On-site solid-oxide fuel cells — "bring-your-own-power" generation that bypasses the grid for AI data centers (Oracle Project Jupiter, 55-day deploy)
AI exposure 55% — derived from reported segment revenue, not asserted.
Physical Infrastructure → Power & Electrical Equipment · figures rebased to Q2 2026, reported 2026-07-28
On-site fuel-cell power-generation equipment a data center installs = B2 Power & Electrical Equipment (peer to gensets/standby power), not a utility/fuel commodity
Bloom Energy ($1,065.4M Q2 2026, +165.5% YoY from $401.2M — a record quarter with FY2026 guidance raised to about $4.2B). Product 87.8% ($935.4M, +215.4%) = primary B2 Power & Electrical Equipment — solid-oxide fuel-cell servers being bought as on-site data-centre generation; Service, installation and electricity 12.2% (service $69M, installation $51M, electricity $9.95M, -22.3%). The product line more than tripled while the electricity line shrank, so Bloom is now selling boxes rather than power.
Moat. First-mover scale in solid-oxide fuel cells + a recurring catalyst-service annuity + hyperscaler validation (Oracle/Brookfield); manufacturing + reliability (99.9%+) lead, though fuel cells face turbine + grid alternatives.
Bottleneck / pricing power. Real pricing power from the time-to-power bottleneck — large-load interconnection queues run 5-7 years in PJM, MISO and CAISO while Bloom deploys in ~55 days — and Q2 2026 proved it monetises: revenue +165% with gross margin EXPANDING 668bps to 33.4%, i.e. taking price rather than buying growth. Treat the ~$20B backlog figure with care: it is unaudited and largely framework, against ~$492M of audited binding performance obligations. Conversion, not the headline, is the evidence.
Role in the AI stack. The on-site (behind-the-meter) primary-power-generation layer — bypasses the grid.
All 174 companies · What moved this quarter · How the map is built