A fuel cell is held back neither by the hydrogen nor by the membrane, but by the sluggishness of a single reaction: oxygen reduction at the cathode. It is accelerated with platinum, a rare and expensive metal, and it has long been known that an ordered platinum-cobalt alloy outperforms platinum alone — provided it is heated enough for the atoms to settle into place. The problem is that the very heat that orders the atoms also fuses the nanoparticles into one another, destroying the active surface one set out to exploit. A team at Washington University in St. Louis has just removed that constraint by changing not the catalyst, but what it rests on.

Source: nature.com

In plain terms

In a fuel cell, electricity arises from the meeting of hydrogen with the oxygen in air. On the oxygen side, the reaction is lazy: a catalyst is needed to get it going, and it is almost always platinum — a metal so costly that it accounts for a large share of the cell's price. It is known that mixing platinum with cobalt, and arranging the two metals in regular layers like a mille-feuille, yields a catalyst that is both more active and more durable. But obtaining that arrangement requires very strong heating — and at that temperature the catalyst grains melt into one another and grow large. Yet a catalyst acts only through its surface: large grains mean less surface for the same amount of platinum, and therefore precious metal wasted.

The researchers' trick: house the grains inside hollow carbon spheres pierced with channels radiating from the centre, like the spokes of a wheel. Each grain ends up locked in its own corridor and can no longer reach its neighbours. Heating can then go beyond 1000 °C — enough to order the atoms — without the grains growing. This is not yet an industrial product: the senior author himself states that catalyst problems remain to be solved, in collaboration with industrial partners.

Technical sheet — Discovery

Technical explanation

1. Why the cathode governs everything. A proton-exchange-membrane fuel cell oxidises hydrogen at the anode and reduces oxygen at the cathode:
   $\mathrm{O_2} + 4\,\mathrm{H^+} + 4\,e^- \rightarrow 2\,\mathrm{H_2O}$O2​+4H++4e−→2H2​O.
   This oxygen reduction reaction (ORR) involves four electrons and the breaking of a strong O=O bond; its kinetics are several orders of magnitude slower than those of hydrogen oxidation. It is what sets the overpotential, hence the efficiency, hence the amount of platinum required. Cutting platinum without losing activity is the central economic stake of the field: nanostructuring already makes it possible, according to the institutional press release, to operate in a low-loading regime — typically less than a quarter of a milligram of platinum per square centimetre of electrode.

2. What chemical order brings — and what it costs. In a disordered PtCo alloy, the two metals occupy lattice sites at random. In the L1₀ intermetallic phase, they organise into alternating planes of platinum and cobalt, a regular tetragonal structure. It is established in electrocatalysis that this order compresses the surface platinum lattice and alters its interaction with oxygenated intermediates, which improves activity; order also makes the cobalt harder to leach out, which improves durability. But order cannot be obtained cold: enough thermal energy must be supplied for the atoms to diffuse and find their sites. Hence the need for high-temperature annealing.

3. The bottleneck: order versus dispersion. Heating triggers sintering — nanoparticles coalesce and grow, through migration then fusion, and through Ostwald ripening in which large particles grow at the expense of small ones. Yet a catalyst acts only through its accessible surface, which for a given mass varies as the inverse of the diameter: $S/m \propto 1/(\rho d)$S/m∝1/(ρd). Doubling the diameter halves the useful surface of the same platinum. Earlier approaches were therefore limited to anneals below 700 °C: they preserved fine particles but achieved only partial chemical order. Raising the temperature delivered order but lost dispersion. It is this trade-off — degree of order versus dispersion — that the study reports lifting.

4. The mechanism of the support. The support is not just any carbon black but a hollow sphere pierced by an ordered, radial array of nanochannels running from the centre outwards. Each PtCo nanoparticle is held inside a channel: according to senior author Gang Wu, this confinement by the carbon host allows it to remain stable at very small size even at high temperature. The most direct mechanism is the hindering of particle migration towards one another — one of the two sintering pathways; the accessible sources do not detail the fate of the other pathway, Ostwald ripening. The support thus ceases to be what caps the annealing temperature. The reported result is the conjunction of three quantities that could not previously be brought together: an ordered L1₀ phase fraction above 80%, particles under 5 nm, and all this at a high platinum content of 40 wt% — the condition for a thin and therefore low-resistance electrode.

