How the IRA Is Reshaping Battery Manufacturing in the United States

The IRA pushed the U.S. industry toward lithium iron phosphate cathode chemistry. FEOC compliance is simpler with LFP because it contains no cobalt, nickel, or manganese. The fixed 45X credit covers a larger fraction of LFP's lower cell cost. The strategic logic is obvious and has been obvious since 2023. What has not been obvious, and what is only now becoming visible as licensed LFP plants attempt to ramp production in the United States, is how profoundly difficult it is to manufacture LFP cathode material well.

The iron phosphate olivine structure has terrible electronic conductivity. Every commercial LFP cathode particle must be coated in a thin carbon shell, roughly two to four nanometers, to function in a cell. This is the step where the entire chemistry either works or does not, and it is the step where the gap between documented process knowledge and accumulated manufacturing intuition is widest.

Carbon coating happens in a rotary kiln or a rotary furnace. The precursor (usually iron phosphate with a lithium source and a carbon source like glucose or sucrose) enters one end, travels through a heated zone under controlled atmosphere, and exits the other end as coated LFP particles. The carbon source decomposes during transit, depositing a carbonaceous layer on the particle surfaces. The thickness and uniformity of that layer depend on temperature profile along the kiln axis, rotation speed, atmosphere composition (nitrogen, with precise control of residual oxygen to prevent oxidation of Fe²⁺ to Fe³⁺), residence time, and the physical characteristics of the precursor feed (particle size distribution, moisture content, degree of pre-mixing).

These variables interact nonlinearly. A batch of precursor with slightly coarser particle size distribution presents less total surface area per unit mass. The same carbon source loading produces a thicker coating on each particle. If the coating exceeds roughly four nanometers, lithium-ion diffusion through the carbon shell slows enough to degrade rate performance. The operator response is to reduce carbon source loading or shorten residence time. But shorter residence time means less complete decomposition of the carbon source, which can leave organic residues that increase cell impedance. So the operator adjusts furnace temperature upward to compensate, which risks pushing Fe²⁺ toward Fe³⁺ if the atmosphere control is not correspondingly tightened. Each adjustment propagates.

CATL and BYD learned to navigate these interactions across more than a decade of continuous production, processing millions of kilograms of precursor through furnaces that aged and drifted and were retuned and replaced. The knowledge that accumulated is not primarily in the form of written procedures. It is in the form of operator heuristics: recognizing a color change in the kiln discharge that indicates incomplete carbon decomposition, knowing that a particular precursor supplier's material runs 0.3 microns coarser in summer months due to humidity effects on their milling process, understanding that after a kiln refractory replacement the temperature profile shifts for the first 72 hours and the residence time needs to be extended by roughly 8 percent to compensate. None of this is secret. It is just not written down, because the people who accumulated it never had reason to write it down. It accreted in the daily practice of running production.

A licensing agreement transfers the documented portion. Equipment specs, target temperature profiles, atmosphere setpoints, carbon source ratios, particle size specifications. A competent engineering team can set up the equipment and hit the setpoints. The cells produced in the first months will work. They will also exhibit higher defect rates, wider performance distributions, and lower first-pass yield than the licensor's own output, because the documented process represents the central tendency of a production system whose actual performance depends on continuous small adjustments that the documentation does not capture.

The yield gap between a licensed LFP plant and the licensor's home factory is not a fixed quantity. It narrows over time as the new plant's operators build their own experience base. The rate of narrowing depends on production volume (more cells means more data points means faster statistical learning), on the quality of the plant's process data infrastructure (whether operators can trace a finished-cell defect back to a specific process excursion three weeks earlier on the coating line), and on whether the licensor's technical support is structured to accelerate learning or to maintain dependency. That last variable is worth dwelling on. CATL's licensing agreement with Ford includes ongoing technical service fees. Every month that the Michigan plant needs CATL's process engineers to troubleshoot a coating quality issue is a month of fee revenue. The incentive to provide excellent support coexists with the incentive to provide support slowly enough that dependency endures. These incentives do not conflict in the short term, which is what makes the dynamic stable and hard to disrupt. CATL will be genuinely helpful. CATL will not be in a hurry.

