The Rain AI patent acquisition gives OpenAI a slice of neuromorphic intellectual property from a startup that came within weeks of total shutdown, after broader takeover talks between the two companies ended without a deal, according to Wired.
Neither company has disclosed the price paid, the number of patents transferred, or which specific technologies are covered. Rain AI has nearly ceased operating and most of its staff have departed.
What Rain AI Actually Built
Rain’s core product was a neuromorphic processing unit (NPU): what the company described at its $25 million Series A as the world’s first end-to-end analogue, trainable AI circuit, according to Design Reuse.
The architecture uses memristors as artificial synapses overlaid on neuron circuits in a sparse pattern that replicates sparse brain connectivity, enabling tens of millions of artificial neurons on a single chip. Rain claimed the NPU could reduce AI computing costs by more than 10,000 times compared to conventional accelerators, and said it had held talks to supply chips to Google, Oracle, Meta, Microsoft, and Amazon, per The Decoder.
The training algorithm, called Equilibrium Propagation, was designed to run directly on the analogue circuit rather than requiring the digital backward-pass that makes transformer training so energy-hungry. Whether those claims held up at scale was never proven commercially.
The Rain AI Patent Acquisition in Context: A Long Entanglement
OpenAI’s connection to Rain predates the patent deal by years. In 2019, OpenAI signed a nonbinding letter of intent to purchase $51 million worth of Rain chips once they became available. Sam Altman had also made a personal investment of more than $1 million in the startup, according to investor disclosure documents reviewed by Wired.
Rain’s $33 million April 2022 round, led by Prosperity7, was partly justified in deal documents by citing Altman’s involvement and the OpenAI letter of intent, The Decoder reported. Y Combinator is also listed among Rain’s backers, per Data Center Dynamics.
The chip order was never completed, and Rain spent the years after that round struggling to reach production scale.
How the Fundraise Unravelled
Rain’s difficulties accelerated in late 2024. A planned $150 million Series B, which would have valued the company at roughly $600 million, was postponed multiple times before collapsing entirely, the New York Post reported.
Co-founder Will Passo stepped down as CEO for personal reasons and was replaced by co-founder Jack Kendall, who acknowledged the leadership change had damaged investor confidence and hampered the Series B effort. Rain had also brought in Jean-Didier Allegrucci, an Apple veteran who helped lead that company’s transition away from Intel silicon, but the hire was not enough to close the round.
Facing a cash cliff after the fundraise collapsed, Rain secured a $3 million bridge loan to maintain operations while exploring a sale, according to reporting aggregated by AIM Research via LinkedIn (no primary wire source independently confirmed this figure). The acquisition talks with OpenAI eventually stalled, leaving the patent sale as the residual transaction.
What OpenAI Gets, and What It Does Not
Purchasing patents without the company means OpenAI acquires IP without Rain’s remaining headcount, liabilities, or unfulfilled commercial commitments. It is a targeted extraction of potentially useful research, not a bet on Rain’s roadmap.
Whether any of the neuromorphic architecture translates into OpenAI’s own chip programme is unconfirmed. OpenAI has not stated whether the Rain patents will feed an internal design effort, a licensing position, or defensive blocking. The analogue-training approach Rain pursued sits some distance from the large-scale CUDA-compatible GPU clusters that OpenAI currently depends on.
Altman’s dual role as Rain investor and OpenAI CEO makes the governance optics awkward regardless of process, though available reporting does not identify a conflict finding or describe his direct involvement in approving the deal.
The more immediate question is whether Rain’s memristor-based IP is worth anything on a timeline that matters. OpenAI’s infrastructure roadmap is moving fast, and analogue neuromorphic research has historically struggled to close the gap with GPU scaling before the frontier moves on.
