WEKA Debuts NeuralMesh Axon For Exascale AI Deployments
PARIS and CAMPBELL, Calif., July 8, 2025 /PRNewswire/ -- From RAISE SUMMIT 2025: WEKA unveiled NeuralMesh Axon, a breakthrough storage system that leverages an innovative fusion architecture designed to address the fundamental challenges of running exascale AI applications and workloads. NeuralMesh Axon seamlessly fuses with GPU servers and AI factories to streamline deployments, reduce costs, and significantly enhance AI workload responsiveness and performance, transforming underutilized GPU resources into a unified, high-performance infrastructure layer.
Building on the company's recently announced NeuralMesh storage system, the new offering enhances its containerized microservices architecture with powerful embedded functionality, enabling AI pioneers, AI cloud and neocloud service providers to accelerate AI model development at extreme scale, particularly when combined with NVIDIA AI Enterprise software stacks for advanced model training and inference optimization. NeuralMesh Axon also supports real-time reasoning, with significantly improved time-to-first-token and overall token throughput, enabling customers to bring innovations to market faster.
AI Infrastructure Obstacles Compound at Exascale
Performance is make-or-break for large language model (LLM) training and inference workloads, especially when running at extreme scale. Organizations that run massive AI workloads on traditional storage architectures, which rely on replication-heavy approaches, waste NVMe capacity, face significant inefficiencies, and struggle with unpredictable performance and resource allocation.
The reason? Traditional architectures weren't designed to process and store massive volumes of data in real-time. They create latency and bottlenecks in data pipelines and AI workflows that can cripple exascale AI deployments. Underutilized GPU servers and outdated data architectures turn premium hardware into idle capital, resulting in costly downtime for training workloads. Inference workloads struggle with memory-bound barriers, including key-value (KV) caches and hot data, resulting in reduced throughput and increased infrastructure strain. Limited KV cache offload capacity creates data access bottlenecks and complicates resource allocation for incoming prompts, directly impacting operational expenses and time-to-insight. Many organizations are transitioning to NVIDIA accelerated compute servers, paired with NVIDIA AI Enterprise software, to address these challenges. However, without modern storage integration, they still encounter significant limitations in pipeline efficiency and overall GPU utilization.
Built For The World's Largest and Most Demanding Accelerated Compute Environments
To address these challenges, NeuralMesh Axon's high-performance, resilient storage fabric fuses directly into accelerated compute servers by leveraging local NVMe, spare CPU cores, and its existing network infrastructure. This unified, software-defined compute and storage layer delivers consistent microsecond latency for both local and remote workloads—outpacing traditional local protocols like NFS.
Additionally, when leveraging WEKA's Augmented Memory Grid capability, it can provide near-memory speeds for KV cache loads at massive scale. Unlike replication-heavy approaches that squander aggregate capacity and collapse under failures, NeuralMesh Axon's unique erasure coding design tolerates up to four simultaneous node losses, sustains full throughput during rebuilds, and enables predefined resource allocation across the existing NVMe, CPU cores, and networking resources—transforming isolated disks into a memory-like storage pool at exascale and beyond while providing consistent low latency access to all addressable data.
Cloud service providers and AI innovators operating at exascale require infrastructure solutions that can match the exponential growth in model complexity and dataset sizes. NeuralMesh Axon is specifically designed for organizations operating at the forefront of AI innovation that require immediate, extreme-scale performance rather than gradual scaling over time. This includes AI cloud providers and neoclouds building AI services, regional AI factories, major cloud providers developing AI solutions for enterprise customers, and large enterprise organizations deploying the most demanding AI inference and training solutions that must agilely scale and optimize their AI infrastructure investments to support rapid innovation cycles.
Delivering Game-Changing Performance for Accelerated AI Innovation
Early adopters, including Cohere, the industry's leading security-first enterprise AI company, are already seeing transformational results.
