What Systematic Facility Delays Mean for US Cell Procurement Timelines

Year-end 2025 installed US lithium-ion cell manufacturing capacity was 295 GWh, up 56% during the year. Pipeline trackers push the number toward 1,172 GWh by 2035 if every announced project materializes at stated scale. Both figures measure something real. Neither answers the question a procurement team is actually asking: how many GWh it can contract from a domestic source, for its product, in 2027 or 2028.

No public source reviewed for this analysis reports a current US total for demonstrated annual cell output — recurring, saleable production as distinct from installed nameplate. The IEA's 2026 analysis separates the two concepts and finds that most plants take more than five years from start of operations to approach nominal output, but it publishes no discrete US production figure. As of late August 2026, the number that would let a procurement team calibrate domestic supply against program needs does not exist in the public record.

What does exist is facility-level evidence: production-start disclosures, workforce actions, construction milestones, conversions, and pauses. Read across the cohort, that evidence shows systematic schedule slippage. The rest of this piece traces what the pattern means for the decisions being made against it.

The slippage taxonomy

Across US cell manufacturing projects with announced nameplate above 20 GWh — 18 projects, plus Microvast Clarksville as a sub-scale comparator — original start-of-production dates have moved. They have moved in four distinguishable ways, and the distinction matters because each category implies something different about when a facility becomes contractable.

Straight delays are the simplest case. LGES Lansing was announced with GM for late 2024 at 50 GWh and declared production on August 18, 2026, nearly two years late, at a revised target above 35 GWh. Panasonic De Soto targeted mass production by March 2025 and opened in July 2025. Hyundai-SK Bartow County was announced for the second half of 2025; commercial production started June 2026. Construction delays, equipment commissioning, workforce training, and yield stabilization stack, and the resulting slips across the cohort run from four months at the short end to two years.

Indefinite stalls remove a facility from the timeline without supplying a replacement date. AESC Florence County announced 30 GWh for 2026 commercial operations. Construction has been suspended since June 2025, the plant is roughly 75% complete, and a $26.2 million contractor lien is on file. AESC Bowling Green announced 30 GWh for early 2025; the company's facilities page still classifies it as under construction. Microvast Clarksville suspended cell-line construction in Q2 2024 for insufficient funding and changed the intended chemistry from NMC to LFP. Combined: roughly 62 GWh of announced capacity with no current production date.

Conversions are the category that most consistently misleads capacity trackers, because the building still exists and the nameplate figure often survives in databases after the original product, customer, and timeline have been replaced.

BlueOval SK's Kentucky complex was announced as 86 GWh of EV cells for 2025. The first plant started EV-cell production in August 2025, Ford ended that production in December 2025, and the complex is being retooled for prismatic LFP storage cells targeting at least 20 GWh in late 2027. The original 86 GWh EV route is gone. L-H Battery Ohio was announced as 40 GWh for Honda EVs by end of 2025 and declared ESS cell mass production in July 2026, routed through LG Vertech, with no disclosed Honda allocation or uncommitted volume. Samsung SDI's New Carlisle plant was originally described as more than 30 GWh of NMC cells for GM in early 2026; Samsung acquired GM's interest in August 2026, said the plant would initially make storage cells, and gave no restart date, revised capacity, or customer.

A conversion preserves the structure and the equipment base. Customer qualification does not transfer with it. The sequence a facility has to walk — customer qualification, first accepted shipment, sustained production, field history — restarts at the beginning for the new product and the new buyer. For procurement purposes a converted facility is a new project that happens to have a roof.

Disrupted operations affect facilities that had already crossed into production. Ultium Warren began producing cells in August 2022 and shipped its 100-millionth cell in December 2024. After a seven-month shutdown, workers returned in August 2026, with production ramp expected after training. SK Battery America's Commerce, Georgia complex laid off 958 workers in March 2026 while pursuing new customers and storage applications. Output during an interruption is zero, and the ramp profile afterward is not predictable from the pre-shutdown run rate.

What a production-start declaration establishes

Four US facilities give publicly dated intervals from a company-declared production start to a first-shipment or delivery claim: Ultium Warren (about three months), Ultium Spring Hill (about one month), Panasonic De Soto (within the same fiscal quarter), and Hyundai-SK Bartow (within roughly six weeks). First shipments follow production declarations quickly.

