Etched AI
Custom frontier inference clusters for massive transformer models at extreme scale.
Etched is a high-stakes, high-reward bet for AI labs and hyperscalers that need breakthrough inference efficiency at extreme scale. The custom A0 silicon, with LVI and CSM, addresses real bottlenecks in long-context and MoE inference, but the hardware is pre-production with no public benchmarks or pricing. Buy only if you can pilot and maintain a parallel GPU strategy. If execution matches claims, it could redefine inference economics.
Verified 8d ago · liveness 54/100 · cite: rightaichoice.com/tools/etched-ai
- AI labs running frontier inference workloads at scale
- Enterprises planning AGI-ready compute infrastructure
- Teams needing low-latency MoE inference for agents
- Hyperscalers seeking alternatives to GPU-based clusters
- Teams requiring production-available hardware today
- Small startups with limited budget or time to pilot
- Organizations locked into CUDA software ecosystems
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Etched if you need production-ready hardware today, require public benchmarks before committing, have a limited budget, or are not prepared to handle a specialized inference-only solution with an evolving software ecosystem.
Hardware is sold via contact sales; you will need to negotiate pricing and commit to minimum orders, which could be in the millions.
Etched's pricing is custom and not published, so you'll need to contact sales. Compared to GPU alternatives like NVIDIA H100 or B200, Etched claims to achieve better cost-efficiency on inference workloads, but total cost depends on your specific volume and negotiated terms. For teams at AI labs or hyperscalers, the potential savings on power and throughput could justify the investment. For smaller teams or startups with constrained budgets, the lack of transparent pricing and the need for
In short
Etched AI — Custom frontier inference clusters for massive transformer models at extreme scale. Best for AI labs running frontier inference workloads at scale, Enterprises planning AGI-ready compute infrastructure, Teams needing low-latency MoE inference for agents. Contact Sales pricing.
Viability Score
How well maintained and how widely used is Etched AI? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: August 2026
How we score →Key Features
- Custom A0 silicon on TSMC N4P
- Low Voltage Inference (LVI) for sustained 80%+ peak FLOPs utilization
- Cluster Scale Memory (CSM) with hybrid HBM/SRAM
- Ultra-low-latency, high-bandwidth scale-up interconnect
- Designed for many-trillion-parameter sparse MoE models
- Optimized for both prefill and decode workloads
- Co-designed rack, cold plate, PCB, power delivery
- Custom scheduling and network tiling algorithms
- Vertical integration: chip, rack, factory, data center
- Scalable to gigawatt-scale deployments
- Tested racks in representative data center deployments
- Production manufacturing with TSMC partnership
- Targets long-context and agentic workloads
About Etched AI
Etched is building a new category of AI hardware—frontier inference clusters—that co-designs chips, racks, software, and manufacturing to run today's largest transformer models with industry-leading throughput, latency, cost, and power efficiency for both prefill and decode workloads. Its A0 silicon, fabricated on TSMC N4P, introduces two key breakthroughs: Low Voltage Inference (LVI), which runs math blocks at under half the standard voltage to sustain 80%+ peak FLOPs utilization without thermal throttling, and Cluster Scale Memory (CSM), a hybrid HBM/SRAM shared memory pool with a proprietary ultra-low-latency interconnect that delivers SRAM-like decode speeds while keeping HBM-scale capacity. The company emerged from stealth with a $10.3B Series C, backed by VentureTech Alliance and Jane Street, and counts Geoffrey Hinton, Andrej Karpathy, and Tri Dao among its advisors. First rack-scale systems ship summer 2025, fulfilling over $1B in customer contracts. Etched targets AI labs, hyperscalers, and enterprises that need extreme inference efficiency for long-context agents and autonomous systems, where GPUs like H100 or B200 hit thermal and memory bottlenecks. Unlike general-purpose GPUs, Etched's hardware is specialized for transformer inference, prioritizing throughput, latency, cost, and power efficiency in a single, vertically integrated system.
