AI Picks and Shovels (Arc 3: Keeping it Alive): Bloom Energy (NYSE: BE) — The Backup Generator

Picks and Shovels Series | Arc 3: Keeping it Alive | Part 3 of 3

Last week Constellation contracted the nuclear baseload that keeps the grid running. This week the individual data-centre site still has to connect to something.

Nvidia designs the brain. Arc 1 covered the foundry, the custom silicon, and the merchant challenger that try to turn that design into chips. Arc 2 covered the switching fabric that moves the traffic, the memory that holds the weights, and the interconnect landlord that houses the racks. Arc 3 has covered the cooling loop that stops the rack melting and the nuclear fleet that supplies the electrons at grid scale. The remaining gap sits between the utility and the fence line.

AI data centres do not wait for the average interconnection study. They wait for the hour the GPUs can be powered. In parts of the United States the queue for a new transmission interconnect now stretches years. Solid-oxide fuel cells convert natural gas (or biogas, or hydrogen) into electricity at the customer’s site. They do not require a new transmission line for the initial capacity. They do not wait for a new substation. They scale in modular blocks. For an operator whose revenue clock starts when the racks are live, the relevant comparison is not the all-in cost of grid power versus fuel-cell power. It is the lost revenue from a quarter’s delay versus the cost of the systems that can be delivered in months.

Bloom Energy (NYSE: BE) sells the on-site fuel-cell systems that let a hyperscaler or neocloud operator (specialist AI-cloud builders) start drawing power without waiting for the utility to build the line.

The Modular Stack

Bloom’s solid-oxide fuel cells operate at high temperature and high efficiency relative to combustion generators. Units are small and stacked into multi-megawatt installations that sit behind the customer’s fence. On-site placement keeps the initial capacity out of the transmission interconnection queue. Planning approvals and delivery have cleared in months: an Oracle site in Utah received power in less than two months; a Nebius project in New Jersey received planning approvals within three months of contract signature. Equinix has deployed more than 100 megawatts across its footprint. CoreWeave has used the systems at an Illinois site. These are commercial volumes.

Oracle has indicated intent to procure up to 2.8 gigawatts of Bloom systems, of which 1.2 gigawatts is already contracted. AEP and Google have a 700-megawatt carbon-capture project that will use Bloom servers. Management has cited customer statements that a quarter’s delay on a 100-megawatt data centre can represent roughly US$1 billion of lost revenue. On that customer framing, the fuel-cell cost is small relative to the opportunity cost of waiting for the grid.

The Numbers and the Reaction

Third-quarter 2026 revenue reached US$1.24 billion, up 138% year-on-year. Adjusted earnings per share of US$0.94 beat consensus. Full-year revenue guidance was raised to US$4.4–4.6 billion. The stock still fell more than 11% on the day after the report, closing near US$193. Investors focused on negative free cash flow of roughly US$180 million in the quarter and the capital required to scale production toward multi-gigawatt annual capacity. Market capitalisation sits around US$57 billion — roughly 12–13 times the midpoint of the raised full-year revenue guidance. Trailing and forward earnings multiples remain elevated relative to the broader industrial and power equipment group. Beta is high.

The revenue growth is real. The customer list is concentrated among the same hyperscalers and neocloud operators that appear throughout this series. Cash conversion has not yet matched the top-line expansion. Scaling manufacturing capacity to multi-gigawatt annual output requires capital. That capital is visible in the free-cash-flow number and in the post-earnings price reaction.

Competition and the Constraint

Gas turbines from suppliers such as GE Vernova and Siemens Energy can also be placed behind the fence, but order backlogs have lengthened as the same data-centre demand has pulled capacity; multi-year lead times are now common for larger units. Reciprocating gas engines and diesel gensets install faster but run at lower efficiency and higher emissions. Other fuel-cell makers exist at smaller scale. Grid interconnection remains the slowest path for large new load in many US regions. Bloom’s differentiator is the combination of modular on-site generation, demonstrated commercial-scale deployments with named hyperscalers, and the ability to start drawing power in months rather than the multi-year timeline of a new transmission interconnect or a constrained turbine slot. The practical constraint is no longer whether the technology works. It is the rate at which manufacturing capacity can be built and the durability of the stacks once installed. Stack replacement and service costs are real. The systems still consume natural gas. Energy-transition politics around gas-fed generation remain contested.

The Aussie Angle

Australia has no operating nuclear plants and limited near-term policy path to them. The same power bottleneck that has driven US interconnection queues is already visible in Australian data-centre planning and industrial load growth. Grid connection timelines for large continuous loads in Sydney and Melbourne are commonly measured in years; firming costs are higher in a system with a larger renewable share and fewer dispatchable baseload options. On-site generation is one of the tools available when the queue is long.