SK Innovation, Panasonic, and Samsung SDI face a different version of the same problem at their own U.S. subsidiaries. The parent company's Korean or Japanese engineers run critical production steps during ramp. They are competent. They are also, at the individual level, being asked to train their own replacements on a management-imposed timeline, and the thoroughness of that training varies with individual temperament, language barriers, documentation habits, and the degree to which the engineer's job security back home depends on the U.S. plant continuing to need support from headquarters. Some engineers transfer knowledge generously. Some transfer the procedure but not the reasoning behind the procedure, which means the American operator can follow the recipe in normal conditions but cannot diagnose or recover from a novel excursion. The difference between these two outcomes unfolds over months and is essentially invisible to management until something goes wrong on the production line at 2 AM and the locally trained shift supervisor either handles it or calls Korea.

Formation is the manufacturing step that determines cell lifetime, and it is the step where new plants are most exposed to the knowledge gap because the feedback loop is physically slow.

A newly assembled cell contains electrode materials, electrolyte, and a separator, but it is not yet a functional energy storage device. The first charge decomposes a thin layer of electrolyte on the anode surface, forming the solid electrolyte interphase. The SEI is a passivation film whose composition, density, ionic conductivity, and adhesion to the graphite anode surface determine how the cell will age for its entire service life. Capacity fade rate, impedance growth, self-discharge, vulnerability to lithium plating during fast charge: the SEI quality imprints all of these characteristics during formation.

The formation protocol specifies a sequence of charge and discharge steps with defined current rates, voltage limits, temperature setpoints, and rest periods. A typical protocol takes two to three weeks to execute per cell. The protocol interacts with everything upstream. Electrode porosity (set during calendering) affects electrolyte infiltration, which affects how uniformly the SEI forms across the electrode surface. Electrolyte composition determines which decomposition products constitute the SEI. Electrolyte fill volume and the wetting time allowed before the first charge affect whether the electrolyte has fully penetrated the electrode pores. Even the ambient temperature in the formation chamber during rest periods influences the SEI morphology.

A protocol that produces excellent SEI quality on cells from CATL's Ningde factory, where the electrode porosity distribution reflects CATL's specific calendering equipment and setup, the electrolyte is mixed by CATL's specific supplier to CATL's specific composition tolerance, and the fill volume is controlled by CATL's specific filling equipment with CATL's specific nozzle geometry, will not necessarily produce the same SEI quality on cells from the Ford-CATL Michigan plant, even though the bill of materials is nominally identical, because the upstream process variation at the Michigan plant is different. Different calendering equipment (or the same model with different wear characteristics) produces a different porosity distribution. Different electrolyte filling equipment produces a different fill volume distribution. Different ambient temperature ranges in a different building in a different climate produce different rest-period thermal profiles.

Tuning the formation protocol to a new plant's specific upstream variation is an iterative experimental process. Change one parameter (say, the initial charge rate), run a batch of cells through the full two-to-three-week formation sequence, perform post-formation electrical characterization, put a subset of cells on calendar aging, and wait weeks or months for the aging data to reveal whether the SEI quality improved or degraded. Each iteration of this loop consumes weeks at minimum. Reaching a well-optimized protocol for a specific plant's cell population takes six to twelve months of sustained experimentation, assuming adequate engineering resources and a robust data pipeline connecting formation outcomes to upstream process records.

During those months, production cells are being formed with a protocol that is known to be suboptimal. The cells pass initial electrical testing. They ship. They go into vehicles or storage systems. Whether they will meet their rated cycle life is uncertain, and that uncertainty will not resolve for years. The first customers of every new battery plant are, in effect, absorbing a quality risk that is invisible at the point of sale and becomes apparent only after thousands of charge-discharge cycles in the field. This is not fraud. It is the unavoidable consequence of the physics of SEI aging combined with the economics of factories that cannot afford to delay shipment for a year while formation protocols are optimized.

Formation equipment is also the largest capital and floor space allocation in a cell factory, often exceeding the combined footprint and cost of the coating and assembly lines. A 30 GWh plant needs millions of cell slots occupied simultaneously for weeks each. The capital tied up in formation chambers, the building space to house them, and the electrical infrastructure to power them exceeds $500 million at that scale. Discussions of gigafactory capital costs that focus on coating lines and cell assembly equipment are discussing the minority of the expenditure.

