The competitor headline leads with percentage gains. That tells you nothing about whether the setup is still intact. A stock up 200% can still be a buy if growth and revisions remain top-quartile versus peers. It can also be a trap if the only thing left is momentum coasting on fading enthusiasm. The question isn't whether the list produced past winners. It's whether the factor stack - valuation relative to growth, earnings trajectory, and sector positioning - still supports adding to these names right now.
Here's the landscape. Big Tech capex is accelerating, not retreating. Microsoft guided to $190 billion in calendar 2026 capex. Alphabet raised its full-year guidance to $180–$190 billion. Meta sits at $125–$145 billion. Combined, the hyperscalers plan $725 billion in capital spending this year - a 77% increase over 2025. That spending doesn't appear anywhere else on these companies' income statements. It flows through semiconductors, networking, and memory. So the primary driver for this sector remains intact. The early-2026 rotation away from AI stocks turned out to be sentiment, not fundamentals. Money flowed back in.
The picks below are ranked by how well the factor stack holds up - not by which story is loudest. Each name has a portfolio role. Without one, it's just data.
1. NVIDIA (NVDA) - Growth anchor
NVIDIA is the AI infrastructure index, and that's the whole point of holding it. Fiscal Q4 2026 revenue was $215.9 billion, up 65% year over year, with gross margins at 75%. The trailing P/E is around 32 and the PEG ratio... sits near 0.3 - meaning the stock is priced at roughly 30 cents per point of expected growth. That is cheap for a company still growing its earnings in the mid-to-high 60s. A PEG below 0.5 is what growth investors look for when they want expansion without paying a premium. NVDA is the only mega-cap name in this group where growth is still outpacing valuation expansion. Role: core growth sleeve holding. Trim only if the PEG pushes above 0.8 or gross margins break below 70%.
2. Micron (MU) - Bottleneck compounding
Micron crossed $1 trillion in market cap, but the story isn't about market cap - it's about pricing power in a supply-constrained niche. Q2 revenue surged 196% year over year. Gross margins hit 74.9%. The company's HBM... capacity is sold out through 2026. Mizuho raised its price target to $800. The commodity memory cycle that used to punish Micron every 18 months has been replaced by structural AI demand. That changes the entire valuation framework: you're no longer buying a cyclical trough; you're buying a priced-in monopoly on a constrained supply chain. Role: momentum-add for growth portfolios. The trigger to reduce is HBM inventory buildup or a hyperscaler capex pullback.
3. AMD (AMD) - Earnings acceleration
AMD's Q1 2026 results nearly doubled net income to $1.38 billion, up 95% year over year on $10.3 billion in revenue. The stock jumped nearly 15% the next day and at least 20 brokerages raised price targets. What matters here is the trajectory: AMD is moving from "AI also-ran" to actual profit contributor. The MI300 product line is gaining design wins, and MI400 is in pipeline. The growth numbers don't yet match NVIDIA's, but the earnings acceleration rate is among the fastest in the sector. The factor question isn't whether AMD beats NVIDIA - it's whether the rate of improvement justifies holding through volatility. In our book, a stock moving faster toward profitability is often more actionable than one already there and slowing. Role: satellite growth position sized smaller than NVDA. Add on sector pullbacks.
4. Arista Networks (ANET) - Networking infrastructure play
Arista reported Q1 2026 revenue of $2.71 billion, up 35.1%, with $1.69 billion in operating cash flow. The company raised its full-year 2026 revenue target to approximately $11.5 billion, implying 27.7% growth, and increased its AI fabrics revenue goal to $3.5 billion. Networking is the plumbing layer most investors skip when they think about AI - but hyperscaler clusters need switching capacity as much as they need GPUs. ANET's 800G and upcoming 1.6T Ethernet switches are the interconnect backbone for data center AI clusters. The business is less visible than chip names but structurally essential. That obscurity is the advantage. Role: diversifier within the growth sleeve. If you already hold NVDA and AMD, ANET adds exposure to the networking sub-segment without duplicating GPU risk.
5. Taiwan Semiconductor (TSM) - Foundry durability
TSMC produced $35 billion in Q1 2026 revenue, up 35.1%, with gross margins of 66.2% - beating expectations. The company forecasts approximately 30% revenue growth for 2026. Every AI chip on this list, including NVIDIA's, AMD's, and most custom hyperscaler silicon, runs through TSM's fab capacity. That makes TSM the toll road of AI infrastructure. The valuation is higher than historical semiconductor averages - it embeds expectations for mid-60s margins and sustained ~30% growth - but the moat is unmatched. There is no credible alternative foundry at advanced nodes. Role: defensive growth holding. Hold through volatility; the geographic concentration risk (Taiwan) is real but has not materialized despite repeated headlines.
6. Broadcom (AVGO) - Yield-and-growth hybrid
Broadcom is the most complex pick on the list. It trades at a trailing P/E of roughly 87, which looks expensive on any spreadsheet, but the forward P/E near 40 reflects expected AI revenue acceleration through its custom chip and networking businesses. Q2 FY2026 earnings are due June 3; consensus calls for about $22 billion in revenue and adjusted EPS growth near 51%. The stock sits at roughly $2.1 trillion in market cap. The valuation gap between trailing and forward multiples is wide because the market is pricing in a structural shift in Broadcom's revenue mix away from legacy semiconductor solutions toward AI custom silicon and VMware integration revenue. That's a bet, not a certainty. Role: growth-income crossover name. The barbell utility: it behaves like a growth stock when AI spending accelerates and has enough yield to cushion when it doesn't.
These six names span the AI infrastructure stack: compute (NVDA, AMD), memory (MU), networking (ANET), foundry (TSM), and custom silicon (AVGO). That's the structure. If capex holds at the levels Big Tech has guided, each layer benefits. If capex disappoints - and here is the real question - the stack compresses, but the names with the widest moats (NVDA, TSM) and the strongest supply constraints (MU) absorb the hit first and recover fastest. The way you position depends on which regime you expect, but the factor stack tells you which names to hold through either outcome. Watch Broadcom's June 3 earnings as the near-term signal for whether AI revenue acceleration is broadening beyond NVIDIA. Watch Micron's HBM pricing as the early indicator of whether demand outstrips new fab capacity. Everything else is noise around those two data points.
