Luma AI raises $900 million led by HUMAIN

Background

Venture funding is usually raised in named stages, and a "Series C" is conventionally the third of the large priced rounds a young company sells to outside investors. Rounds at that stage are normally less about proving an idea than about scaling something that already works: hiring, infrastructure, and the machinery needed to serve customers at size. Luma AI announced on 19 November 2025 that it had raised a $900 million Series C ^[1]. The company's About page records the same $900 million round as a Series C led by HUMAIN ^[2], and states that Luma has raised more than $1.1 billion across five rounds since 2021 ^[2].

Who put money in

The round was led by HUMAIN, with significant participation from AMD ^[1]. Andreessen Horowitz, Omniva, Amplify Partners and Matrix Partners also took part ^[1]. In a venture round the lead investor is the one that negotiates the terms and normally anchors the deal with the largest commitment; the other participants join on the same terms. Here the list of participants combines venture firms with AMD, and sets the new lead investor alongside established funds.

Project Halo

The funding announcement came together with an infrastructure plan. Luma and HUMAIN announced Project Halo, a 2 GW compute supercluster for training and inference, deploying from Q1 2026 ^[1].

Each of those terms describes something fairly concrete. "Training" is the phase in which a model is shown large amounts of data and its internal parameters are adjusted so that its outputs get better at the task; "inference" is the phase in which a finished model is actually used to answer requests, which is where most of the ongoing electricity bill of a deployed AI service ends up. A "supercluster" in this context means a very large group of machines wired together with fast networking so that they can be scheduled as one system rather than as separate computers. And a gigawatt is a unit of electrical power: a 2 GW figure is a statement about how much power the facility is designed to draw. For AI data centres, power is a practical stand-in for capacity, because every accelerator in the building draws electricity and the surrounding cooling draws more.

The deployment date matters as much as the capacity. Starting from the first quarter of 2026 means the cluster is intended to come online shortly after the money is announced, rather than at some indefinite point in the future.

Why it matters

The announcement is a compact illustration of where money in AI has been going. A funding round and a power-plant-scale computing facility are announced as a single package: capital is converted into electricity and machines, and the machines are then used to train and serve models. That makes the size of a round a rough proxy for the size of the compute a company expects to command, and it makes energy and hardware supply the practical constraints on what a model developer can do next.

The pairing of a lead investor with a compute project is also a structural point. In this case HUMAIN leads the round and co-announces the supercluster, so the same relationship supplies both money and infrastructure ^[1]. For a company that has raised across five rounds since 2021 ^[2], a round of this size is a step up in scale, and it commits the company to an operating cost — running and powering a large cluster — that did not exist at earlier stages.

The concrete test comes when Project Halo begins deploying from Q1 2026 ^[1]. Until then, the $900 million round is the part that is settled: a Series C led by HUMAIN, with AMD, Andreessen Horowitz, Omniva, Amplify Partners and Matrix Partners participating ^[1].