Etched AI
Frontier inference clusters for extreme-scale transformer workloads.
Etched is a high-stakes, pre-production bet that could redefine inference economics if the claims hold. It's only for teams with billions in compute spend who can afford to pilot and keep a GPU fallback. Investors are betting $21B, but buyers should wait for public benchmarks before committing. If you need production-available hardware today, stick with NVIDIA GPUs or consider other specialized inference providers like Groq or Cerebras.
Verified 7d ago · liveness 68/100 · cite: rightaichoice.com/tools/etched-ai
- AI labs running frontier inference at extreme scale
- Hyperscalers seeking alternatives to GPU-based clusters
- Teams needing low-latency MoE inference for agents
- Enterprises planning AGI-ready compute infrastructure
- Teams needing production-available hardware today
- Organizations locked into CUDA software ecosystems
- Small startups with limited budget or time to pilot
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, cannot absorb a long pilot phase, or rely on CUDA-based software stacks without the engineering resources to port them.
There is no published pricing; expect negotiation and potential for significant upfront capital outlay for rack-scale deployments.
Etched's pricing is custom and not published, but given the target volume (racks at gigawatt scale), it aims to be cost-effective for hyperscalers and large AI labs. Compared to general-purpose GPU clusters, Etched claims lower total cost of ownership for inference workloads. However, without public benchmarks, direct cost comparison remains speculative.
In short
Etched AI — Frontier inference clusters for extreme-scale transformer workloads. Best for AI labs running frontier inference at extreme scale, Hyperscalers seeking alternatives to GPU-based clusters, Teams needing low-latency MoE inference for agents. Contact Sales pricing.
What people actually say about Etched AI — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
23 mentions across 3 sources (Hacker News, YouTube, Lemmy) · researched Aug 30, 2026.
Average across the 3 sources that answered — each source counts once, not each post.
- +Unique ASIC design purpose-built for transformer inference, unlike general-purpose GPUs.
- +Low Voltage Inference claims sustained 80%+ FLOPs utilization without thermal throttling.
- +Cluster Scale Memory offers HBM-scale capacity with SRAM-like speeds for long-context workloads.
- +$1B+ in customer contracts booked before launch signals strong early market interest.
- +Vertical integration from chip to data center ensures optimized performance and control.
- −No shipped product or independent benchmarks yet, making performance claims unverifiable.
- −Inference-only focus is a major limitation for teams needing flexibility for training.
- −Lack of ecosystem and software tooling forces heavy customization and expertise.
- −Reported 56k tokens/sec is only from FPGA demo, not the real A0 silicon.
- −Very high cost of entry with contact-based pricing, likely out of reach for SMBs.
- • Custom software development and integration likely required
- • Potential long-term contracts with minimum commitment
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: September 2026
How we score →Key Features
- Custom A0 silicon fabricated on TSMC N4P
- Low Voltage Inference (LVI) for 80%+ peak FLOPs utilization without thermal throttling
- Cluster Scale Memory (CSM) hybrid HBM/SRAM shared memory pool
- Proprietary ultra-low-latency, high-bandwidth scale-up interconnect
- Optimized for many-trillion-parameter sparse MoE models
- Supports long-context and agentic workloads
- Co-designed racks with cold plates, PCBs, and power delivery
- Custom scheduling and network tiling algorithms
- Vertical integration: chip, rack, factory, data center
- Scalable to gigawatt-scale deployments
- Optimized for both prefill and decode workloads
- First rack-scale products shipping summer 2025
About Etched AI
Etched builds a new category of AI hardware: frontier inference clusters. Instead of selling a standalone chip, the company co-designs chips, racks, software, and manufacturing so that the largest transformer models can run with best-in-class throughput, latency, cost, and power efficiency for both prefill and decode. The A0 silicon, fabricated on TSMC N4P, targets many-trillion-parameter sparse MoE models, long-context agents, and autonomous systems. It is built for AI labs, cloud providers, and enterprises that have hit the thermal and memory limits of general-purpose GPUs like H100 or B200. Two architectural breakthroughs define Etched's approach. Low Voltage Inference (LVI) runs the chip's math blocks at under half the voltage of most AI chips, enabling sustained 80%+ peak FLOPs utilization without thermal throttling on trillion-parameter sparse MoEs. Cluster Scale Memory (CSM) creates a hybrid HBM/SRAM shared memory pool across a scale-up domain, connected by a proprietary ultra-low-latency, high-bandwidth interconnect, delivering SRAM-like decode speeds with HBM-scale capacity. These are not incremental tweaks; they require co-designing everything from the transistor to the token, including splittable math arrays, power delivery networks, cold plates, and scheduling algorithms. Etched has raised $800M across four financings, including a strategic investment from VentureTech Alliance, and recently announced a $700M raise at a $21B valuation led by Jane Street. The company is vertically integrated, with a Taiwan factory, a San Jose data center, test house, and NPI prototyping lab. Its first rack-scale products ship summer 2025, with over $1B in customer contracts already booked. Advisors include Geoffrey Hinton, Andrej Karpathy, and Tri Dao. Unlike GPU clusters that juggle training and inference, Etched is purpose-built for inference only. That specialization is both its promise and its limitation. For teams running frontier transformer workloads at extreme scale, Etched may offer a path to dramatically lower cost and latency, but it's a high-stakes bet that requires patience and substantial resources.
