Steven Fiorillo
Seeking Alpha Analyst, Basis Point Podcast Co-Host
Energy Transfer offers data center exposure because its pipelines run through key regions where data centers are being built, and natural-gas turbines and fuel cells are a big part of powering them. The case got stronger after Iran's strikes knocked out Middle East capacity. x.comÂ
Energy infrastructure looks like the next memory trade, as the AI race shifts from software to energy. China produces about three times as much energy as the US, and Musk and Jensen Huang said every 5GW of new energy adds about 1% to US GDP. AI needs power, cooling and infrastructure, so the companies enabling the buildout, from Vistra, Constellation and GE Vernova to Bloom Energy, Vertiv, BWXT, Cameco, Enphase and Tesla, are worth accumulating on weakness. x.com x.comÂ
CoreWeave's results confirm the AI supercycle, and the next wave of capital is moving beyond semiconductors after their big runs. CoreWeave reported $2.58 billion of revenue, up 122%, a $104 billion backlog and more than $25 billion of new customer commitments already in Q3. That capital is flowing into data centers (Nebius, IREN, Riot, Applied Digital), networking (Arista, Marvell), photonics (Lumentum, Coherent, Applied Optoelectronics) and power (Vertiv, Constellation, Vistra, Oklo). x.com … Show more
Nextpower is building on power for AI, the foundation of the AI secular bull market, which is sturdier than boom-and-bust software and chips. It is buying Prevalon Energy, a utility-scale battery storage company already engaged with hyperscalers, extending its rebrand from solar trackers into end-to-end solar and power platforms. Its last quarter brought 34% sales growth, a 2.25 GW Bisha Solar Project commitment, $953 million of cash with no debt and a $500 million buyback. x.comÂ
Bloom Energy's addition to the S&P 500 on AI-power demand confirms the power thesis through the index itself. Chips and power remain the areas to lean into despite a hot 162K jobs report that lifted rate-hike odds to about 65%, while staying selective in software, keeping cash and buying gradually. x.comÂ
Navitas Semiconductor's GaN chips address the power bottleneck in AI data centers, where rising GPU power draw has pushed silicon power chips to their physical limits. Its CEO puts GaN's design value at $10,000-$15,000 per MW, and high-power market revenue rose 35% year over year in Q1. After a brutal 2025 of weak consumer demand, dilution and losses, the Navitas 2.0 pivot and Q1's return to sequential growth suggest the worst is over, with its AI power line moving into system validation. x.comÂ
The tech rally shows little sign of overheating because AI fundamentals remain strong, with robust token usage, ARR and capex. Big 5 cloud capex is set to rise 82% in 2026, with upward revisions for Microsoft and Meta, and 2027 capex is expected to grow about 30%, above the 26% consensus. TSMC CoWoS capacity forecasts were raised to 145K and 175K wafers per month by end-2026 and 2027, and price hikes are spreading across CPUs, memory, mature nodes, copper foil and optical lasers. x.comÂ
Alpha and Omega Semiconductor's pivot to AI data-center power is gaining traction beneath muted overall revenue. AI-focused power delivery in its advanced computing division grew over 40% year over year and doubled sequentially, and CEO Stephen Chang reports rising demand from hyperscalers and supercomputing operators. High-margin power ICs grew 30% to nearly 40% of product sales. Insider selling and valuation froth may cause volatility, and second-half results need to validate AI growth. x.comÂ
The AI trade is much broader than semiconductors, and the biggest winners could come from every layer of the buildout. AMD gets paid for compute, Amazon for cloud infrastructure, ServiceNow for enterprise AI deployment, Vistra for the electricity, and Brookfield for financing and building the infrastructure behind it all. x.comÂ
If recursive AI makes intelligence cheaper, demand explodes and the bottleneck moves to the physical world, which favors suppliers of GPUs, HBM, interconnect and power such as Nvidia, Bloom Energy, Eaton, Vertiv and Wolfspeed. Recursive self-improvement is debated as a software question, but that framing is wrong. x.comÂ