Earnings season has arrived with significant force, with roughly 40% of the S&P 500 set to report and five of the largest mega-cap names releasing results in a single week. Volatility is fully expected, but the more interesting question is what the market actually cares about right now. Are investors focused on earnings, the Federal Reserve narrative, or the inflation picture? The answer will likely emerge in pieces over the coming days, with Wednesday providing the first real glimpse.
From an earnings perspective, the most important variable to track is artificial intelligence capital expenditure. There is no supply problem in this market — there is a demand story. The four major hyperscalers — Amazon, Google, Microsoft, and Meta — are projected to spend roughly $680 billion on capex in 2026, a 64% year-over-year increase, and those estimates are still moving higher. Including Oracle, the figure climbs further. The questions worth asking heading into earnings are: where is that AI demand showing up, where is the spend being directed, and is forward guidance reinforcing the narrative that AI demand continues to outpace supply?
With that framing in mind, three stocks stand out — one as a defensive ballast, and two as direct plays on emerging AI infrastructure bottlenecks.
T-Mobile: A Dividend and Growth Crossover
T-Mobile is the outlier here, the one pick that sits outside the AI infrastructure theme. It surfaced on two different screens in the same week — a growth screen and a high-quality dividend yield screen — which is always a signal worth paying attention to. The fundamentals are clean: a dividend yield of 2.23%, a payout ratio of 39%, and trailing twelve-month free cash flow per share of $21.76. That combination matters. The yield is comfortably covered by free cash flow, the low payout ratio leaves meaningful room for reinvestment into growth, and the overall profile suggests a company that can smooth out portfolio volatility without sacrificing upside.
Technically, the picture is more mixed. The chart shows a downward-sloping trend line connecting subsequent highs, and prices have not yet returned to test that line. The stock recently approached its 52-week low near $181.36 but did not break through, instead consolidating slightly above that level. All the major moving averages are sloping lower with price below them, and the 5-day exponential moving average sits around $188. There is, however, a small positive signal: the relative strength index is compressing into a triangular pattern and showing bullish divergence heading into earnings. Momentum often slows ahead of a known event, so this can be read as constructive rather than ominous. The volume profile points to heavy trading activity around $198, which would act as the next meaningful resistance level if earnings catalyze upside. The 29% decline from the highs looks dramatic on screen, but in context it is less severe than it first appears.
Credo: Solving Inference Bottlenecks
Credo has had a remarkable run — up nearly 290% over the past year — but recent weakness across AI-exposed names has dragged it down. Some of that pressure stems from a leaked internal memo at a major AI lab indicating that internal revenue targets had not been met, alongside a noticeable shift in user behavior between leading models. Even when capabilities like advanced video generation are temporarily pulled, the underlying issue is the same: a compute and inference problem.
This is precisely where Credo's positioning becomes interesting. Memory bottlenecks limit AI inference scalability, and Credo addresses one of the most expensive failure modes in large-scale AI training. In massive GPU clusters, a single "link flap" — a millisecond-long signal disconnect — can crash an entire training run, costing millions of dollars in lost compute time. Their ZeroFlap solution uses predictive telemetry to anticipate and prevent these failures, dramatically improving uptime and reliability for large-scale machine learning models. The acquisition of Dust Photonics expands their silicon photonics capabilities, which aligns with the broader industry shift away from copper toward photonics for high-speed GPU connectivity. As AI competition intensifies and the cost of compute continues to climb, solutions that protect the integrity of training runs become structurally valuable.
The chart reflects the volatility of a high-beta name. Credo recently posted a new closing high but stumbled at the same $194 level that marked an earlier intraday high from October, suggesting that resistance is real. Old highs around $165 are now acting as support, with another supportive zone in the $135 to $149 range tied to a prior gap. The stock sits mostly above its moving averages, though the short-term 5-day exponential moving average has begun pointing downward — typically the earliest signal of a possible trend change. The relative strength index has broken its own trend line and slipped below overbought territory, indicating a slowdown in momentum. The volume profile's point of control sits between roughly $136 and $150, marking the most important downside node to watch.
Marvell: The Switzerland of Connectivity
Marvell sits at the same structural intersection but from a different angle. The thesis here connects directly to the dual-track interconnect strategy that has been outlined publicly by major chip industry figures: push copper bandwidth as far as it can go, and use photonics for everything beyond that. Copper's effective range tops out around two meters, which constrains it to certain rack configurations. For genuine scale-up — connecting hundreds of thousands of GPUs across racks — photonics becomes mandatory.
Marvell is well positioned as a beneficiary of this transition. Their acquisition of Celestial AI advances optical scale-up connectivity and high-speed data transfer capabilities. They have explicitly positioned themselves as neutral across hyperscalers — the "Switzerland of connectivity" — which is an attractive structural stance in a market where customers want to avoid lock-in. There is also a competitive dimension: their optical switching solutions could serve as alternatives to certain offerings from dominant incumbents in the GPU connectivity space.
Recent news has weighed on the stock. Marvell cancelled a slew of orders with Poet, with allegations of confidentiality breaches surrounding the dispute. There is some speculation that those orders are simply being redirected to other customers, which would itself reflect the intense competition for the data and resources required by large language models. Either way, the structural thesis — Marvell as a solution provider for an emerging compute and connectivity bottleneck — remains intact.
The technical picture, however, calls for caution. Marvell ran from a long sideways period into a very narrow, very steep upward channel, peaking at $170.84 before breaking down. The general principle here is reliable: the sharper and steeper the rally, the less sustainable it tends to be. That does not preclude further upside over a longer horizon, but it does create the conditions where traders grow nervous about pullbacks after extended runs. A gap between $140 and $144 now serves as a potential support area, with deeper support at $122 to $128 — though that is well below current trading. The relative strength index is hanging just at 70, with a close below that level potentially signaling a bearish turn. The volume profile shows a relatively thin node between $130 and $137, with most of the heavy historical volume sitting much lower in the $70 to $95 range.
The Underlying Theme
What ties this trio together is a recognition that the AI buildout is no longer just a story about chips. The next leg of the story is about everything around the chips — the connectivity, the photonics, the inference reliability, the data movement. As AI demand continues to outpace supply and the major hyperscalers pour hundreds of billions of dollars into capex, the bottlenecks shift. They move from raw compute to memory, to interconnect, to the optical infrastructure required to scale beyond the physical limits of copper. Companies that solve those bottlenecks become structurally important, even if their charts move violently in the short term.
Pairing those exposures with a defensive cash-generative name like T-Mobile is a way to acknowledge the volatility inherent in this trade without abandoning the upside. The AI narrative remains constructive, but how the market chooses to weight earnings, inflation, and Fed policy in the coming days will determine whether that narrative has room to keep running — or whether it needs to consolidate before the next push.