5. What the geometry adds beyond confinement. According to senior author Gang Wu, the open structure of the channels also facilitates uniform ionomer distribution and the circulation of protons, oxygen and water within the electrode. This point belongs to mass transport rather than to catalysis proper: in a real electrode, an intrinsically active catalyst remains useless if the reactants cannot reach it and the water produced cannot escape.

6. What the durability measurement establishes. The reported test — 85% of performance retained after 150,000 voltage cycles, an equivalence estimated by the institutional press release at roughly 25,000 hours of operation — is a voltage-cycling protocol, that is, an accelerated ageing test: the potential is varied so as to induce in a short time the degradation that real use would produce over years. The actual duration of the test is not stated by the sources consulted. This figure must be read for what it is: a projected equivalence — the press release itself writes "likely equivalent" — and not a cell actually monitored for 25,000 hours.

Why It Worked

The lever is not the catalyst but its support. Containing sintering through carbon architecture is not new — confining supports have already been reported, notably PtCo nanoparticles encaged in graphene pockets (Zhao et al., Nature Nanotechnology, 2022). What changes here is not the principle of confinement but the thermal ceiling it permits: according to the institutional press release, earlier approaches annealed below 700 °C, a temperature at which chemical order sets in only partially — the sources consulted do not specify the annealing regime proper to each of those works. What the study reports is a support whose geometry — an ordered array of radial channels — remains viable beyond 1000 °C, and at high platinum content. High-temperature annealing ceases to be a hazard to avoid and becomes an available tool, making it possible to go after L1₀ order without paying in dispersion.

The gap between what is demonstrated and what is announced must remain sharp. What the study establishes: a support material and a synthesis route yielding, simultaneously, high order, fine particles and high platinum loading, with durability measured under an accelerated cycling protocol. What is not established: manufacturing this carbon at scale and at acceptable cost, performance in a full cell over real time, and integration into an industrial stack. The senior author himself states, in his institution's press release, that "catalyst problems" remain to be solved, in collaboration with industrial partners.

Causal Chain

Electrification and rising energy demand from data centres → interest in local electricity generation by fuel cell → but cost dominated by the cathode's platinum → need to increase activity per platinum atom → recourse to ordered L1₀ PtCo intermetallics (more active, cobalt less prone to leaching) → chemical order requires atomic diffusion, hence high-temperature annealing → but high temperature causes migration, coalescence and Ostwald ripening → particle growth → collapse of active surface per unit mass → historical trade-off between degree of order and dispersion → support made of hollow carbon spheres with ordered radial nanochannels → geometric confinement of each particle in its channel → hindering of the migration pathway → annealing beyond 1000 °C made admissible → L1₀ order > 80% and particles < 5 nm at 40 wt% platinum → 85% of performance retained after 150,000 cycles → (horizon) low-platinum fuel cells for heavy mobility and for powering energy-hungry sites.

Anecdote

Platinum owes its name to contempt: platina, "little silver" in Spanish, the nickname given by the gold panners of the Chocó, in Colombia — a name the scholar Antonio de Ulloa recorded on the spot in 1735 and made known to Europe in 1748. This white, refractory metal, impossible to melt with the means of the day, contaminated gold-bearing alluvium and was thrown away. A few decades were then enough for it to become precious, and two and a half centuries for its scarcity to make it the bottleneck of an energy technology — to the point that research today consists, very largely, in finding how to use as little of it as possible.

Legacy and Current Data

The stated motivation is the electricity consumption of data centres: according to an Electric Power Research Institute estimate relayed by the press release, they accounted for 4% of total US electricity demand in 2023 and could reach up to 9% of annual electricity generation in 2030 — two percentages that do not rest on the same base, and which therefore do not describe a homogeneous trajectory. Generating electricity on site from hydrogen would relieve the grid. This argument is framing rather than measured result, and the consumption projections remain estimates. On the technical substance, the work fits into a long trajectory — reducing the platinum loading of cathodes without sacrificing durability — in which activity gains have often been paid for in stability. The participation of two US national laboratories (Brookhaven, Lawrence Berkeley), which host the large structural characterisation instruments needed to establish a degree of order, places the work on the side of heavy characterisation. What is not quantified here — cost and scalability of the carbon support, performance in a full cell, behaviour under real start-stop conditions — conditions any scaling up.