Cohere is among WEKA's first customers to deploy NeuralMesh Axon to power its AI model training and inference workloads. Faced with high innovation costs, data transfer bottlenecks, and underutilized GPUs, Cohere first deployed NeuralMesh Axon in the public cloud to unify its AI stack and streamline operations.
"For AI model builders, speed, GPU optimization, and cost-efficiency are mission-critical. That means using less hardware, generating more tokens, and running more models—without waiting on capacity or migrating data," said Autumn Moulder, vice president of engineering at Cohere. "Embedding WEKA's NeuralMesh Axon into our GPU servers enabled us to maximize utilization and accelerate every step of our AI pipelines. The performance gains have been game-changing: Inference deployments that used to take five minutes can occur in 15 seconds, with 10 times faster checkpointing. Our team can now iterate on and bring revolutionary new AI models, like North, to market with unprecedented speed."
To improve training and help develop North, Cohere's secure AI agents platform, the company is deploying WEKA's NeuralMesh Axon on CoreWeave Cloud, creating a robust foundation to support real-time reasoning and deliver exceptional experiences for Cohere's end customers.
"We're entering an era where AI advancement transcends raw compute alone—it's unleashed by intelligent infrastructure design. CoreWeave is redefining what's possible for AI pioneers by eliminating the complexities that constrain AI at scale," said Peter Salanki, CTO and co-founder at CoreWeave. "With WEKA's NeuralMesh Axon seamlessly integrated into CoreWeave's AI cloud infrastructure, we're bringing processing power directly to data, achieving microsecond latencies that reduce I/O wait time and deliver more than 30 GB/s read, 12 GB/s write, and 1 million IOPS to an individual GPU server. This breakthrough approach increases GPU utilization and empowers Cohere with the performance foundation they need to shatter inference speed barriers and deliver advanced AI solutions to their customers."
"AI factories are defining the future of AI infrastructure built on NVIDIA accelerated compute and our ecosystem of NVIDIA Cloud Partners," said Marc Hamilton, vice president of solutions architecture and engineering at NVIDIA. "By optimizing inference at scale and embedding ultra-low latency NVMe storage close to the GPUs, organizations can unlock more bandwidth and extend the available on-GPU memory for any capacity. Partner solutions like WEKA's NeuralMesh Axon deployed with CoreWeave provide a critical foundation for accelerated inferencing while enabling next-generation AI services with exceptional performance and cost efficiency."
The Benefits of Fusing Storage and Compute For AI Innovation
NeuralMesh Axon delivers immediate, measurable improvements for AI builders and cloud service providers operating at exascale, including:
"The infrastructure challenges of exascale AI are unlike anything the industry has faced before. At WEKA, we're seeing organizations struggle with low GPU utilization during training and GPU overload during inference, while AI costs spiral into millions per model and agent," said Ajay Singh, chief product officer at WEKA. "That's why we engineered NeuralMesh Axon, born from our deep focus on optimizing every layer of AI infrastructure from the GPU up. Now, AI-first organizations can achieve the performance and cost efficiency required for competitive AI innovation when running at exascale and beyond."