Behind the Verdict
Etched is one of the most ambitious hardware companies to emerge from the AI boom. Its core thesis—that inference, not training, will become the dominant compute bottleneck—is well-aligned with industry trends. The company's co-design approach, from chip to rack to scheduling, is a genuine differentiator. The two architectural breakthroughs, LVI and CSM, directly address the thermal and memory bottlenecks that throttle GPU inference today. If the benchmarks hold, Etched could offer a step-change in cost and latency for trillion-parameter MoE models. However, the risks are equally high. The company is pre-production, with no public benchmarks. The hardware is not yet available, and the software ecosystem is unproven. Teams will need to invest significant time and resources to port their workloads, and there's no guarantee of compatibility with existing CUDA-based tooling. The $21B valuation, led by Jane Street, signals strong investor confidence but also sets a high bar for execution. For whom is Etched a fit? AI labs running frontier inference at extreme scale, hyperscalers seeking alternatives to GPU clusters, and enterprises planning AGI-ready compute infrastructure. These are organizations with the engineering depth to co-design with Etched and the financial runway to absorb delays. For small startups or teams needing production-available hardware today, Etched is not yet a viable option. GPU clusters, or specialized inference providers like Groq or Cerebras, may be safer bets. In summary, Etched is a bold bet that could pay off handsomely for early adopters, but it's not for the cautious. Keep an eye on their summer 2025 launch and public benchmarks; that will be the moment of truth.
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.
Need to run a 3-trillion-parameter MoE model for a research project but hitting GPU memory and latency limits.
Outcome: Collaborate with Etched to pilot a rack, port the model using their custom tooling, and achieve faster inference with lower power draw, enabling larger experiments.
Run a large-scale, real-time chatbot inference service that is consuming significant GPU resources and hitting thermal throttling.
Outcome: Replace a GPU cluster with Etched's front-line inference clusters, reducing per-token cost and latency while improving power efficiency.
Exploring AGI-ready computing infrastructure but concerned about the environmental and cost impact of GPU farms.
Outcome: Evaluate Etched's vertical integration and gigawatt-scale scaling plans; prototype on their hardware to assess long-term viability.
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 2026-09-22
Limitations
- Etched is a hardware company building custom silicon and inferencing clusters for frontier models.
- It is currently validating its first rack-scale product with customers and has not published public pricing.
- The company has raised $800M across four unannounced financings, including a strategic investment from VentureTech Alliance, and demonstrated A0 silicon on TSMC N4P.
- Specific limitations on software tooling and availability are not detailed in the provided evidence.
as of 2026-08-30
Verification history
We have re-verified Etched AI 19 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-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-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 19 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, but given the target volume (racks at gigawatt scale), it aims to be cost-effective for hyperscalers and large AI labs. Compared to general-purpose GPU clusters, Etched claims lower total cost of ownership for inference workloads. However, without public benchmarks, direct cost comparison remains speculative.
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 early access, expect weeks to months of contract negotiation and hardware provisioning. Pilot programs with Etched involve co-design and testing, so time to first value could be 3-6 months or more.
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 NVIDIA GPU clusters: Plan for custom model compilation and optimization using Etched's software stack, which will require dedicated engineering effort.
- ↗To NVIDIA GPU clusters: Likely to require re-optimization for CUDA if you ever need to move back, as Etched's proprietary architecture won't support GPU software directly.
Resources & Guides
Tutorials & Learning
YouTube returned 2 videos for “Etched AI”, and we withheld 2: 2 could not be judged, because “Etched AI” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Etched AI.
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.
Anyscale Endpoints
Anyscale Endpoints runs distributed training, batch inference, and multimodal data curation on managed Ray GPU clusters.
Groq
Groq is an inference neocloud built for sub-200ms LPU inference — fast open-weight model serving for real-time chat, voice, and agent workloads.
Cerebras
Cerebras delivers ultra-fast AI inference on wafer-scale hardware for latency-critical agents and apps.
Alternatives to Etched AI
View all
Anyscale Endpoints
Anyscale Endpoints runs distributed training, batch inference, and multimodal data curation on managed Ray GPU clusters.
Frequently Asked Questions
Categories
Best-of guides
Topics
Used Etched AI? Help shape our editorial sentiment research.