AI doubts hit hardware
The first cracks in the AI trade usually show up in semis, and that’s what happened. Samsung and SK Hynix slid to roughly 16‑month lows, and the message is straightforward: investors are shifting from “AI spend is infinite” to “how much capex is too much, and when does it pay back?” Memory sits at the intersection of hype and physics. When it rolls over, sentiment tends to follow.
This isn’t “AI is over.” It’s the market treating upstream hardware less like a secular cheat code and more like cycle beta—still strategically important, but priced with a longer payback window and less forgiveness on the next quarter’s demand signal. If hyperscaler spend keeps growing but at a more scrutinized pace, the most capex‑tethered parts of the stack are where the multiple compression shows up first.
Software keeps shipping
While hardware got hit, the product machine didn’t slow down. Google pushed Gemini 4 Argon to developers and businesses globally. That contrast matters: capability and distribution are accelerating, but public markets are starting to price the infrastructure side like a maturing buildout where ROI needs to show up faster and cleaner.
Dispersion remains the trade. Platforms that can point to usage, retention, and dollars per user can keep their footing. The supply chain names tied to incremental capex are the ones that take the hit when expectations get clipped.
Agents, not slogans
One deal headline fit the tape. Decagon reportedly held acquisition talks with ServiceNow and others around its AI agent business. Even as chatter, it maps to what buyers want right now: agent workflows tied to measurable outcomes.
“This saves time and money” sells. “Trust us, the TAM is infinite” doesn’t.
Late-cycle pressure
The broader setup looks familiar: S&P 500 earnings growth gets harder to deliver at current expectations, which narrows the path for stocks. “Beat and raise” is the clean lane; anything else invites questions about volumes, pricing power, and margin cadence. With rate volatility in the background, multiples don’t have much patience. Stocks sold, not debated.
That also helps explain why M&A keeps showing up where it does. If organic growth is harder to manufacture, buying product capability or distribution starts to look rational again—especially when “AI” can be translated into near-term productivity or cost-out. The Decagon/ServiceNow angle sits right in that lane: large enterprise platforms would rather bolt on proven agent workflows than wait through longer build cycles, and they’ll pay for anything that can credibly point to ROI without a three-year spreadsheet.
Rates and supply
Rates and fiscal stayed as the backdrop nobody can mute. Ray Dalio warned the U.S. Treasury selloff may have meaningful downside remaining on higher borrowing needs and weaker marginal demand. You don’t have to agree to respect the setup: heavy issuance plus less reliable marginal buyers keeps yields elevated and jumpy, and that’s not friendly for long-duration equity. The market only needs a little extra term premium to make growth multiples behave badly.
Europe added its own reminder with France opening 2027 budget discussions amid deficit and debt concerns. In the U.S., the Social Security timeline (projected insolvency in six years) stays in the background as steady fiscal friction. Not a day-trading catalyst, but it’s part of the context for future issuance and policy choices.
On commodities, the cleanest “real economy” headline was supply. Guyana is set to exceed 1 million bpd in oil exports for the first time, with Uaru grade shipments starting in December—another credible source of incremental non‑OPEC barrels. Separately, the EPA proposed a complete reallocation to offset biofuel waiver cuts—technical, but meaningful through compliance costs and demand allocation.
What mattered today
- AI trade split: hardware (Samsung/SK Hynix) is pricing a longer, messier capex-to-ROI path; software platforms keep shipping (Google’s Gemini 4 Argon).
- M&A logic strengthens: Decagon/ServiceNow chatter fits the market’s preference for agentic AI tied to measurable workflow outcomes.
- Macro still matters: issuance and rate volatility hang over multiples; fiscal headlines aren’t noise, they’re context.
- Flows stayed “in the market”:all 11 U.S. sector ETFs saw net inflows over the week, while gold ETFs led net outflows.
The market isn’t shorting AI—it’s demanding receipts.