Availability
NeuralMesh Axon is currently available in limited release for large-scale enterprise AI and neocloud customers, with general availability scheduled for fall 2025. For more information, visit:
About WEKA
WEKA is transforming how organizations build, run, and scale AI workflows through NeuralMesh™, its intelligent, adaptive mesh storage system. Unlike traditional data infrastructure, which becomes more fragile as AI environments expand, NeuralMesh becomes faster, stronger, and more efficient as it scales, growing with your AI environment to provide a flexible foundation for enterprise and agentic AI innovation. Trusted by 30% of the Fortune 50 and the world's leading neoclouds and AI innovators, NeuralMesh maximizes GPU utilization, accelerates time to first token, and lowers the cost of AI innovation. Learn more at www.weka.io, or connect with us on LinkedIn and X.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Korea Herald
9 hours ago
- Korea Herald
SK hynix racks up 70% of sales in US on AI chip boom
Chip giant SK hynix has reported that its US revenue accounted for nearly 70 percent of its total sales in the first half of the year, boosted by surging demand for AI-critical high-bandwidth memory chips, helping the chipmaker cut debt and increase investments. The company reported revenues in the US of 27.83 trillion won ($20.02 billion) from January to June, representing 69.8 percent of its total sales of 39.87 trillion won, according to a regulatory filing on Sunday. The figure includes sales at its US unit, SK hynix America, which posted 24.7 trillion won in sales, more than double from a year ago when it recorded 12.2 trillion won. The sharp surge in revenue is attributed to high demand for HBM, a type of DRAM chip used for AI accelerators. SK hynix is regarded as a front-runner in the lucrative HBM market, supplying clients such as US tech giant Nvidia. 'Demand for AI memory continued to grow, driven by aggressive investments from global big tech companies,' Song Hyun-jong, president and head of the Corporate Center at SK hynix, said in the earnings call after announcing the Q2 results last month. SK hynix's sales are expected to grow further, as the company said it aims to double HBM sales and shipments this year versus last year. By contrast, the company's revenue in China fell to 7.36 trillion won in the first half, down more than 1 trillion won from 8.61 trillion won a year earlier. Buoyed by explosive growth in HBM sales, SK hynix was able to improve its balance sheet by repaying more than 800 billion won in borrowings in just six months. Total borrowing came to 21.84 trillion won at the end of June, down 842 billion won from the beginning of the year, and down 3.38 trillion won from a year earlier. With strengthened finances, SK hynix is stepping up investment in future growth. R&D spending hit 3 trillion won in the first half, already exceeding over 60 percent of 2024's total. Capital expenditure nearly doubled to 11.2 trillion won during the same period.
Korea Herald
6 days ago
- Korea Herald
ANALYSIS- 'Trump's Nvidia deal raises new corporate, national security risks'
ANALYSIS US President Donald Trump upended decades of US national security policy, creating an entirely new category of corporate risk, when he made a deal with Nvidia to give the US government a cut of its sales in exchange for resuming exports of banned AI chips to China. Historically, the US government made decisions to control the export of sensitive technologies on national security grounds. Those decisions were viewed as non-negotiable; if a technology was controlled, companies could not buy their way around those controls, no matter how lucrative the foregone foreign sales. On Monday, Trump raised the prospect of ending that era, saying he would allow Nvidia to sell its H20 chips to China in exchange for the US government receiving a 15 percent cut of the company's sales of some advanced chips in that country. He made a similar deal with Nvidia's smaller rival AMD. He also told reporters he was open to allowing Nvidia to sell a scaled-down version of its current flagship Blackwell chips to China. Months earlier, his own administration had banned the sale of H20 chips to China, reversing the decision in July as part of what the government said were negotiations on rare earths. The latest move drew condemnation from US lawmakers in both parties who warned that it risked creating a pay-for-play framework for the sale of sensitive technologies to US adversaries, a concern echoed by analysts and legal experts. "Export controls are a frontline defense in protecting our national security, and we should not set a precedent that incentivizes the government to grant licenses to sell China technology that will enhance its AI capabilities," said US Representative John Moolenaar, a Michigan Republican who chairs the House Select Committee on China. Representative Raja Krishnamoorthi of Illinois, the ranking Democrat on the same committee, said that "by putting a price on our security concerns, we signal to China and our allies that American national security principles are negotiable for the right fee." To be sure, the Trump administration has said the national security risks of resuming H20 sales are