Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteSambaNova cut 77 California employees on or around April 22, 2025—approximately 15% of a workforce estimated at 500—as it shifted emphasis from large-scale model training toward fine-tuning, inference and cloud-first deployment of open-source models. The figure comes from a California WARN filing reported by Data Center Dynamics; SambaNova separately described the reduction as “around 75 employees,” a rounded figure reported by EE Times.
What happened at SambaNova?
The layoffs were a California reduction, not proof of an exactly 15% worldwide cut. The WARN filing identified 77 affected employees and an April 22, 2025 effective date. The approximately 500-person denominator produces the commonly reported 15% figure. Public reporting does not establish how many employees outside California were affected, which departments absorbed the deepest cuts, the severance terms, the savings target or whether products and customer commitments were canceled.
SambaNova said the reorganization reflected “today’s market conditions,” a move from a primarily training-oriented business toward fine-tuning and inference, and a cloud-first effort to help customers deploy open-source models at scale. That explanation establishes the company’s stated rationale; it does not independently prove that the inference pivot alone caused every position elimination.
Training, fine-tuning and inference are different businesses
| Workload | What it does | Typical commercial concern |
|---|---|---|
| Training | Processes very large datasets while adjusting a model’s parameters. | Large, periodic and capital-intensive clusters. |
| Fine-tuning | Adapts an existing model to a company, domain, task or behavior. | More targeted workloads that overlap training and production serving. |
| Inference | Runs a trained model to generate responses for applications and users. | Latency, throughput, uptime, utilization, cost per token, power and data location. |
SambaNova’s technical paper describes training as large-scale data processing and inference as primarily a data-movement and serving challenge. That is the company’s technical framing, not an industry-neutral benchmark; see its SN40L paper.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11#1 Best Overall
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
The distinction affects revenue. Training hardware may be sold for occasional, enormous bursts of demand. Hosted inference can generate continuing consumption revenue as applications process requests or tokens. Buyers may also prefer a managed service to purchasing, installing and operating a specialized cluster.
Why inference was attractive to SambaNova
- Recurring usage revenue: Cloud services can bill by requests or tokens rather than relying only on system sales.
- Enterprise demand: Many companies want to use established models without building a training supercluster.
- Optimization opportunities: Inference systems can be designed around latency, batching, power, cooling and cost per token.
- Open-model adoption: SambaNova’s cloud messaging emphasizes serving models from the open-source ecosystem.
- Data-center constraints: Power availability and cooling increasingly influence where AI capacity can be deployed.
Inference is not automatically a higher-margin or easier market. Demand is price-sensitive, customers can switch providers, and a provider must operate reliable capacity, software, support and networking in addition to designing silicon.
SambaNova’s product strategy
SambaNova Cloud
Announced in September 2024, SambaNova Cloud offered API access to models including Llama 3.1 8B, 70B and 405B on its SN40L processor, with free, developer and enterprise tiers. In February 2025, the company said its paid Developer Tier used token billing and included $5 in introductory credits; that historical promotion should not be treated as current August 2026 pricing. Product information is available through the launch announcement, the Developer Tier post and SambaNova Cloud.
Rank #2
- High-Performance AI Processing: The MX3 is designed to handle the most demanding AI computer vision workloads, delivering exceptional performance and efficiency.
- Flexible Integration: The MX3 can be easily integrated into your existing systems via its M.2 M-key form factor and support for Linux operating systems.
- Energy Efficient: The MX3 is designed to provide high performance while minimizing power consumption.
- Comprehensive Software Development Kit (SDK): The MX3 is supported by a comprehensive SDK that simplifies development and deployment.
- Hardware compatability: The MX3 is compatible with the PCI-SIG M.2 M-key 2280 Specification. It can be used with the Raspberry Pi 5 with a M-key 2280 HAT.
SambaCloud, SambaStack and SambaManaged
In July 2025, SambaNova described a three-part portfolio: SambaCloud for hosted inference, SambaStack for enterprise AI infrastructure and SambaManaged for a managed inference cloud deployed in a customer or partner data center. The SN40L RDU chip underpins the platform. The strategy is therefore broader than “sell a chip”: it combines silicon, systems, software, cloud operations and deployment services. See SambaNova’s portfolio description.
Recommended Free Tools
SambaManaged is marketed as a turnkey service for data-center operators. SambaNova advertises deployment in roughly 90 days on its product page, while a related datasheet says “as little as 30 days.” These are marketing estimates, not guaranteed implementation schedules. The product page is available here; the datasheet is here.
RDU roadmap
SambaNova’s current product page calls the SN50 its fifth-generation inference processor and claims five times more compute and four times more network bandwidth than the fourth-generation SN40. Those are vendor specifications, not independently verified performance results; see the RDU product page.
Rank #3
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
Did SambaNova abandon training?
