The post-Nvidia wobble is being mistaken for an AI unwind when the better read is leadership broadening across the same capex cycle. Dell’s guidance reset, plus continued strength in memory, foundry, and custom silicon, argues for selectivity inside AI infrastructure rather than a blanket bubble call.
The mistake in the current debate is treating AI as if it were a one-stock trade. NVDA remains the flagship, but the market rarely pays the same premium to the first winner forever once the spending cycle matures. As hyperscalers and enterprises move from buying accelerators to building full systems around them, the beneficiaries naturally widen to servers, networking, memory, foundry capacity, and custom silicon. That is why the right question now is not whether AI is over, but which layer of the stack is getting the next dollar of spend.
Start with valuation versus operating reality. NVDA still looks expensive at 33.06x earnings and 20.63x sales, but those multiples sit on top of 65.5% revenue growth and a 63.0% net margin. That is not the profile of a business cracking under its own hype. The more interesting tell is that some of the supposed "next leg" names are actually richer on earnings despite lower profitability or less dominant economics. AMD trades at 169.75x earnings with a 13.4% net margin, while AVGO sits at 85.81x. If this were a true AI unwind, investors would not be rotating into adjacent names with their own stretched multiples; they would be exiting the complex altogether.
The better evidence is in where fundamentals are accelerating. DELL is the clearest second-order beneficiary because it sits at the rack level, where AI enthusiasm becomes real purchase orders. Its stock is up 223.6% YTD, but the more important figure is the guidance reset: management did not just talk up demand, it materially raised the revenue range. That matters because it shows the spend is moving from chips in isolation to integrated systems. A market that was truly losing faith in AI demand would not reward a server vendor for saying the buildout is getting bigger.
The same broadening shows up across the supply chain:
That mix matters. MU and TSM are not meme appendages to the AI story; they are core bottlenecks in memory and manufacturing. AVGO adds another layer through custom silicon and connectivity, and its long-dated customer commitments reinforce the point that AI demand is being institutionalized, not abandoned. When the tape starts rewarding memory, foundry, and systems names alongside the GPU leader, that is usually a sign of a capex cycle deepening through the stack.
Yes, the bulls can overstate this and the skeptics are right about one thing: parts of the group are crowded. Semis and memory have done a huge amount of the market's work this year, and when a theme gets that consensus, a pullback can hit everything at once. There is also a real strategic shift underway from training toward inference and custom silicon, which could chip away at NVDA's singular dominance. But that is exactly why calling this a broken AI trade misses the point. A weakening one-stock monopoly case is not the same thing as weakening AI infrastructure demand.
If anything, the current setup looks more like a classic late-stage leadership handoff inside a still-intact boom. The first phase was obvious: own the company selling the indispensable accelerators. The next phase is messier and more selective: own the companies that assemble the servers, provide the networking, manufacture the wafers, and supply the memory that lets those systems scale. That is why ANET, MU, TSM, AVGO, and DELL keep showing up in the conversation. Investors are not asking whether AI exists anymore; they are asking where the next margin pool sits.
The clean read is that the AI trade is broadening, not breaking. NVDA can wobble after earnings and still leave the larger capex story intact if the surrounding ecosystem keeps posting stronger orders, better guidance, and durable growth. Right now, that is exactly what the evidence suggests.
What would change our mind? A real rollback in data-center spending, not just a valuation shakeout in semis. If server demand cools, memory tightness eases materially, and foundry commentary stops pointing to AI-led capacity pressure, then the rotation case weakens fast. Until then, the smarter stance is not a blanket bubble call but selective exposure to the second-order winners of the same buildout.
This week’s Q2 earnings recap showed that strong EPS results were not enough to lift every stock. ServiceNow and T-Mobile rallied on beats and upbeat growth stories, while Intel, American Express, and Tesla fell despite key business wins and management optimism.
US data painted a split picture: initial jobless claims fell to 187,000, the lowest since 1969, while the July PMI showed stronger growth and hotter price pressures. Rising Treasury and mortgage rates tightened conditions, keeping the market focused on a soft landing with a stubborn inflation problem.
Private credit is still attracting capital, but falling direct-lending activity is making deployment, underwriting and liquidity more important than fundraising totals. The risk is not an immediate default crisis; it is pressure to put money to work as eligible deals shrink and marks become harder to trust.