Tiny Corp

What Tiny Corp does

Tiny Corp is a privately held, for-profit software and hardware company founded by George Hotz in 2022-2023 and based in San Diego, California. The company builds and maintains tinygrad, an open-source neural network framework written in Python with a PyTorch-like API that decomposes complex networks into three operation types (ElementwiseOps, ReduceOps, MovementOps) and supports autodiff, lazy tensor fusion, and JIT compilation. Tinygrad targets multiple hardware backends including NVIDIA (CUDA), AMD (LLVM/HIP, plus a custom AM userspace driver), Apple Metal, CPU, QCOM OpenCL, and WebGPU, and is used in production by comma.ai's openpilot to run the autonomous-driving model on Snapdragon 845 GPUs. Supporting developer tools include the HCQ (Hardware Command Queue) low-level API and the AM driver for AMD RDNA3/RDNA4 GPUs, and the company ships tinyos, a custom Ubuntu-based OS image pre-installed with tinygrad and PyTorch.

On the hardware side, Tiny Corp sells the TinyBox line of pre-configured deep learning computers through a direct-to-consumer Shopify storefront (tinycorp.myshopify.com) with wire-transfer-only payment, no sales team, and no customization options. Shipping SKUs include the TinyBox Red v2 (778 TFLOPS FP16, 4x AMD 9070XT, 64GB GPU RAM) and the TinyBox Green v2 Blackwell (3086 TFLOPS FP16, 4x RTX PRO 6000 Blackwell, 384GB GPU RAM), with the ExaBox (~1 EXAFLOP, 720x RDNA5 AT0 XL GPUs) available as a preorder for 2027 delivery. In April 2026 Apple officially approved TinyGPU, a macOS kernel extension that enables external AMD and NVIDIA GPUs to function as AI accelerators on Apple Silicon via USB4/Thunderbolt, restoring eGPU support for AI workloads.

The business model is product-led and self-serve: hardware is the only disclosed revenue stream, customers are individual developers, researchers, and SMBs ordering directly online, and software adoption (tinygrad) is the primary hiring and community-growth funnel via GitHub contributions and paid bounties. The company's stated mission is to "commoditize the petaflop," and its strategy positions the open-source framework as a vendor-neutral alternative to NVIDIA-dominated AI infrastructure, with the TinyBox hardware line serving as both revenue and a reference platform for the framework.

Name

Tiny Corp

Legal name

Tiny Corp

Website

https://tinygrad.org

Company type

Private

Founded year

2023

Operating status

Operating

Headcount range

1–10 employees

Short description

Tiny Corp develops tinygrad, an open-source neural network framework, and sells TinyBox pre-built deep learning computers for AI developers, researchers, and practitioners seeking hardware-agnostic, cost-efficient GPU compute.

Ownership category

akta.pro rank

Product category

AI Deep Learning Hardware

NAICS

Electronic Computer Manufacturing (334111)

SIC

Electronic Computers (3571)

akta.pro industry

PRIMARYHDACAAAI

Mini PCs & Small Form Factor Desktops

Hardware, Technology & Telecom/Computing Hardware & Peripherals/Personal Computing Devices (Desktops, Laptops, Workstations)

SECONDARYHDAAAJAB

On-Device Inference Runtimes & SDKs (mobile/embedded)

Hardware, Technology & Telecom/Artificial Intelligence & Machine Learning/Edge AI & On-Device Machine Learning

Keywords

• Deep learning hardware
• Neural network framework
• AI compute systems
• GPU workstations
• Open-source ML framework

Headquarters

HQ city

San Diego

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2C

Offering type

Hardware or Manufacturing

Cost components

Supply Chain, Operations, Personnel, Technology or R&D, Infrastructure, Marketing or Sales

Revenue model

1. Hardware Sales (TinyBox): Sale of pre-configured deep learning computers (TinyBox red v2, green v2 blackwell, and exabox). Products ship within one week of payment receipt. Wire transfer is the only accepted payment method. Pickup available in San Diego plus worldwide shipping.

Pricing tiers

Go-to-market motion1 record

Distribution channels1 record

Marketing channels5 records

Core offering

Tiny Corp designs, manufactures, and sells petaflop-scale AI compute workstations (TinyBox Red, Green v2 Blackwell, and ExaBox) powered by their open-source tinygrad neural network framework. The company produces computers in order to commoditize the petaflop, delivering turnkey deep learning hardware with bundled software stack to AI/ML developers and researchers at transparent pricing.

Product overview

Tiny Corp operates as a dual product company offering both software and hardware. The core software product is tinygrad, an open-source neural network framework written in Python that supports multiple hardware backends (NVIDIA, AMD, Apple Metal, CPU, QCOM, WebGPU). The hardware product line consists of tinybox systems—deep learning computers available in three tiers: Red v2 (AMD-based, 778 TFLOPS), Green v2 Blackwell (NVIDIA-based, 3086 TFLOPS), and ExaBox (AMD-based, ~1 EXAFLOP for 2027). Supporting products include TinyGPU (eGPU driver for Apple Silicon Macs), the AM Driver (AMD userspace driver), and HCQ (Hardware Command Queue API for runtime development). The company also provides tinyos, a custom Ubuntu-based OS for tinybox hardware. The company is funded ($5.1M raised in May 2023) and founded by George Hotz.

