Bloom Energy — Manufacturing Capacity Deep Dive

1. The capacity story timeline

The capacity story has moved fast. Here is what management has actually said, in chronological order, with sources.

2. The physical facilities

Bloom's 2025 10-K gives a newer and better facility map than the March 2024 press release. The company's primary manufacturing facilities are in Fremont, California and Newark, Delaware; it also maintains a full-assembly facility in Korea through the SK ecoplant joint venture.^[11]

The key footprint math: excluding headquarters, Bloom discloses 958,000 square feet of principal U.S. manufacturing / R&D / warehousing footprint (326,000 Fremont + 454,000 Newark leased + 178,000 Newark owned). Including headquarters, the principal property footprint is 1.141 million square feet.^[11] That is materially larger than the 524,000 sq ft figure in the March 2024 press release, but square footage is not the same thing as line output.

The exact split of the 5 GW claim across Fremont, Newark, and Korea is not disclosed. The only clean public split is that Fremont is targeted for 2 GW by end-2026. Newark clearly matters because it is the Energy Server assembly base and owns the copy-exact expansion template, but the company does not disclose Newark's current GW output. Korea adds optional capacity, but counting it inside the 5 GW number is an inference, not a disclosed fact.

3. What capex tells us — the most concrete evidence

If the 5 GW number required a traditional new-factory buildout, it should show up as a large PP&E spike. It does not. Bloom's PP&E purchases have been modest relative to revenue, though Q1 2026 did show a step-up as the company kept investing in Fremont and Delaware capacity.^[3][12]

This has three significant implications:

1. Bloom can probably expand further at low incremental capex. If the current scaling model holds, the balance sheet is unlikely to be the binding constraint on additional capacity. The harder constraints are supplier qualification, customer timing, manufacturing yield, field installation labor, and whether demand stays strong enough to justify adding lines.
2. The cost-down curve is more credible. Low capex intensity per GW means each new MW shipped does not require a proportional fixed-cost recovery. Operating leverage is more durable.
3. The "copy exact" model is not just call rhetoric. The 2025 10-K says the owned Newark facility was designed for copy-exact duplication, and the Q1 2026 call says that model underpins capacity beyond 5 GW.^[11][1]

4. The Korea JV — Bloom SK Fuel Cell

This is the part of Bloom's manufacturing footprint that gets the least attention but matters for two reasons: it adds to total capacity, and the JV is a source of the equity-method losses that drag down GAAP net income.

• Established: January 2020 as a joint venture between SK Ecoplant (Korean construction conglomerate) and Bloom Energy^[6]
• Location: Gumi, North Gyeongsang Province, South Korea
• Construction completed: October 2020
• Products: Solid oxide fuel cells (SOFCs) and, more recently, solid oxide electrolysis (SOEC) equipment for Asian markets
• Capacity timeline: originally planned 50 MW (2021) → 200+ MW (post-2025); revised in October 2021 agreement to 200+ MW starting from 2023^[6]
• Current output: not publicly disclosed in recent quarters

Why this matters for the manufacturing picture: the Korea JV adds capacity that doesn't show up in Bloom's US capex figures (it's funded jointly by SK). If the "5 GW" claim implicitly includes this facility scaled to several hundred MW, the actual US-owned-and-funded capacity may be closer to 4.5 GW. This doesn't change the demand-side conclusion but matters for understanding Bloom's actual financial commitments.

The downside of the JV structure is that Bloom's share of JV losses ($17-21M per quarter recently) flows through equity-method accounting and reduces net income — even though the joint manufacturing is helping fulfill demand. See the main BE summary for the GAAP vs net income reconciliation.

5. What the actual bottleneck is now

If management is telling the truth and Bloom is at 5 GW capacity with no constraint, then the binding limits on revenue growth become demand-side, not supply-side. Per CEO KR Sridhar on the Q1 2026 call:

"Today, we are not order constrained and not capacity constrained. The pace of our revenue growth is decided by how fast our customers can build their greenfield sites." — K.R. Sridhar, CEO, Bloom Energy Q1 2026 earnings call (April 28, 2026)^[1]

Translated: the bottleneck has moved from "Bloom can't build them fast enough" to "the hyperscaler can't break ground on the data center fast enough to install them." That includes:

• Site preparation — land acquisition, permits, foundation work, utility connections (typically 6-18 months)
• Greenfield civil construction — pads for the Energy Server units, electrical infrastructure
• Customer financing and approval cycles — board approvals on multi-hundred-million-dollar orders
• Long-lead-time components Bloom doesn't make — power electronics, control systems, rare-earth elements, specialty alloys, and third-party processes such as compaction, sintering, brazing, and specialty-material manufacturing^[11]

The Oracle deal is the best concrete evidence of this. Oracle signed a master agreement for up to 2.8 GW, with an initial 1.2 GW contracted and deployment "underway and continuing into next year."^[7] Separately, on the Q1 2026 call, Sridhar described Oracle's Project Jupiter as an up-to-2.45-GW power block using Bloom servers.^[1] These are multi-year site deployments; Bloom is saying customer site readiness, not factory output, is the pacing item.

6. Where to push back on the 5 GW claim

This is the part of the analysis where honesty matters most. The 5 GW figure is management's own statement on an earnings call, repeated in transcripts. It has not been independently verified by a tour, an audit, an analyst day with a factory walk-through, or a third-party engineering review.

Reasons to apply skepticism:

1. The jump from 2 GW (Aug 2025) to 5 GW (Apr 2026) is dramatic. Eight months of construction time is not a normal industrial buildout. The most generous explanation is that the 2 GW number was conservative and the 5 GW reflects automation/yield/shift expansion that does not require physical construction.
2. Annual revenue is still tracking to "only" $3.4-3.8B in 2026.^[13] At $3M/MW system price plus service/installation, that is roughly 1-1.2 GW of shipments. If capacity is truly 5 GW, the utilization rate is 20-25%. That is unusually low for a company with a multi-year backlog, suggesting that either (a) the 5 GW number is "nameplate" rather than achievable run-rate, or (b) Bloom is deliberately running below full capacity to fulfill the demand-pacing constraint.
3. "Nameplate capacity" in industrial manufacturing typically requires 18-24 months to ramp to from a cold start. Even if Bloom has the physical lines and equipment for 5 GW of output, achieving that throughput requires worker training, supply chain ramp, qualification testing, and yield optimization. Real-world capacity often runs 60-70% of nameplate for the first two years.
4. No independent verification. I found no public factory audit, analyst-day walkthrough, or third-party engineering report verifying 5 GW. The firmest physical evidence is the 2025 10-K footprint: 958,000 sq ft of principal U.S. manufacturing / R&D / warehousing footprint excluding headquarters. That supports several-GW plausibility, but it does not prove 5 GW practical output.
5. Bloom has a history of optimistic projections that have slipped. The Korea JV was originally planned to hit 200+ MW after 2025; the timeline was revised forward in 2021 and the current actual output is not disclosed in recent quarters.^[6]

7. Implications for the 2029 cash flow model

The question tree's bear case assumed Bloom would cap at 1-1.5 GW/yr of shipments through 2029 because of manufacturing constraints. That bear case is now too conservative. The actual cap is probably 3-5 GW of shipments depending on which figure you believe.

Updated framing for the 2029 manufacturing question:

At $3M/MW blended hardware price plus service/installation, the 2029 revenue ranges become:

These numbers are higher than the question tree's original ranges by roughly 50% across the board. That doesn't change the fundamental shape of the asymmetry analysis — the market is still pricing the bull/lottery scenario — but it does mean the base case is now more defensible at a higher revenue figure than I previously modeled.