Neo Semiconductor has developed three-dimensional DRAM with added neuron circuitry that it says will accelerate AI processing by avoiding data transfers from high-bandwidth memory to GPUs.
Neo’s 3D DRAM technology is the basis for its 3D X-AI 300-layer, 128 Gbit DRAM chip, with 8,000 neurons and 10 TBps of AI processing per die. 3D X-AI chip capacity and performance can be scaled 12x with up to 12 x 3D X-AI dies stacked like high-bandwidth memory (HBM), providing 192 GB (1,536 Gb) capacity and 120 TBps processing throughput.
Andy Hsu.
A statement from Andy Hsu, Founder and CEO of NEO Semiconductor, said: “Typical AI chips use processor-based neural networks. This involves combining high-bandwidth memory to simulate synapses for storing weight data and graphical processing units (GPUs) to simulate neurons for performing mathematical calculations. Performance is limited by the data transfer between HBM and GPU, with back-and-forth data transfer lowering AI chip performance and increasing power consumption.”
3D X-AI simulates artificial neural networks (ANNs), including synapses for weight data storage and neurons for data processing, making it, Neo claims, ideally suited to accelerate next-generation AI chips and applications.
Hsu added: “AI chips with 3D X-AI use memory-based neural networks. These chips possess neural network functions with synapses and neurons in each 3D X-AI chip. They are used to drastically reduce the heavy workload of data transfer data between GPU and HBM when performing AI operations. Our invention drastically improves AI chip performance and sustainability.”
NAND suppliers such as SK hynix and Samsung have experimented with computational memory but it has not become mainstream technology and productized. The use cases are too niche to justify mass production.
Neo will be hoping that AI processing is going to become so widespread that it leaves such nicheness far behind.
An alternative approach to accelerating AI processing is to use specialized processors such as Google’s TPU and Groq’s Tensor Stream Processor (TSP). The Neo 3D X-AI chip can be used with standard GPUs, with the claim of providing faster AI processing with less expense. Now Neo has to find AI processing system or system component builders willing to take a bet on its technology.
Neo will present its 3D X-AI technology at FMS 2024 in Santa Clara, August 6 – 7, and showcase it at booth 507. Check out a 3D X-DRAM slide deck here.
DapuStor storage systems will run NoSQL databases faster through a Pliops data processing unit partnership, we’re told.
Israel-based Pliops has developed its XDP (Extreme Data Processor) AccelKV, with key:value store technology. This PCIe card offloads and accelerates low-level storage stack processing from a host x86 server for applications like RocksDB and RAID. Pliops has recently partnered with Hammerspace and announced it is in merger talks with French DPU developer Kalray. DapuStor is a Chinese firm, based in Shenzhen, supplying SSDs with its own ASIC controllers, and storage systems. It has more than 500 customers and was the number four enterprise SSD supplier in China in 2023.
Tony Afshary
Tony Afshary, Global VP of Products and Marketing at Pliops, said in a canned quote: “We’re thrilled to partner with DapuStore and accelerate their best-in-class NVMe storage solutions.”
The two companies are pushing the use of KVRocks, an open source key-value store alternative to Redis, which is built on top of the RocksDB engine. KVrocks users in China include Baidu, Trip.com Group, Meitu, RGYUN, U-NEXT, and Baishan Cloud. The pair claim that, at petabyte levels, RocksDB meets performance and latency constraints. The Pliops XDP card and KVrocks software combo does not suffer from this, we’re told. Pliops’ XDP-Rocks library extends RocksDB functionalities, optimizing KVrocks integration with features like more efficient data manipulation and advanced dataset scanning.
DapuStor solution architect Grant Li said in a statement: “The DapuStor R5101 is a high-performance PCIe4.0 NVMe SSD that offer superior reliability, low latency, and excellent power efficiency, optimising TCO for enterprise IT and cloud facilities.”
“It achieves 1,750,000 random read and 240,000 random write IOPS” at a 3.84TB capacity point with a PCIe gen 4 x 4 NVMe interface. Li claimed that, when “combined with Pliops’ XDP and library, the solution delivers a high-performance key-value storage interface.”
The Dbbench database benchmark tool was used to compare R5101 SSDs and Pliops Accel KV performance for KVRocks against RocksDB. This assessment involved a substantial dataset comprising 1.2 billion objects, and showed that Pliops KV outperforms traditional RocksDB by a factor of two.
Specifically, at least according to the claimed benchmark, tail latency drops between 66 percent and 87 percent in the mixed workload with 50 percent read and 50 percent write. It declined between 10 percent and 28 percent in a workload with 70 percent read and 30 percent write.
LI said this “brings a 53 percent reduction in price performance due to performance acceleration.”
Assuming the Kalray-Pliops merger/acquisition goes ahead then Kalray will inherit the DapuStor-Pliops partnership, plus the Hammerspace-Pliops integration and Pliops’ HPE server use validation. The DapuStor deal could also provide Kalray with a road into the Chinese market for its own DPU technology and products.
CTERA has raised $80 million in primary and secondary funding from private investor PSG Equity.
