Stelian Mitu @stelianx: "𝗧𝗵𝗲 𝗔𝗜 𝗲𝗺𝗽𝗲𝗿𝗼𝗿𝘀' 𝗻𝗲𝘄 𝗰𝗹𝗼𝘁𝗵𝗲𝘀?..."

Romania

𝗧𝗵𝗲 𝗔𝗜 𝗲𝗺𝗽𝗲𝗿𝗼𝗿𝘀' 𝗻𝗲𝘄 𝗰𝗹𝗼𝘁𝗵𝗲𝘀? 𝗖𝗹𝗼𝘂𝗱𝘀, 𝗳𝗼𝗴 𝗮𝗻𝗱 𝗺𝗶𝗿𝗿𝗼𝗿𝘀 Big Tech is on track to pour 650billion into AI investment in 2026 alone $MSFT (Microsoft) $AMZN (Amazon.com Inc) $META (Meta Platforms Inc) etc (Bloomberg, Feb 2026). Hyperscalers keep announcing campus after campus, with roughly 100 GW of new global capacity between 2026 and 2030 as a $3 trillion "supercycle" (JLL 2026 Global Data Center Outlook). That's the headline. The plumbing tells a more sober story. 🔌 𝗪𝗵𝗮𝘁 𝘁𝗵𝗲 𝗱𝗮𝘁𝗮 𝘀𝗵𝗼𝘄𝘀 Per Sightline Climate (Feb 2026 outlook, refreshed May 2026, cited by Bloomberg): - About 16 GW is slated to come online in 2026 across ~140 projects, global but US-heavy. - Only ~5 GW, roughly one in three, is actually under construction. - An estimated 30-50% of the 2026 pipeline is expected to slip into 2027 or later, with some cancelled outright. 𝙄𝙣 𝙥𝙡𝙖𝙞𝙣 𝙩𝙚𝙧𝙢𝙨: 𝙢𝙤𝙨𝙩 𝙤𝙛 𝙩𝙝𝙚 "𝙗𝙪𝙞𝙡𝙙-𝙤𝙪𝙩" 𝙘𝙪𝙧𝙧𝙚𝙣𝙩𝙡𝙮 𝙡𝙞𝙫𝙚𝙨 𝙤𝙣 𝙨𝙡𝙞𝙙𝙚𝙨, 𝙥𝙧𝙚𝙨𝙨 𝙧𝙚𝙡𝙚𝙖𝙨𝙚𝙨 𝙖𝙣𝙙 𝙡𝙖𝙣𝙙 𝙤𝙥𝙩𝙞𝙤𝙣𝙨 𝙧𝙖𝙩𝙝𝙚𝙧 𝙩𝙝𝙖𝙣 𝙥𝙤𝙪𝙧𝙚𝙙 𝙘𝙤𝙣𝙘𝙧𝙚𝙩𝙚. Or rather in the clouds, not on the Earth. Cloud AI computing, literally. 𝗧𝗵𝗲 𝗿𝗲𝗮𝗹-𝘄𝗼𝗿𝗹𝗱 𝗯𝗹𝗼𝗰𝗸𝗲𝗿𝘀 Grid interconnection queues running several years in major hubs (JLL flags 4+ years, and far longer in London and Amsterdam). Multi-year lead times on transformers, switchgear and battery systems. A backlash that's still building: cancellations roughly quadrupled in 2025 (25 projects vs 6 in 2024, per Baird/Heatmap), and at least 20 more (around 3.5 GW and $41.7 billion) were shelved in Q1 2026 following local opposition (Heatmap Pro, May 2026). 𝗠𝘆 𝗿𝗲𝗮𝗱 (𝗼𝗽𝗶𝗻𝗶𝗼𝗻, 𝗻𝗼𝘁 𝗳𝗮𝗰𝘁) I don't think the supercycle is cancelled. The demand pull is real, and inference workloads look set to grow hard from 2027, which keeps the silicon story $NVDA (NVIDIA Corporation) $AVGO (Broadcom Inc) firmly in the conversation. But the next two to three years will likely separate operators with executable, power-secured projects from those holding capital and a press release. The binding constraint has quietly shifted from chips and money to electrons, copper and permits, which is exactly where the electrical picks-and-shovels names like $VRT (Vertiv Holdings Co) $ETN (Eaton Corp PLC) sit. So: the biggest infrastructure lag in a generation, or are hyperscalers about to pull a rabbit out of the hat with on-site generation and SMRs like $GEV (GE Vernova LLC) $OKLO (Oklo Inc) ? I lean towards "lag now, catch-up later", but I'd like to read your take. 👇 Signal over noise, as always. Clouds, fog and mirrors, for now. Sources: Sightline Climate Data Center Outlook (Feb/May 2026), Bloomberg (Feb and May 2026), JLL 2026 Global Data Center Outlook, Heatmap Pro / Baird (Q1 2026). 𝗬𝗼𝘂𝗿 𝗰𝗮𝗽𝗶𝘁𝗮𝗹 𝗶𝘀 𝗮𝘁 𝗿𝗶𝘀𝗸 | 𝗖𝗼𝗽𝘆 𝗧𝗿𝗮𝗱𝗶𝗻𝗴 𝗱𝗼𝗲𝘀 𝗻𝗼𝘁 𝗮𝗺𝗼𝘂𝗻𝘁 𝘁𝗼 𝗶𝗻𝘃𝗲𝘀𝘁𝗺𝗲𝗻𝘁 𝗮𝗱𝘃𝗶𝗰𝗲 | 𝗬𝗼𝘂𝗿 𝗶𝗻𝘃𝗲𝘀𝘁𝗺𝗲𝗻𝘁𝘀 𝘃𝗮𝗹𝘂𝗲 𝗺𝗮𝘆 𝗴𝗼 𝘂𝗽 𝗼𝗿 𝗱𝗼𝘄𝗻

Not investment advice. The author may have financial interests in the mentioned instruments.