fail or be absorbed by 2030.

Highly leveraged neocloud company CoreWeave’s (CRWV) shares fell over 20% during the past week, as its 5-year credit default swap spread surged as much as +74.5 basis points on Thursday to a record 1,194.5 basis points, implying a more than 60% probability of default within 5 years. Even after Nvidia CEO Jensen Huang pushed back against concerns over CoreWeave’s financial position, the selloff continued into Friday.

 

According to Morningstar, neocloud operators could face higher debt costs as competition for financing intensifies, with hyperscalers also seeking substantial amounts of capital to expand their own AI infrastructure. They pointed out, however, that Nebius (NBIS), a neocloud competitor to CoreWeave, has a stronger balance sheet than its peers, where the company has raised capital through equity offerings and convertible securities instead, thus limiting its reliance on debt.

The way I see it, there are possible ways to plug the massacre in Neoclouds.

On one hand, if oil prices start moving lower, it will ease inflation fears, push down bond yields and the US dollar, and improve liquidity conditions. This will improve the financing landscape for highly leveraged neocloud companies like CoreWeave with access to capital at reasonable rates.

On the other hand, if neocloud companies can demonstrate that they are converting their backlog into revenue and cash flows at a faster rate by bringing more capacity online, that could temporarily soothe investor anxiety as well. This is where I think hyperscaler earnings will matter a lot for neoclouds and the overall state of lending conditions in AI, because if they continue to say that capacity is struggling to catch up to demand, it will validate neoclouds’ existence at least in the near term.

4. A Successful Anthropic IPO Could Stabilize the AI Trade

The third catalyst is a potential Anthropic IPO next month or so, which I believe could serve as a major bullish trigger, drawing significant liquidity into the markets ahead of its listing.

Anthropic reported Q3 revenue of $16.8B, up 45% QoQ. But despite reporting adjusted profitability, investors weren’t impressed, as it marked a significant slowdown from the 100% sequential growth it posted in Q2.

But here is the thing. Anthropic is planning to spend more than $500B on computing and AI infrastructure over the next decade. This equates to an average of $50B annually, which will trickle through GPUs, memory, networking equipment, optical components, power infrastructure, and more.

While some claim that it is better that Anthropic delays its IPO, as its expected valuation has come down from $2T to $1.5T, it still represents a rich revenue multiple of over 20x its annualized revenue, which can align market expectations towards a predictable baseline and reduce volatility, in my opinion.

How I’m Positioning, Risks & Conclusion

The past week, as the S&P 500 retreated 1.8% from its all-time highs, we saw capital rotate heavily out of the AI trade into sectors like Financials and Energy, like I said it would in my previous macro post.

It is important to note, however, that the S&P 500 has not broken any major support level that I had outlined, with 21 WMA (exponential weekly moving average) at 7562 as the first stop.

So, this is how I see it all playing out in the coming weeks.

In the first scenario, the S&P 500 holds support and bounces higher to new all-time highs. This is a possible scenario, as the fear index is in “Extreme Fear” territory. In this case, we will see SMH also move higher with the S&P 500, but unlike the S&P 500, SMH will likely create a lower high. Should this scenario play out, it will increase the chances of yet another painful correction in SMH after the S&P 500 creates a new all-time high.

In the alternative scenario, the S&P 500 pulls back between 7597 and 7271, a decline of another 1-5% from current levels. In this case, SMH continues lower to 200 DMA as a first support level, followed by 50 WMA as the final support level, representing a decline of another 5-10% from current levels.

As I explained in my previous post, neither of these scenarios indicates bear markets; instead, I view them as phenomenal buying opportunities for investors.

Even BofA analyst Vivek Arya said Nvidia, Broadcom (AVGO), and Micron (MU) are already priced like Cisco (CSCO) after the dot-com crash. “Investors are paying post crash multiples for a cycle that hasn’t crashed,” he wrote in a note dated Friday, per Benzinga.

As for my game plan, I started last week with 18% cash in The Portfolio, which I had planned to draw down by a third should SMH fall between $565 and $545 levels. During the week, I drew my cash down to 13% of the portfolio as SMH came to my first support level by nibbling back into companies like AMD (AMD), Cerebras (CBRS), Vertiv (VRT), MaxLinear (MXL), and more. I also raised some by booking multibagger profits from Palantir (PLTR) and CrowdStrike (CRWD) holdings.

Despite the heightened volatility in the AI trade, my highly disciplined and selective capital deployment strategy backed by an ironclad game plan continues to help The Portfolio deliver superior returns both on an MTD and YTD basis, compared to other growth benchmarks, as can be seen below.

 

From where we currently stand, my game plan forward is simple and straightforward.

If the S&P 500 continues lower towards the two support zones outlined above, followed by SMH finding support at 200 DMA, followed by 50 WMA and 501, I will deploy 50% of my remaining cash position at each of the support levels for SMH across my high conviction picks in the portfolio.

On the other hand, if the S&P 500 finds support at current levels and starts to move higher, I will monitor the nature of the bounce in SMH and may raise cash for a possible deeper pullback at the end of the year or start of next year.

One of the biggest risks that I am monitoring is the state of liquidity conditions, because if oil refuses to move lower, the US dollar could strengthen further, which would not bode well for the liquidity environment in the near-term picture of the AI trade. That doesn’t mean that the fundamental AI thesis breaks down. On the contrary, it is at these times that it is important to separate price action from thesis deterioration, and for investors like myself who have a game plan ready before volatility shows up with plenty of liquidity on the sidelines, this is a phenomenal opportunity to double down before the next leg of the rally starts.