Differentiator

Problem solved

Functional benefit

Brands

• tinygrad: An open-source neural network framework developed and maintained by Tiny Corp. It is described as a simple and powerful framework that breaks down complex neural networks into three OpTypes.
• Tinybox
• TinyGPU

Products and services

• tinygrad Open-source neural network framework and full stack computer infrastructure that uses the entire GPU and includes a lazy tensor evaluation system and JIT runtime. Designed for AI/ML developers and researchers who want a hardware-agnostic deep learning stack supporting both NVIDIA and AMD GPUs.
• TinyBox Red v2 Multi-GPU deep learning workstation bundled with tinygrad and tinyos, offering petaflop-scale compute at $25,000. Targets AI/ML developers and researchers seeking self-hosted AI training hardware at 10x price-performance versus competing devboxes.
• TinyBox Green v2 Blackwell Higher-end multi-GPU deep learning workstation with petaflop-scale compute bundled with tinygrad and tinyos, priced at $48,000. Targets buyers needing greater performance than the Red v2 for larger AI training workloads.
• TinyBox ExaBox Enterprise-scale AI compute system offering exaflop-class compute, positioned for organizations that need self-contained petaflop to exaflop GPU cluster hardware. Sold via multi-year contract with a $50,000 deposit.
• TinyGPU Apple-approved eGPU driver for macOS that enables use of AMD and NVIDIA eGPUs with Mac Pro, Mac mini, and Mac Studio machines for AI workloads, addressing Apple's blocking of third-party eGPU support.

Quantifiable outcome

• 10x better price-performance than comparable systems in MLPerf Training 4.0 benchmarks
• +1 more outcomes

Named customers1 record

Segments3 records

Ideal customer profiles3 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration8 records

AI capability6 records

Feature5 records

Partnerships

Two partnerships are on record, tiered core.

• ApplecoreTechnology or Integration · 1 April 2026Apple officially approved TinyGPU, a driver developed by Tiny Corp that enables external AMD and NVIDIA graphics cards to work with Apple Silicon Macs via Thunderbolt or USB4. This restores eGPU support originally removed with the Apple Silicon transition.
• AMDcoreTechnology or Integration · 24 May 2023Tiny Corp raised $5.1 million with the goal to demonstrate AMD's technology on MLPerf benchmark using the tinygrad framework. The company develops software for AMD chips to compete with NVIDIA in the AI compute space.

Scale indicators4 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Lambda Labs: Sells GPU workstations, servers, and a cloud platform purpose-built for deep learning with a similar price-performance-led positioning. The most direct competitor to TinyBox in the pre-configured AI workstation segment.
• Boxx Technologies: Specialty workstation manufacturer selling high-performance computing systems for AI, VFX, and engineering. Comparable to TinyBox as a pre-configured deep learning workstation vendor, though with broader customization and enterprise sales motions.
• Puget Systems: Custom workstation and server builder with deep learning configurations, known for tightly tested Linux-based AI workstations. Comparable on product form factor and target customer (AI practitioners wanting pre-built workstations).

Emerging players

• CoreWeave: GPU cloud provider originally built on commodity GPUs to offer AI compute below hyperscaler pricing. Shares Tiny Corp's philosophy of building a NVIDIA-alternative compute layer, but delivered as cloud rather than boxed hardware.
• Cerebras Systems: AI accelerator startup building wafer-scale chips for training and inference. Comparable as an emerging AI compute vendor pursuing NVIDIA-alternative architectures, though at much larger scale and with proprietary silicon.
• RunPod: GPU cloud platform offering on-demand AI compute at lower price points than hyperscalers. Comparable on the cost-disruptive AI compute positioning, though delivered as cloud rather than boxed hardware.

Broad incumbents

• NVIDIA: Dominant supplier of AI GPUs (RTX PRO 6000 Blackwell) used inside TinyBox Green v2 and the broader market. The incumbent against which Tiny Corp explicitly positions itself as an AMD-based, software-disaggregated alternative.
• AMD: Supplier of the GPUs (9070XT, RDNA5) inside TinyBox Red v2 and ExaBox and a strategic partner for MLPerf benchmarking. Both enables Tiny Corp and competes for end-customer mindshare as an AI accelerator vendor.
• PyTorch: Dominant open-source neural network framework that tinygrad explicitly benchmarks against and aims to outperform. The most direct incumbent on the software side; tinygrad's API is PyTorch-compatible to ease migration.
• JAX: Google-backed open-source ML framework with a similarly minimalist design philosophy (functional transformations, NumPy compatibility). Comparable as a competing alternative-framework position to PyTorch in the research community.

Market position

Strengths4 records

Weaknesses4 records

Competitive moat5 records

Key risks6 records

Key highlights6 records

Customer concentration

• Xx.com/__tinygrad__
• GitHubgithub.com/tinygrad/tinygrad
• LinkedInlinkedin.com/company/the-tiny-corp

Revenue estimate

Valuation estimate

Number of profiles

Profiles1 record

Funding overview

Funding rounds1 record

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Tiny Corp