With whey protein concentrate costing more than twice what it did a year ago, brands are replacing part of it with milk protein concentrate, soy or pea. No single protein can stand in for whey across solubility, viscosity, heat stability and flavor, so the replacement is a blend, and the blend has to be designed. Teams that treat it as a multi-variable formulation problem from the start reach a workable hybrid in fewer trials than teams that swap one ingredient at a time.
How tight the whey market has become
Whey prices have eased from their summer peak but are still far above last year. According to market data firm Expana, cited by FoodNavigator, US high-protein whey concentrate cost about $12.25 a pound at the end of August, up 134% year on year. European prices ended August about 122% higher than a year earlier (FoodNavigator, 20 October 2026).
To put that in formula terms: a ready-to-drink shake that uses 30 g of whey concentrate per serving carries about $0.81 of whey at $12.25 a pound. At a year-ago price of roughly $5.24 (the same figure divided by 2.34), it carried about $0.35. On a run of one million bottles, that gap is about $460,000 before any other cost moves.
Demand is the main driver. FoodNavigator points to fitness culture's focus on "proteinmaxxing" and to people on GLP-1 medications trying to preserve muscle. ADM told the publication that 66% of consumers globally want to increase their protein intake, up from 57% in 2023.
Supply can't respond quickly. Whey is a by-product of cheese, so processors can't make more protein without making more cheese, and some suppliers are sold out for the rest of the year (ProFood World). US processors are investing more than $11 billion in new and expanded capacity across 19 states, according to the International Dairy Foods Association, but new plants take time to come online.
Brands are moving to hybrid protein systems
Manufacturers are blending dairy and plant proteins to reduce how much whey each product needs. ProFood World names the main alternatives as milk protein concentrate (MPC), soy and pea, and NutraIngredients adds collagen and yeast protein in sports nutrition (NutraIngredients, 13 October 2026).
Ingredient suppliers have launched products aimed at exactly this gap. FoodNavigator lists several from 2026:
- ADM launched eight soy and pea products in the first half of the year, including ProFam 883, a soy protein isolate for dairy hybrids, dairy-forward ready-to-drink beverages and sports nutrition.
- Puris released H2-IZO in July, a pea protein for high-protein beverages that it says delivers 32 to 40 g of protein per serving with a lighter texture.
- Roquette launched Natralys Pea 850F, a pea protein isolate for dairy alternatives with fewer vegetal notes.
- NNB Nutrition and Axiom Foods introduced PeptiClear in August, hydrolyzed plant proteins aimed at solubility, clarity and flavor in beverages and frozen desserts.
The launches tell you where suppliers see the problems: texture, off-notes and clarity. Those are the properties a formulator loses first when whey comes out.
Why no single protein replaces whey
Whey does several jobs at once in a formula, and each alternative covers some of them. NutraIngredients describes whey as still setting "the benchmark for amino acid quality, leucine content, digestibility, functionality and consumer recognition." Replacing it means checking each of those jobs separately:
- Solubility and clarity. Plant proteins are prone to grittiness, opacity and sediment, which is why clear and hydrolyzed plant proteins are a 2026 launch theme.
- Viscosity. A protein that thickens more can push a shake past what consumers will drink, or help a pudding set. The same property is a defect in one format and a benefit in another.
- Heat stability and gelation. Retort, UHT and baking all denature proteins differently. Whey's behavior under heat is often what the existing process was built around.
- Flavor. Pea and soy bring vegetal and beany notes that change the flavor system, and masking them can add cost back.
- Nutrition. Soy has a PDCAAS of 1, ADM notes, but collagen has a different amino acid profile and should not be presented as nutritionally equivalent to whey.
The grade matters as much as the source. In a Wageningen study, researchers compared three pea protein isolates made by different processes against whey protein isolate. A mildly processed isolate matched whey's solubility at pH 7 and its gel strength up to 15% protein, while two others thickened more and dissolved less. Any of the three could replace up to half the whey without losing gel strength, but above 15% protein whey still made the strongest gels (Kornet et al., Food Hydrocolloids, 2021).
The practical result: "pea protein" is not one ingredient, and how far a partial swap works depends on the protein level and the format. A formulator choosing between two pea isolates and a soy isolate at three blend ratios already has a problem with many interacting variables.
Real products show the cost of getting it wrong. Canadian baking-mix maker HelloAmino, which uses whey in all 30 of its high-protein mixes, bought replacement isolate at a 50% premium when a supplier ran out, and the new isolate dried out its baked goods, according to the Financial Post via ProFood World.
Designing the blend instead of guessing it
A protein blend is a mixture problem: the proteins share a fixed total, usually set by the protein claim on the label, so raising one lowers another. Mixture designs, a branch of design of experiments (DOE), are built for exactly this, and they cut the number of trial blends sharply.
Here is the arithmetic for a four-protein system of whey concentrate, MPC, soy isolate and pea isolate:
- Testing every blend in 10% steps means 286 different blends, which no pilot plant will run.
- A simplex-centroid design tests 15: each protein alone, every 50:50 pair, every equal three-way blend, and an equal four-way blend. That is enough to fit a model of how each property responds to the mix, including two-way interactions such as whether pea thickens more when MPC is present.
- Real constraints shrink the space further. If whey must stay at 40% or more of the protein for leucine and flavor, and pea at 30% or less for taste, a constrained design can cover the remaining region in a similar number of runs.
A few practices make the design pay off:
- Measure every response on every blend: solubility or sediment, viscosity, stability after the real heat process, sensory scores and cost per serving. A blend that wins on viscosity and fails on flavor is still a failure.
- Fix the grade, not just the source. Record the supplier, lot and isolate grade, since two pea isolates can behave like different ingredients.
- Run the heat step you will use in production. A blend that is stable at bench temperature can gel or sediment after retort or UHT.
- Build cost into the response. At today's whey prices, the cheapest blend that passes every specification is often the real target, and the model can find it directly.
- Use the results for the next round. Once the first 15 to 20 blends are measured, Bayesian optimization can choose the next few blends most likely to improve on the best so far, without rerunning the whole design.
Done this way, a whey shortage leaves something behind: a model of the protein system that is ready when prices swing again, or when a pea supplier changes its process.
What this means for F&B and nutrition R&D teams
- Which products in your portfolio carry the most whey per serving, and what does each cost at today's prices compared with last year's?
- Do you know which protein claims (grams per serving, "complete protein", leucine content) each product must keep through a reformulation?
- If a pea or soy supplier changes grade, could you find every formula and test result that used the old one?
- Are trial blends being chosen one at a time, or from a design that covers the whole blend space?
- When whey prices fall again, will you know which hybrid formulas are worth keeping on cost and which should switch back?
For the nutrition side of the same demand shift, see how GLP-1 drugs are changing food formulation.
How Uncountable approaches protein reformulation
Uncountable is the AI-native end-to-end product development platform. Teams use its DOE and predictive tools to set up constrained mixture designs, record every blend with its ingredient lots and test results, and model solubility, viscosity, stability, sensory and cost together. Each new round of trials builds on the last, so a protein system can be re-optimized when prices or suppliers change.
.png)
.png)
.png)