Behind the Verdict
Etched is a serious, well-funded attempt to rethink inference hardware from the ground up, but it's not for everyone. If you're an AI lab or hyperscaler pushing multi-trillion-parameter MoE models with long context and agentic workloads, the promise of sustained 80%+ peak FLOPs and SRAM-like decode latency at HBM-scale capacity could solve real pain points that GPUs like H100 or B200 struggle with. The $10.3B Series C and $1B in customer contracts are strong signals, but the hardware hasn't shipped to general customers yet, and there's no public performance data to evaluate. That means you'd be betting on unproven silicon and a new software stack. If you need production-ready inference today, Etched isn't an option—stick with GPUs or other shipping accelerators. The vertical integration, including a Taiwan factory and San Jose test house, suggests they're serious about scaling, but it also means you're locking into a vendor-specific ecosystem, not a drop-in CUDA replacement. For teams with the budget and tolerance for risk, piloting Etched alongside existing GPU infrastructure could be smart. The lack of public pricing is a hurdle—contact sales, but expect enterprise-level costs. In practice, this is a bet on the team's execution as much as the technology; if they deliver on their claims, it could shift inference economics, but that's a big if.
Researching Etched AI? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Etched AI actually fits — and what changes day-one when you adopt it.
Deploying a large MoE model for long-context agent experiments
Outcome: Research team moves from GPU cluster with thermal throttling to Etched cluster, sees sustained 80%+ FLOPs utilization and lower latency for real-time interaction, enabling more complex agent loops.
Evaluating dedicated inference hardware to replace general-purpose GPUs
Outcome: Engineer tests Etched racks in a representative data center environment, validates SOTA throughput and power efficiency on production inference workloads, then negotiates a multi-year contract for gigawatt-scale deployment.
Planning AGI-ready compute infrastructure for autonomous systems
Outcome: CTO allocates budget for pilot deployment, runs proof-of-concept with Etched's cluster, sees potential for reduced latency and power costs, and integrates Etched into long-term infrastructure roadmap alongside GPUs.
Use Cases
- Deploy high-throughput LLM inference for customer-facing chatbots at lower cost
- Run production MoE models with reduced latency and power consumption
- Scale diffusion model inference for image generation services
- Replace GPU clusters for transformer inference in data centers
- Enable real-time AI applications with dedicated transformer hardware
Models Under the Hood
as of 2026-08-15
Limitations
- Etched is a hardware company building custom silicon for frontier inference.
- It is currently validating its first rack-scale product with customers and has not published public benchmarks or pricing.
- The company has announced a $10.3B Series C financing and demonstrated A0 silicon on TSMC N4P, but specific limitations on software tooling and availability are not detailed in the provided evidence.
as of 2026-08-01
Verification history
We have re-verified Etched AI 15 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
Showing the 6 most recent of 15 verification passes.
Free to cite with attribution — this page re-verifies continuously.
Where the pricing makes sense
The company stage and team size where Etched AI's pricing actually pencils out — and where peers do it cheaper.
Etched's pricing is custom and not published, so you'll need to contact sales. Compared to GPU alternatives like NVIDIA H100 or B200, Etched claims to achieve better cost-efficiency on inference workloads, but total cost depends on your specific volume and negotiated terms. For teams at AI labs or hyperscalers, the potential savings on power and throughput could justify the investment. For smaller teams or startups with constrained budgets, the lack of transparent pricing and the need for
Setup time & first value
How long it actually takes to get something useful out of Etched AI — broken out by persona, not the marketing-page minute.
For a pilot: expect several weeks to negotiate contracts and receive hardware, then a few days to deploy and integrate. For full-scale production: months for sizing, software integration, and infrastructure changes.
Switching to or from Etched AI
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From GPU clusters: Plan to port inference workloads from CUDA to Etched's specialized stack—expect a learning curve and potential code changes.
- ↗To GPU clusters: If Etched underperforms, you may need to migrate back to GPUs, but contracts may have exit penalties.
Resources & Guides
Tutorials & Learning
Official links
Tools that pair well with Etched AI
Common stack mates teams adopt alongside Etched AI, with the specific reason each pairing earns its keep.
MAX Engine
GPU-agnostic GenAI inference framework for serving, customizing, and optimizing open-source models.
Cerebras
World's fastest AI inference on wafer-scale chips for real-time agents and multimodal models.
Anyscale Endpoints
Managed Ray platform for distributed training and batch inference at scale.
Alternatives to Etched AI
View allMAX Engine
GPU-agnostic GenAI inference framework for serving, customizing, and optimizing open-source models.
Cerebras
World's fastest AI inference on wafer-scale chips for real-time agents and multimodal models.
Anyscale Endpoints
Managed Ray platform for distributed training and batch inference at scale.
Frequently Asked Questions
Categories
Best-of guides
Topics
Used Etched AI? Help shape our editorial sentiment research.