Meta said it will spend more on AI and Tesla said it will spend less, yet chip stocks fell on both. Should I sell my semiconductor ETF now or hold on?

Why chips got hit whether spending went up or down

Meta raised spending and Tesla cut it, yet semiconductors were sold on both. It looks contradictory, but the market was asking one question: "Where will the money to keep investing in AI come from?"

Meta raised a question about 'funding,' not 'demand'

Meta's Q3 capex of $34.9 billion exceeded operating cash flow of $32.8 billion, turning quarterly free cash flow negative for the first time since its 2012 IPO. Future spending commitments related to data centers rose from $700 billion in July to $825 billion, and Meta created non-voting Class C shares, opening a path to raise capital through equity (FXStreet, 2026-10-22). Meta's 5-year CDS rose above 119bp, the highest since trading began (FXStreet citing Bloomberg, 2026-10-22).

This is not just Meta's problem. UBS estimates that hyperscaler capex this year is consuming nearly 100% of operating cash flow, compared with a 10-year average of about 40% (AI Weekly citing UBS). Investment once covered by cash is now starting to rely on debt and equity issuance. The bond market is already pricing it.

"Bonds of AI-related issuers trade at spreads of about 115bp relative to similarly rated peers, wider than the 78bp for investment grade overall" — Lon Erickson, Thornburg (cited by AI Weekly)

For chipmakers, a structure in which customers borrow money to buy chips means leaving order durability to the credit market. That is why both Meta's "we'll spend more" and Tesla's "we'll spend less" were read to the same conclusion.

The Lam Research shock is different in nature

Lam Research's decline reflects a supply bottleneck, not demand worries. The CFO said December-quarter guidance ($8.6 billion) would have been "a little higher" without component constraints and expects semiconductor equipment spending to exceed $200 billion in 2027. Delayed orders will ship in the March quarter (24/7 Wall St., 2026-10-22). The CEO said it is "completely different" from the pandemic-era shortages and that Lam will catch up soon. That AMAT (-6%) and ASML (-4%) fell less than Lam also signals selling concentrated on Lam-specific issues (24/7 Wall St., 2026-10-22). Part of the equipment stock decline is a matter of delayed shipment timing rather than earnings damage, so it should be viewed separately from memory and AI chips.

What the signal dashboard says

SMH's recent trend as of 10/20 is as follows.

On 10/20 the close was already below the MA150, the buffer to the exit line was in the 1% range, and the risk flag was on. The 10/22 close of 523.35 is about 4% below 10/20, so it has likely fallen below the exit line (estimated, since 10/21 and 10/22 signal values are not yet reflected). The MA10 falling below the MA50 also means the short-term trend has become weaker than the medium-term trend.

So what to do with semiconductor ETFs?

• Hold off on new entries. Entry conditions are off on the signals, and both rebounds in October faded within a day. It is better to confirm at least a close back above the MA150 (about 551) before building in tranches.
• Reducing 3x leverage (such as SOXL) comes first. SOXL mentions on Reddit jumped from 37 to 99 in a day, but in a volatile downtrend leveraged ETFs leave losses even if the index returns to where it started.
• If you hold, distinguish between equipment and memory. Equipment faces a timing issue of shipment delays, while memory and AI chips face a structural issue of customer financing. Equipment is likely to recover first.
• The next checkpoint is Amazon earnings on 10/29. If even Amazon shows higher capex combined with deteriorating cash flow, the "funding problem" will harden into a hyperscaler-wide issue. Conversely, if accelerating cloud revenue is confirmed, the recoverability of spending will come back into focus.

Detailed charts and signals for other tickers are available on the /signals dashboard.