The Foreign Entity of Concern provisions restrict 30D eligibility for vehicles with batteries containing FEOC-sourced minerals or components. The compliance architecture this has generated (licensing structures, tolling agreements, corporate restructuring to distance nominal ownership from Chinese entities) has been extensively documented in trade press and legal commentary. The part that has not been adequately examined is the human dimension.

The knowledge that new American battery plants need most urgently is held by mid-level technical staff at Chinese battery companies. Production supervisors who have run LFP coating lines through five years of process variation. Quality engineers who have traced thousands of cell defects back to specific upstream causes. Equipment maintenance leads who know the behavioral quirks of specific machine models under specific operating conditions. These people are not executives who sign licensing deals. They are not in anyone's press release. They hold the tacit knowledge that makes the difference between a plant that produces cells and a plant that produces cells at 93 percent yield.

They are staying in China. Their employers are reluctant to deploy them to American facilities in the FEOC environment. The technology leakage risk is real: process knowledge that leaves through an individual employee's head is harder to control than knowledge that leaves through a documented license. The employees themselves face personal calculations that mostly resolve against relocation. Compensation at a new U.S. plant is comparable to what they earn at home, but the cost of living is higher, the family disruption is significant, the visa situation introduces career uncertainty, and the role at the American plant is to make themselves unnecessary, which is not an attractive job description.

The FEOC framework thus restricts the flow of the specific expertise that would most accelerate the ramp of the factories it incentivizes. The licensed process documents cross the Pacific. The judgment that makes those documents effective production tools stays at the originating facility. This is not a flaw in the FEOC design that could be fixed with a carve-out. It is an inherent tension in a policy that simultaneously restricts Chinese participation in the U.S. battery chain and depends on Chinese-origin knowledge to make that chain functional.

Cell manufacturing requires process disciplines analogous to semiconductor fabrication: controlled environments, tight tolerances, statistical process control, root-cause analysis of defects that manifest far downstream of their origin. The U.S. has almost no existing labor pool with these skills applied to electrochemical manufacturing.

The CHIPS Act is pulling from the same labor categories simultaneously. Intel in Ohio, TSMC in Arizona, Samsung in Texas, Micron in New York, GlobalFoundries in Vermont. Process technicians, cleanroom operators, quality engineers, equipment maintenance specialists. The salary inflation in regions where battery and semiconductor hiring overlap has reached 15 to 25 percent for these roles as of 2025. Two policies designed to rebuild American manufacturing in different sectors are competing for the same workers, and neither policy's labor models accounted for the other's demand. The constraint is acute in specific geographies and specific skill tiers, and it will worsen through 2027 as both buildouts accelerate.

Cell production equipment comes from a small number of specialized manufacturers in Japan, Korea, and China. Lead times exceed eighteen months. There is no significant U.S. manufacturing base for coating lines, calendering machines, electrolyte filling systems, stacking equipment, winding equipment, or formation chambers. The IRA's domestic battery supply chain depends entirely on imported Asian equipment. The supply chain dependency did not disappear. It shifted from finished cells to the machines that make cells.

The Battery Belt concentrated in the humid Southeast and upper Midwest for rational reasons: automotive customer proximity, state incentives, labor pools, logistics. Cell manufacturing dry rooms maintained below negative 40°C dewpoint consume 8 to 12 percent of total production cost in humid climates and 3 to 5 percent in arid climates. The 45X credit pays the same per kWh regardless of location. The site selection criteria that drove the Battery Belt's geography did not weight operating thermodynamics, and the resulting cost penalty is permanent for the life of each building. It is survivable within the credit window. Whether it is survivable after the credit phases down depends on how much yield improvement and other cost reductions the plant achieves in the interim.

Cathode and cell manufacturing attracted the investment. The supply chain between them did not.

Battery-grade graphite (natural and synthetic) processing is overwhelmingly Chinese. Domestic anode material capacity is negligible relative to cathode capacity under construction. Spheronization yields for natural graphite run 30 to 40 percent. Synthetic graphite requires graphitization above 2800°C with major energy and permitting implications. Syrah Resources' Vidalia plant is the most prominent non-Chinese anode project in the U.S. and has had a difficult ramp. A cathode plant without matching anode supply ships nothing.

