Investment in AI inference chips accelerated sharply through 2024 and 2025, with capital commitments spanning asset-backed loans, pre-IPO rounds, and venture deals collectively exceeding several billion dollars. The trend reflects a broader recalibration among investors who previously concentrated exposure on GPU-centric hardware and are now directing capital toward purpose-built inference silicon.
One of the most structurally significant transactions came from General Compute, an AI inference cloud startup that secured a $400 million loan from tech investment firm Upper90. The financing uses inference-specific ASIC chips as collateral, marking one of the first asset-backed deals structured around non-GPU AI hardware. General Compute’s SN50 inference chips reportedly deliver up to 16 times faster inference than GPU-based cloud alternatives while eliminating the water-cooling demands typical of GPU deployments.
General Compute’s $400 million loan treats inference-specific ASIC chips as collateral — a Wall Street first.
The capital is earmarked for deploying hardware optimized to run pre-trained open AI models at scale, and the deal is widely interpreted as evidence that dedicated inference silicon is emerging as a bankable infrastructure asset class on Wall Street.
South Korean AI chip startup Rebellions completed a $400 million pre-IPO round led by Mirae Asset Financial Group and the Korea National Growth Fund, pushing its valuation to approximately $2.34 billion. The company, which specializes in Neural Processing Units designed for data center inference workloads, raised roughly $650 million within a six-month window when combined with a prior $250 million Series C, bringing total fundraising to around $850 million.
The new capital funds mass production of the next-generation Rebel chip, which targets large language model inference and retrieval-augmented generation applications.
Enterprise inference software platform Fireworks AI raised $250 million in a Series C round at a $4 billion valuation. The round drew participation from Lightspeed Venture Partners, Index Ventures, Sequoia Capital, Nvidia, AMD, and Databricks, signaling that incumbent chipmakers are actively hedging by backing inference-layer software alongside their own hardware businesses.
At the infrastructure level, Cerebras Systems secured $1.1 billion in a Series G round, while Groq raised $750 million to advance its low-latency AI accelerator systems. Both companies focus on inference performance as a primary design objective rather than treating it as secondary to training workloads. General Compute’s use of SambaNova SN50 chips, which are backed by Intel, further illustrates how alternative chipmakers are gaining credibility as viable infrastructure providers in the inference market.
Quarter-over-quarter funding data reinforces the scale of activity: AI and quantum chip startups collectively raised over $2.5 billion in Q3 2025 alone, with AI accelerators and inference-focused silicon accounting for the dominant share.
The pattern across these transactions points to a structural shift in how AI hardware is financed and valued. GPU clusters, which defined the first wave of AI infrastructure investment, are increasingly viewed as general-purpose assets. Specialized inference chips, by contrast, are being underwritten as targeted infrastructure with measurable performance and cost advantages, a distinction that is reshaping both venture and debt financing strategies across the sector.