No. The reported shift was away from a training-led emphasis, not an announcement that training hardware or model development had ended. SambaNova’s later materials continue to describe training, fine-tuning and deployment use cases, including work associated with Argonne National Laboratory; see the company’s account. A customer can therefore use the reorganization as evidence of changed priorities, but not as proof of a complete exit from training.
Competitive and financial pressure
SambaNova competes in a market led by Nvidia GPUs and also faces AMD accelerators, hyperscalers’ in-house chips, Groq, Cerebras and other specialized providers. Nvidia’s advantage includes a broad software ecosystem and general-purpose compatibility. Specialized silicon can be compelling for selected models or serving patterns, but customers must consider model support, framework compatibility, capacity, migration costs and vendor reliability.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
SambaNova has promoted SN40L systems as air-cooled and substantially less power-intensive than conventional GPU racks. In an October 2025 announcement, it claimed 10 kW per rack versus up to 120 kW for traditional GPU systems. Those figures are company claims, not an independent comparison; see the announcement. Its broader benchmark claims involving GPUs, Groq and Cerebras are likewise vendor-authored; see SambaNova’s benchmark article.
At the time of the layoffs, reporting placed SambaNova’s total funding above $1.1 billion and its valuation above $5 billion after the 2021 Series D. Those figures indicate substantial capital backing but do not reveal profitability or cash burn.
What the layoffs do—and do not—prove
What they do indicate
- A meaningful organizational reallocation toward inference, cloud services and managed infrastructure.
- Management believed the existing workforce mix no longer matched its near-term market priorities.
- The company was trying to align its cost base with a different revenue model and customer set.
What they do not establish
- They do not prove insolvency, product failure or that SambaNova was leaving the chip business.
- They do not establish a precise global headcount reduction because the filing covered California employees.
- They do not prove that inference had universally replaced training across the AI industry.
- They do not show that the pivot caused the layoffs independently of broader market and operating pressures.
What happened after the layoffs?
| Date | Development |
|---|---|
| July 2025 | SambaNova presented SambaCloud, SambaStack and SambaManaged as a unified strategy. |
| October 2025 | The company announced sovereign-AI deployments involving Australia, Europe and the United Kingdom. |
| February 2026 | Reuters later reported that SambaNova had raised $350 million. |
| July 2026 | Reuters reported a $1 billion round led by General Atlantic at an $11 billion post-money valuation, plus an inference-infrastructure relationship with JPMorgan Chase. |
The later financing and product expansion show that SambaNova continued pursuing inference infrastructure after the reduction. They do not prove that the 2025 reorganization produced profitability, market leadership or a guaranteed commercial outcome. The financing report is available through Reuters coverage republished by Investing.com.
What an enterprise buyer should evaluate
SambaNova’s shift matters most to organizations deciding how to purchase inference capacity. A responsible comparison should examine:
Best Value
- Memory Size: 16 GB GDDR6 ECC.
- Memory Bus Width: 128-bit.
- Memory Bandwidth: 200 GB/s.
- CUDA Cores: 1280.
- Peak Single Precision floating point performance: 18 Tflops (GPU Boost Clocks).
- Cost per token at the buyer’s actual model, context length and utilization.
- Latency, throughput, batching behavior and service-level guarantees.
- Supported models, frameworks, quantization methods and portability.
- Cloud, on-premises and managed-data-center deployment choices.
- Data residency, sovereignty, security controls and operational support.
- Power, cooling, networking and capacity commitments.
- Migration options if pricing, availability or the vendor roadmap changes.
Nvidia remains the broadest default for mixed training, fine-tuning and inference requirements; its data-center starting point is Nvidia’s official site. AWS, Microsoft Azure and Google Cloud offer consumption-based accelerator capacity. Groq and Cerebras may suit selected low-latency or high-throughput workloads, while self-managed open-source serving with tools such as vLLM can prioritize portability. Current public information is insufficient for a responsible apples-to-apples price comparison with SambaNova’s enterprise offerings.
Timeline
- 2017: SambaNova was founded.
- April 2021: It raised a reported $676 million round led by SoftBank Vision Fund 2.
- September 10, 2024: SambaNova announced SambaNova Cloud.
- February 8, 2025: It announced the paid Developer Tier.
- April 22, 2025: The California WARN filing date for 77 layoffs.
- April 25, 2025: EE Times reported the company’s explanation of the cuts.
- July 8, 2025: SambaNova outlined its SambaCloud, SambaStack and SambaManaged portfolio.
- October 22, 2025: It announced sovereign-AI partnerships.
- July 8, 2026: Reuters reported the $1 billion financing at an $11 billion valuation.
The Bottom Line
SambaNova’s April 2025 reduction was a 77-person California layoff—about 15% of an estimated 500-person workforce—paired with a strategic move toward recurring inference, cloud and managed-infrastructure revenue. It signaled a serious reorganization and operating pressure, but not a shutdown or complete abandonment of training. The company’s later fundraising and continued product expansion show that the inference strategy remained active, while leaving profitability and long-term competitive success unresolved.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