None of those observations includes first-quarter shipment GWh, customer-accepted volume, stabilized yield, or uncommitted allocation. The interval from "production started" to "you can contract recurring qualified cells from this facility" is longer and harder to measure, and no public data series supports it. The IEA's five-year finding describes the far end of the ramp. The near end — the crossing from initial shipments to reliable contractable volume — is facility-specific, chemistry-specific, and customer-specific.

This publication's earlier inventory of US addressable cell supply found that no US facility publicly quantified cell supply that was simultaneously producing, facility-specific, and available to an unaffiliated buyer. Lansing, L-H Ohio, and Bartow have all declared production since. None has disclosed the fields that would change the finding.

A realistic 2027–2028 arrival picture

The procurement-relevant quantity is: GWh qualified for this product and customer, uncommitted to another program, deliverable within the decision window. Nameplate minus current production does not equal available capacity. Underutilized equipment may be reserved by an existing customer, restricted by qualification status, held for changeover, or bottlenecked upstream. With that caveat, the public record supports a rough sorting of the cohort by proximity to contractable supply.

The top five rows are facilities with 12 or more months of declared production by end of 2027, and they are the only ones where a new buyer could plausibly begin qualification work against stabilizing output. None has disclosed its current run rate, utilization, or open allocation. Below them sit the 2026 production declarations, early enough in ramp that qualification samples pulled today may not represent steady-state output; qualification work done against those samples may have to be repeated. A 2027 date announced by a facility that is not yet producing has the same evidentiary standing as the 2025 and 2026 dates that moved. The 2028-and-later tier is not supply for current planning purposes.

Decision implications

For procurement: stress-test the dual-sourcing assumption against the actual candidate pool. If a program depends on two or more qualified domestic cell suppliers by 2028, the facility evidence says the set of candidates with enough operating history to support parallel qualification is smaller than nameplate figures imply. Specify the contingency now: what the program does if only one domestic source qualifies on time. Contract structures should price ramp uncertainty explicitly. Volume-commitment schedules indexed to a facility's announced nameplate are mismatched to how these plants actually come online; schedules that allow volume flexibility through the first 18 to 24 months of operation, with increases triggered by demonstrated output rather than by calendar date, fit the observed pattern better.

For strategy: a constrained supplier pool changes what an existing qualification is worth. If the realistic set of qualified domestic suppliers in 2027–2028 is three to five facilities, and several of those carry captive or partially captive customer relationships, companies competing for domestic allocation are not competing in a market with surplus supply. Make-versus-buy looks different when "buy domestic" means competing for allocation from a handful of ramping plants. The cost of starting a new supplier qualification in 2027 includes the 12-to-18-month qualification cycle and the risk that the facility's uncommitted volume is absorbed by a buyer who started the same process a year earlier. Teams modeling 2028–2029 programs against announced domestic capacity should model them instead against the qualified-and-uncommitted subset, which is a fraction of nameplate that no public source currently quantifies.

For investment: delay costs compound against a fixed compliance calendar. The 45X advanced manufacturing production credit carries material assistance cost ratio thresholds — the share of a battery component's value that must come from manufacturing outside a foreign entity of concern — and those thresholds escalate on a set schedule: 60% in 2026, 65% in 2027, 70% in 2028, 80% in 2029. Facility schedules do not track that calendar. A cell line that was supposed to contribute domestic content in 2026 against a 60% threshold, but does not produce until 2028, meets a 70% threshold instead. Each year of slippage costs more in compliance difficulty than the year before it, because the bar moved while the facility did not. IRS Notice 2026-15 provides interim calculation and certification procedures; the safe-harbor tables and additional effective-control guidance remain pending against a statutory deadline of December 31, 2026. Project finance models whose economics depend on domestically sourced cells — for credit capture, for compliance, or because a customer requires it — should carry a 12-to-18-month delay scenario consistent with the cohort's observed range, and that scenario should use the threshold in effect at the delayed arrival date rather than the original one.

The substitution to stop making

Qualification, sourcing, contract, and financing decisions are being made against a capacity picture that treats announced nameplate as a proxy for available supply and announced start-of-production as a proxy for procurement-ready output. The facility record supports neither substitution. Domestic cell capacity is arriving; several plants are producing, and the 2025–2026 cohort is real. But it is not the pipeline number, on the pipeline schedule, with open allocation for whoever wants it. Plan against the arrival schedule the evidence supports, and build contract and financial structures that absorb the gap between announcement and delivery.