LiPF6 electrolyte salt synthesis requires hydrogen fluoride handling. HF permitting in the U.S. adds years to project timelines. Domestic production is minimal. Cell plants will import LiPF6.

Separator manufacturing (biaxial stretching of PE or PP to 12-25 micron thickness at sub-micron tolerances) is controlled by Asahi Kasei, Toray, SK ie Technology, and Senior Technology. No major U.S. production announced. A separator pinhole causes thermal runaway. The quality bar is absolute and the process expertise is deep and concentrated.

The 45X credit structure incentivizes cells, modules, and electrode active materials. It does not meaningfully address electrolyte salt, separator, or anode graphite processing, because the specialized manufacturing barriers (HF chemistry, sub-micron polymer processing, low-yield mechanical milling and high-temperature graphitization) make the domestic business case unattractive at current incentive levels. The "domestic" supply chain being constructed under the IRA has import-dependent middle layers that the policy does not cover.

Section 48E created a storage market large enough to absorb cells that fail automotive quality specifications. Automotive OEMs require tight cell-to-cell matching. Storage integrators accept wider variation. Below-spec cells that would otherwise be scrapped can flow to storage projects as a second revenue stream. Several manufacturers are designing quality systems around this tiered model. The practical effect is to improve the economics of plants that have not yet reached automotive-grade yield consistency by ensuring that even imperfect output has a customer. Without 48E, that output is waste.

The IRA's recycling provision (domestically recovered minerals count as U.S.-sourced for 30D compliance) turned recycling into an immediate supply chain tool. Current feedstock is predominantly manufacturing scrap from new plants rather than end-of-life vehicle batteries, because the EV fleet is too young. Recyclers building adjacent to gigafactories are capturing waste from facilities that have not yet learned to run efficiently. Feedstock volume correlates inversely with the factory's yield. As yield improves, scrap drops, and the recyclers' input stream thins until end-of-life volumes arrive at scale in the mid-2030s.

The credits are statutory. Repeal requires legislation, and once factories operate in enough congressional districts the political cost is high. The realistic vulnerability is implementation inconsistency: shifting Treasury FEOC interpretations, changing IRS audit postures, delayed DOE disbursements. A gigafactory needs five to seven years from site selection to full production. Regulatory ambiguity during peak construction phase does real damage even if the underlying credits remain law.

Permitting is the constraint money cannot fix. Lithium mines, graphite processing plants, LiPF6 facilities: all require environmental review and local approval on timelines the credit window does not accommodate.

Yield is the variable that matters most. A plant at 93 percent first-pass yield is competitive with credits and plausibly viable without them. A plant stuck at 85 percent survives only because credits cover the cost of scrapped material. The 45X phasedown beginning in 2030 is the moment when the distinction between a competitive factory and a subsidy-dependent factory becomes financially lethal. Every plant that has not reached the low 90s by then faces a margin crisis that no amount of policy advocacy can resolve, because yield is a manufacturing problem and manufacturing problems are solved on the production floor, not in Washington.

The staffing ratio of expatriate technical personnel to locally trained staff in critical production roles, measured at individual plants three years after the first cell ships, will indicate whether knowledge transfer succeeded or stalled. The data will never be published. The companies have every reason to obscure it. But it is the number that determines whether the IRA built domestic manufacturing capability or financed Asian-operated production capacity on American soil. The distinction matters because the first is self-sustaining and the second requires permanent subsidy or permanent dependence on foreign technical support, either of which defeats the policy's stated purpose.

The construction is real. The investment is enormous. The policy architecture is sophisticated. The question the IRA cannot answer, because it is not a question policy can answer, is whether the United States can learn to make battery cells at competitive quality and cost fast enough for the learning to compound into self-sustaining capability before the training wheels retract. Cell manufacturing is a domain where competence accrues through sustained repetition under conditions of relentless process variation, and the operators who will determine the answer are right now being trained on production lines that are themselves still being tuned, using equipment that arrived late, in dry rooms fighting the wrong climate, by expatriate engineers who will eventually go home. Whether what they leave behind is sufficient is the open question of American industrial policy in the 2020s, and nobody, including the people running the factories, knows the answer yet.