minimal because the chip was sold widely in China. US Commerce Secretary Howard Lutnick last month described the H20 as Nvidia's "fourth-best chip" in an interview with CNBC. He said it was in US interests for Chinese firms to keep using American technology. But the deal is extremely rare for the US and marks Trump's latest intervention in corporate decision-making, after pressuring executives to invest in American manufacturing and demanding the resignation of Intel's CEO, Lip-Bu Tan, over his ties to Chinese companies. It is unclear whether Trump's move is legal. The US Constitution prohibits Congress from levying taxes and duties on articles exported from any state. Trade lawyer Jeremy Iloulian said it is hard to tell if this would be considered an "export tax" or some other form of payment without knowing more about the agreement. "Up until today, there has never been a consideration of how much companies need to pay to receive an export license," Iloulian said. Added Kyle Handley, a professor at the University of California San Diego School of Global Policy and Strategy: "It sure looks like an export tax to me ... they can call it whatever they want. It really looks a lot like the government is skimming a little bit off the top." When asked if Nvidia had agreed to pay 15 percent of revenue to the US, a company spokesperson said, "We follow rules the US government sets for our participation in worldwide markets." "While we haven't shipped H20 to China for months, we hope export control rules will let America compete in China and worldwide," the spokesperson added. A spokesperson for AMD said the US approved its applications to export some AI processors to China but did not directly address the revenue-sharing agreement and said the company's business adheres to all US export controls. 'I think it's fair to say that everything now in this administration seems negotiable in ways that were not the case before," said Sarah Kreps, a professor at the Brooks School of Public Policy at Cornell University. "I don't think this is unique in that this will be the last kind of deal like this that we see.' Equities analysts said the levy could hit margins at chipmakers and set a precedent for Washington to tax critical US exports. "It feels like a slippery slope to us," said Bernstein analysts, who expect the deal to cut gross margins on the China-bound processors by 5 to 15 percentage points, shaving about a point from Nvidia and AMD's overall margins. "Naturally, not only chipmakers but also companies selling other strategic products to China will wonder if the remittance model could apply to their industries," said Hendi Susanto, a portfolio manager at Gabelli, which holds shares in Nvidia. "For sellers of strategic products to China, remittance could be a burden -- or a lifeline to preserve market access to huge and growing opportunities in China," Susanto said. (Reuters)
Korea Herald
6 days ago
- Korea Herald
Trump holds meeting with Intel CEO after demanding he resign
US President Donald Trump said he met with Intel CEO Lip-Bu Tan on Monday, days after seeking his resignation, praising Tan and calling the meeting "a very interesting one." Shares of the chipmaker rose 3 percent in extended trading. Last week, Trump had demanded the immediate resignation of Tan, calling him "highly conflicted" over his ties to Chinese firms, injecting uncertainty into the chipmaker's years-long turnaround effort. Trump said he met with Tan, along with Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent. His cabinet members and Tan were going to bring suggestions to him next week, Trump said in a post on Truth Social. "His success and rise is an amazing story," Trump said about Tan. Tan had invested in hundreds of Chinese firms, some of which were linked to the Chinese military, Reuters reported exclusively in April. It is not illegal for US citizens to hold stakes in Chinese companies unless they have been added to the US Treasury's Chinese Military-Industrial Complex Companies List, which explicitly bans such investments. Tan has been tasked to undo years of missteps that left Intel struggling to make inroads in the booming AI chip industry dominated by Nvidia, while investment-heavy contract manufacturing ambitions led to hefty losses. In the roughly six months as Intel CEO, Tan made major strategic shifts that included divesting assets, laying off employees and redirecting resources. But the demand for Tan's resignation will only distract him from that task, investors and a former senior employee have told Reuters. Tan is now making an effort to reassure Trump that he remains the right person to revive the storied American chipmaker. Tan met with Trump for a candid and constructive discussion on the company's commitment to strengthening US technology and manufacturing leadership, Intel said in a statement. The company said it would work closely with the administration to "restore this great American company." Trump's intervention marked a rare instance of a US president publicly calling for a CEO's ouster and raised questions about his control over corporate affairs. This was also evident in an agreement calling for Nvidia and AMD to give the US government 15 percent of revenue from China sales. (Reuters)
