Confirmation That We Remain in the Early Innings
The latest results from the dominant AI accelerator vendor offer about as clear a signal as the market is going to get that we are still very much in the early phase of the artificial intelligence cycle. The quarter was a blowout — a beat on the print and guidance raised — and it answered a number of lingering questions that had been weighing on sentiment. As the next-generation Rubin platform ramps, margins are holding, which is a key data point given how often investors fear that a product transition will compress profitability. Supply remains tightly constrained relative to demand, and the outlook stretches not just a couple of quarters out but multiple years.
Equally interesting is the evolution on the capital allocation front. With the stock somewhat stuck in a trading range, the leadership has turned its attention to buybacks and dividends — moves designed to widen the aperture and make the company palatable to value-oriented investors who have historically sat on the sidelines of the AI trade. That is a meaningful shift in shareholder strategy and acknowledges that, at this scale, future ownership will have to come from a broader investor base than the growth-momentum crowd alone.
The Picks and Shovels Are Spreading the Wealth
One firm's capital expenditure is another firm's revenue, and the beneficiaries of this buildout extend well beyond the headline name at the top of the food chain. Several companies are receiving prepayments and direct support as the supply chain is constructed. The semiconductor capital equipment names — Applied Materials, Lam Research, and ASML — are core picks-and-shovels plays. The leading foundry, TSMC, is an obvious beneficiary, as are the memory suppliers Micron and SK Hynix, which are essential as workloads grow more memory-intensive.
But the more nuanced and arguably more important story is unfolding one layer up, in the application stack. Inference workloads are only beginning to ramp, and as they do, value recognition will increasingly accrue to companies whose products embed AI directly into enterprise business processes. ServiceNow is a prime example — a platform positioned to host the agentic workflows that will progressively be woven into how large organizations operate. Notably, the dominant accelerator vendor's commentary highlighted strength in its CPU business, which correlates closely to AI workload growth and reinforces just how broad the demand for compute capability has become.
The Software Re-Rating
The market is beginning to sort out which software businesses are at risk and which are not. Not every category will be equally exposed: project management and collaboration tools sit closer to the bullseye of disruption from foundation-model vendors. By contrast, mission-critical systems of record that orchestrate enterprise business processes look considerably more durable, and sentiment is starting to reflect that distinction.
Cybersecurity is undergoing a parallel re-rating. Recent attention around a powerful new frontier model — and the prevetting work being done across the security landscape — underscores that securing AI itself becomes increasingly critical as the application and inference sides scale. That benefits players like CrowdStrike, Cloudflare, and Palo Alto Networks, whose offerings sit in the path of every enterprise's AI rollout.
A Bear Case Worth Examining
There is a contrarian argument worth confronting: that a frontier model can become so powerful and so potentially dangerous — to the point of drawing scrutiny from Washington and summoning major banks to discuss its implications — that access ends up restricted to a small handful of approved entities, undermining the monetization thesis. It is not a frivolous concern, but it overstates the case. Governance and data-security issues are real, and innovation is moving so quickly that good actors and bad actors are leapfrogging one another in real time. Yet the industry has shown a meaningful capacity for self-regulation, engaging proactively with federal regulators and with peers to figure out how to scale these capabilities safely. That cooperative posture is itself a bullish signal.
Quantum Steps Out of the Lab
The most underappreciated development sits adjacent to AI: quantum computing. Roughly eighteen months ago, the prevailing view at the top of the chip industry was that quantum was getting ahead of itself — a position quietly softened in the months that followed. Now the government has begun to lean in, and that matters a great deal.
Washington has always played a role in selecting tech winners and losers, from the early days of relational databases to the procurement decisions that anchored AWS as a primary cloud provider, to the visible backing of Palantir, to the more recent transformative support extended to Intel as a domestic advanced-semiconductor manufacturer. This week brought another inflection. The Commerce Department extended the CHIPS and Science Act, deploying billions of dollars into the market. IBM, a major beneficiary, ramped sharply in response, while smaller pure-play quantum names like D-Wave and Rigetti participated in the move. IBM is also launching a new chip foundry with federal backing, aimed at secure semiconductor manufacturing.
The industry appears to be learning from past cycles. Over-reliance on a single offshore foundry has had real consequences — most visibly in the leading-edge silicon export issues that have constrained the dominant accelerator vendor's ability to ship to China. Building domestic capacity now, before quantum reaches scale, is a deliberate hedge against repeating that mistake. The most telling signal is that the conversation around quantum has shifted from research and development to manufacturing and supply chain. That is what an industry sounds like when commercialization is closer than the consensus believes.
The Constraint That Defines the Cycle
For all the optimism, the binding constraint of this entire build-out remains capacity — particularly in memory, where supply tightness is already shaping the pace at which AI infrastructure can be deployed. That bottleneck deserves close attention, because it dictates not only near-term earnings power for the suppliers but also the rate at which the broader application and inference economy can mature. The buildout is real, the beneficiaries are widening, and the next leg — quantum and a more secure domestic compute base — is already being laid. Whether the supply chain can keep up is the question that will shape the next several quarters.