The researcher's view — open questions

(Interpretation, not results of the study.) Three experiments would settle matters. (1) A synthesis balance sheet at scale: ordered radial-nanochannel carbon is a sophisticated structure; making it by the kilogram, at a cost and reproducibility compatible with industry, is the first unknown — a support more expensive than the platinum it saves would make no sense. (2) A long-duration full-cell test, under real start-stop conditions and with air impurities, to confront the 25,000-hour projection with a direct measurement rather than an accelerated protocol. (3) A post-mortem mapping of particle state after cycling: does the 15% loss come from residual sintering, from cobalt leaching, from corrosion of the carbon support, or from transport within the electrode? The mechanism of the remaining degradation determines where to direct the next effort.

Carbone classique — recuit < 700 °C




→
particules fines, ordre partiel


à 1000 °C : coalescence, surface perdue

Sphère creuse à nanocanaux radiaux — recuit > 1000 °C














Chaque particule PtCo est retenue dans son canal :
la migration d'une particule vers ses voisines est entravée.
→ ordre L1₀ > 80 % ET particules < 5 nm
→ à 40 % de platine en masse
Selon l'auteur senior Gang Wu, les canaux ouverts facilitent
le transport des protons, de l'oxygène et de l'eau.

Sources

References verified during the fact-checking audit (accessed 10 August 2026).

• Gao, L. et al. Radial nanochannel-array carbon enables high-performance intermetallic fuel cell catalysts. Nature Nanotechnology, 6 August 2026. DOI: 10.1038/s41565-026-02244-8. (Peer-reviewed article — primary source. Title, author list and statement of the result verified; the full text, including Methods, was not consulted.)
• Washington University in St. Louis. Platinum powers the future. The Source, August 2026. source.washu.edu (Press release from the authors' institution — cited for the device values, the quotes from Gang Wu and the data-centre framing.)

Background references

(Context for the general mechanism, distinct from the primary source.)

• Zhao, Z. et al. Graphene-nanopocket-encaged PtCo nanocatalysts for highly durable fuel cell operation under demanding ultralow-Pt-loading conditions. Nature Nanotechnology 17, 2022. DOI: 10.1038/s41565-022-01170-9. (Prior work cited in "Why It Worked": the confinement of PtCo nanoparticles by a carbon architecture had already been reported — this is not the novelty claimed here.)
• The notions used here — slow kinetics of oxygen reduction, effect of L1₀ chemical order on activity and on resistance to cobalt leaching, sintering by migration-coalescence and Ostwald ripening, dependence of specific surface area on particle diameter — belong to standard electrocatalysis and materials science, predate this study and are not claimed by it.

---

Confidence statement. Solidly established: the publication, on 6 August 2026 in Nature Nanotechnology (peer-reviewed article, DOI resolved), of a support made of hollow nanocarbon spheres with an array of radial nanochannels enabling the simultaneous achievement of an ordered L1₀-PtCo intermetallic phase fraction above 80% and particles below 5 nm at a platinum content of 40 wt%, after annealing beyond 1000 °C, with 85% of performance retained after 150,000 voltage cycles (the numerical durability values being taken from the institutional press release, not from the full text). Transparency: the full text (Methods, polarisation curves, mass activity) was not consulted; provenance breaks down as follows: the public abstract of the article states the fraction of highly ordered L1₀ phases (> 80%), the particle size (< 5 nm), the platinum content (40 wt%) and the annealing temperature (> 1000 °C); the institutional press release provides the durability (85% after 150,000 cycles, ≈ 25,000 h), the earlier 700 °C threshold, the platinum loading, the data-centre framing and the senior author's quotes. Each value is attributed to its source in the text — no data relating to the study has been added from memory; the historical anecdote on the origin of the word platinum and the general notions of electrocatalysis are domain knowledge, flagged as such. Calibration: the "150,000 cycles ≈ 25,000 hours" equivalence comes solely from the institutional press release, which itself qualifies it ("likely equivalent"); it is reported here as a projection and not as a measurement of real duration; the characterisation of voltage cycling as an accelerated ageing protocol is a domain reading added by the editorial team, not a term used by the sources consulted; the value of less than 0.25 mg·cm⁻² ("typically less than a quarter of a milligram per square centimetre") is given as the common low-loading regime described by the institutional press release, not as a measured performance of this catalyst; the electrocatalysis mechanisms invoked (effect of L1₀ order, sintering) are flagged as domain knowledge and not as findings of the study. Not yet established: cost and scalability of the support's synthesis, performance in a full cell over real time, industrial integration — the authors themselves point to later work with industrial partners. Uniqueness verdict: UNIQUE — no close key in the registry (A001–A007); first article on the site covering electrocatalysis and fuel cells.