Competitors

DIRECT COMPETITOR. Pure inference-as-a-service, 10T tok/day, 10K+ customers (Cursor, DoorDash). Must match or beat $0.20/M pricing via energy cost advantage.

DIRECT COMPETITOR. Token Factory = managed inference at $0.13/M (lowest published). 69% gross margin. Must undercut via energy advantage or compete on sovereign/compliance.

Non-GPU inference at 40M+ tok/sec threatens GPU-based cost assumptions. IPO Q2 2026. Evaluate as potential compute partner or technology licensor.

Nvidia acquisition signals inference hardware consolidation. LPU's deterministic latency (877 tok/s) sets benchmark. Pricing is higher than infrastructure-first providers' target range.

Nvidia's $150M investment signals intent. Custom C++ engine targets enterprise inference workloads. Expanding to training creates a full-stack competitor.

DIRECT COMPETITOR. Now a managed inference platform, not just GPU cloud. MemoryAlloy: 9.9x TTFT vs vLLM. BYOM custom model support. 5x bookings growth. Product team led by ex-Google Cloud AI SVP (Erwan Menard, ex-Vertex AI).

Price floor leader at $0.03/M input. 8,000x volume growth since seed. SOC2 + ISO 27001 certified. Lean team (~15 employees) with Blackwell GPU advantage. Must monitor as cost benchmark.

Commercializing vLLM, the open-source engine behind 400K+ GPUs. Meta, Google, Amazon in production. The software layer MARA likely uses. Monitor commercial offerings closely.

Distribution moat: 332K paying customers, 50K+ models via Replicate acquisition. Edge inference in 330+ cities. Threat is distribution, not raw performance.

Model-specific chips deliver 73x H200 performance at 1/10th power. Trade flexibility for extreme efficiency. Air-cooled PCIe form factor fits modular containers. Evaluate as compute partner.

Crypto-to-AI pivot. $55.6B backlog is GPU rental/training, not managed inference. Potential partner for GPU supply. Watch for inference API launch.

Prices at ~breakeven with FlashAttention optimization. Energy cost advantage is key to sustainable margins. Potential integration partner for model serving.

Distribution channel opportunity: list inference endpoints on OpenRouter for demand generation. Their a16z 100T token study shows inference demand shifting to code + reasoning.

Acquired by Cloudflare Nov 2025. 50K model marketplace is distribution play. Cold-start latency (60s+) limits production use. Now part of $30B+ Cloudflare edge network.

Acquired by NVIDIA Apr 2025. Rebranded as DGX Cloud Lepton. Founded by Caffe creator (Yangqing Jia). Now NVIDIA's multi-cloud GPU marketplace connecting devs to CoreWeave, Crusoe, Lambda.

Developer-first serverless GPU platform. Built in Rust with sub-1s cold starts. $1.1B unicorn, in talks for $2.5B round. 90% workloads are inference. Different approach: compute platform vs managed inference.

Fastest ARR growth in inference: $25M to $200M in 12 months. Media/multimodal focus today, expanding to LLMs. Sequoia + NVIDIA backed. Watch for LLM pivot.

Largest European AI infrastructure company. $1.1B Series B. Stargate Norway with OpenAI (100K GPUs). $14B Microsoft deal value. Sovereign cloud competitor and potential model for MARA's European expansion.

Pure GPU rental with zero egress fees. Not in managed inference today. Potential GPU supply partner. Monitor for inference API announcements.

Cautionary tale: $5B peak valuation collapsed to $1.6B Intel offer. Validates GPU-agnostic approach over custom silicon lock-in. Potential acqui-hire talent pool.

Marketplace model for custom LLM inference. Potential distribution partner. Claims 90% cost reduction. a16z + Multicoin backing.