Luminal Raises $25M Series A, Claims 90 Trillion Tokens Monthly on Multi-Chip Inference Cloud
AI inference startup Luminal announced a $25 million Series A on November 12, 2026, led by Standard Capital, with seed investors such as Y Combinator, Felicis, SagaVC, and Liquid 2 doubling down [1]. The San Francisco company launched its public inference cloud after a $5.3 million seed round in November 2025 led by Felicis with angels including Paul Graham and Guillermo Rauch [1][4]. Luminal says it now processes approximately 90 trillion tokens per month [1].
The token-volume figure, if verified, would be a rare concrete metric for actual production inference demand. However, outlets cite different numbers: SiliconANGLE reported roughly 45 trillion tokens per month in its November 12 profile, as noted by explainx.ai [2], while Technode and Technwz repeat the 90 trillion figure [1][3]. The discrepancy has not been reconciled.
Compiler-first approach targets Nvidia alternatives
Luminal's product is a search compiler paired with a hosted inference cloud. The compiler uses equality saturation — a technique that represents many mathematically equivalent implementations of a model and searches for the fastest on a given chip — to generate optimized kernels ahead of time. The company argues that most serving engines still interpret models at runtime, leaving throughput on the table [2][5].
Co-founder and CEO Joe Fioti, formerly at Intel AI hardware, said the team started from the observation that silicon capabilities sit idle because software does not exploit them [2][5]. The pitch is software-led total cost of ownership: extract more tokens per dollar from chips already purchased, including non-Nvidia accelerators.
A large share of Luminal's current token volume runs on Amazon's Trainium chips under what the company describes as a deepening partnership [1]. The compiler also targets GPUs and other ASICs, with the ambition of letting customers choose hardware based on workload rather than software ecosystem [5].
Benchmark claims await independent reproduction
Luminal's website publishes a chart comparing tokens per second on GPT-OSS 120B across eight H100 SXM GPUs, showing 36,000 tok/s for Luminal, 28,000 for TensorRT-LLM, and 26,000 for vLLM, alongside a claimed 3.2× advantage over vLLM, sub-10ms p99 latency and zero runtime overhead [2]. The company also offers a licensed deployment option for on-premises or VPC use [2].
Explainx.ai notes it has not independently reproduced these figures. The stated tok/s numbers imply a 1.38× advantage (36,000/26,000), not 3.2×; it is unclear whether the 3.2× claim refers to a different configuration or model. Critical details — exact checkpoint hash, GPU driver and power cap, batch size, sequence length, concurrency settings, and p50 versus p99 latency breakdowns — are not disclosed on the marketing page [2]. No public independent benchmarks exist.
What is known and what is not
- Funding: $25M Series A led by Standard Capital, announced Nov 12, 2026. Seed was $5.3M in Nov 2025 led by Felicis [1][4][5].
- Token volume: Company claims ~90T tokens/month; one outlet cites ~45T. No third-party verification [1][2][3].
- Hardware mix: "Large share" on Amazon Trainium; compiler targets GPUs and ASICs broadly [1][5].
- Technique: Equality saturation compile-time search; artifacts cached for repeated inference [2][5].
- Benchmarks: Vendor-published only; missing independent reproduction, configuration transparency, and breadth of model coverage. Published tok/s figures do not align with the claimed 3.2× advantage [2].
- Pricing: Not listed transparently for either cloud or licensed tiers [2].
Why it matters
If Luminal's compiler delivers consistent gains across diverse hardware, it could give procurement teams leverage against Nvidia's CUDA ecosystem — not by replacing Nvidia silicon, but by making alternatives viable without rewriting agent stacks [2]. The category is crowded: Nvidia's own TensorRT-LLM pushes compile-time search, Baseten's Rust engine targets one model family on Blackwell, and the SGLang consortium advances open runtimes [2].
For now, the 90 trillion token claim and the benchmark charts are vendor statistics. Buyers should run A/B tests on their own prompts and hardware, with full reproducibility scripts, before changing autoscaling rules or latency SLAs [2].
Sources
- Luminal raises $25M Series A to cut AI inference costs ... (technode.global)
- Luminal $25M: Compiler Claims 36k tok/s vs vLLM (2026) (explainx.ai)
- Startup Funding News: Databento, Luminal and More (technwz.com)
- Announcing our $5.3M Seed Round - by Jonah Samost (blog.luminal.com)
- Luminal Series A led by Standard Capital - Standard Capital (standardcap.com)