For an Australian investor the usual overlays apply. The equity can be held directly via an international broker or indirectly through US indices and superannuation vehicles. Any capital gain or loss is measured against an AUD cost base, so the AUD/USD rate on entry and exit matters. There are no franking credits. Dividends, if paid, are subject to US withholding tax (typically 15% under the treaty with a valid W-8BEN). Holding Bloom does not change the domestic reality of Australian grid connection queues. It does expose the holder to the US dollar and to the AUD cost base on any capital gain. The fuel cells sit next to a US data centre. The tax return is in Australian dollars.

The names that sit inside the AI cluster — whether held directly or through the S&P 500 or Nasdaq exposures common in Australian super and SMSF portfolios — now carry an implicit dependence on the speed at which power can be delivered. A stock such as Bloom is a pure expression of that constraint. Its returns will be driven by the rate at which customers choose on-site generation over waiting for the grid, and by the cost of capital required to build the manufacturing capacity. Those returns will also be measured in Australian dollars.

The Investment Case — In One Line Per Side

Bull case: Time-to-power has become a scarce resource and Bloom is one of the few suppliers that can deliver modular on-site generation at commercial scale without a multi-year grid queue. Customer statements frame a quarter’s delay on a 100-megawatt site as roughly US$1 billion of lost revenue, making the fuel-cell cost small relative to the opportunity cost.

Bear case: The systems still consume natural gas, stack replacement and service costs are real, energy-transition politics around gas-fed generation remain contested, cash conversion has not yet matched the revenue growth, and the valuation already prices a large share of the AI power narrative at roughly 12–13 times guided full-year sales.

The honest read: On-site fuel cells solve the problem of a cluster that needs electrons behind the fence on a timeline shorter than a transmission upgrade. The valuation assumes manufacturing capacity scales without major slippage, that customers continue to prefer the speed of fuel cells over waiting for the grid or competing on-site options, and that free cash flow eventually follows the revenue. That is a high bar, not an impossibility.

🏁 Nine Pieces, One Anatomy

This is the final piece of a nine-part series. It is worth pausing to say what these nine businesses have in common — and what they reveal about how an AI cluster actually gets built and kept alive.

• Taiwan Semiconductor: the only factory that can build the frontier chips at scale.
• Broadcom: the co-design firm that turns hyperscaler architectures into manufacturable silicon.
• AMD: the merchant challenger selling a general-purpose GPU off the shelf.
• Arista: the switching fabric that turns a warehouse of GPUs into one training run.
• Micron: one of three suppliers of the high-bandwidth memory the cluster needs to hold context.
• Equinix: the building that houses the racks and the neighbours they talk to.
• Vertiv: the thermal systems that stop a 140-kilowatt rack from melting the chip.
• Constellation: the nuclear fleet that contracts carbon-free baseload to the same labs that buy the racks.
• Bloom Energy: the on-site fuel cells that start drawing power in months when the grid queue stretches years.

None of these are the chip itself. Every one of them sits on a physical or contractual constraint that turns a design into a running system — a foundry that must be booked, a fabric that must switch, a building that must have power and cooling, a queue that must be bypassed.

That is the pattern. And it matters more than the next earnings print or the latest capacity announcement.

The gap between what Aussie investors think they know about the AI build-out and what is actually required to keep a cluster alive is wide. It costs real money over time — in concentrated positions that ignore the power bottleneck, in portfolios that treat the S&P 500 or Nasdaq exposure as pure semiconductor beta, in tax returns measured in Australian dollars on a US-dollar cost base.

That gap is what this publication exists to close.

Data Sources:

• Bloom Energy Q3 2026 results and full-year guidance (revenue US$1.24 billion, adjusted EPS US$0.94, guidance US$4.4–4.6 billion)
• Oracle, Nebius, Equinix, CoreWeave, and AEP/Google customer announcements and deployment timelines
• Management references to customer statements on revenue impact of interconnection delays
• Share price, market capitalisation (approximately US$57 billion), and free-cash-flow figures as of the Friday close prior to publication
• Company filings and investor materials on solid-oxide fuel-cell technology, stack life, and capacity expansion plans
• Public commentary on gas-turbine lead times (GE Vernova, Siemens Energy) and interconnection queue data

Wall St. Down Under | Australia

Subscribe | wallstdownunder.com.au

Disclaimer: Wall St. Down Under is an independent financial newsletter for informational and educational purposes only. Nothing published here constitutes financial advice, a recommendation to buy or sell any security, or a solicitation of any investment decision. Always do your own research and consult a licensed financial adviser before making investment decisions. Australian investors should consider their own financial situation, objectives, and risk tolerance. Past performance is not indicative of future results.