A week after cloud file services supplier Nasuni secured a majority investment of up to $501 million, with some existing shareholders replaced by three new equity groups, competitor CTERA has brought in fresh funding.
Liran Eshel.
Liran Eshel, founder and chairman of CTERA, claimed in a statement: “We believe CTERA is setting the standard for the modern hybrid data platform, with military-certified security and unparalleled performance. The strategic partnership with PSG will enable us to further drive our expansion while delivering top service to our customers, and implement our vision for AI data services.”
PSG Equity managing director Ronen Nir – who joins the CTERA Board – said: “We are excited to partner with CTERA as they embark on capturing, what we believe will be an inflection point in the hybrid cloud data market, leveraging both organic and inorganic opportunities to strengthen CTERA’s position and deliver even greater value to its customers.”
The two themes we see here are taking advantage of the current AI opportunities and growing CTERA by acquiring other businesses; the inorganic opportunities Nir mentioned.
Ronen Nir.
CTERA reckons that unstructured data represents about 80 percent of organizational data distributed across branch offices, endpoints, on-premises, and cloud data centers. Hybrid cloud file storage is one of the fastest-growing segments, with adoption rates expected to triple from 20 percent in 2023 to 60 percent in 2027, according to Gartner. New enterprise AI platforms depend on timely access to corporate data for training models and for augmented data retrieval (RAG) to ensure relevance and accuracy.
PSG’s $80 million is split between primary and secondary funding. Primary funding goes to CTERA to spend on growing its business. Secondary funding is used to buy out existing shareholders. This can include existing VC investors needing an exit and employees wanting cash for their share options. We asked CTERA’s chairman and co-founder Liran Eshel about this split, AI and inorganic opportunities.
Blocks & FiIes: What are the proportions of the primary and secondary funding in the $80 million PSG Equity investment?
Liran Eshel: The company is not going into the details of the primary/secondary breakdown. The main purpose of the secondary was to buy out existing long-time shareholders, for example Venrock, that have left the board.
Blocks & FiIes: How will CTERA enable its data stores’ content to be used in RAG for LLMs?
Liran Eshel: AI foundation models require access to data to make them relevant and accurate. While general LLMs are trained over public data, to create the corporate AI brain, organizations need to feed it with their private data.
This is not as trivial as it seems. First, you need to gain access to the data, and as we know, 80 percent of it is unstructured and much of it is distributed across edge locations. The next challenge is AI ingestion, which can be highly intensive in terms of compute and networking if not designed properly. And lastly, there is security – connect AI to your private data, and it is hacker’s heaven. Just ask any question and get all the classified information summarized for you.
The CTERA Data Intelligence Service is intended to solve these exact problems. The CTERA hybrid cloud data platform, and the investments we have been making in the last two years in content services and DataOps, place us in a unique position to provide a best-in-class solution to this challenge of connecting large scale unstructured data repositories with AI.
See more information in the architecture diagram below that shows how we address the problem using distributed ingestion and end-to-end permission enforcement:
Blocks & FiIes: What inorganic opportunities is Ronen Nir talking about?
Liran Eshel: “CTERA is not sharing additional information at this time about non-organic opportunities.”
***
CTERA’s last fund raise was six years ago, with a $30 million D-round taking total funding to around $105 million. In April 2022 Eshel told us it was close to becoming profitable and self-funding. He then had no plans to raise more VC funding. Now things have changed.
Its main cloud file services competitor Nasuni has had a transition to private equity ownership. PSG has holdings in a large number of other companies and inorganic opportunities may exist amongst them.
The third competitor in the cloud file services startup trio is Panzura. It was bought by private equity in 2020 and recently appointed a new CEO, Dan Waldschmidt, to galvanise growth.
Data protector Acronis says its flagship Acronis Cyber Protect Home Office product is reverting to its original name – Acronis True Image – with a new version release. It has enhanced performance and cyber security – real-time protection, illicit crypto mining protection, vulnerability assessments, and antivirus scans. Acronis True Image is specifically designed for home users.
…
New guidelines designed to provide a robust framework for data sharing in smart cities have been published as a British standard by BSI. “Guidance to establishing a decision-making framework for sharing data and information services” (BS ISO IEC 17917) is the international adoption of “Smart cities – Guide to establishing a decision-making framework for sharing data and information services” (PAS 183:2017). It was developed at the request of the Cities Standards Institute in part to support a transparent approach to making decisions. It was designed to support “smart city decision-makers” to create data and service assets, used to improve the quality of life for citizens and visitors.
…
CloudCasa by Catalogic announced its partnership with IONOS Cloud to deliver enhanced Kubernetes data protection, migration, and disaster recovery solutions for IONOS Cloud platform users. CloudCasa will integrate cloud data protection and migration services with the cloud infrastructure of IONOS to ensure its cloud-native applications are safeguarded against data loss and cyber threats.
…
There is a live FCIA webinar, “Fibre Channel Data Center Interconnects (DCI): 64G FC and More,” on July 18, 2024, when experts will discuss how Fibre Channel can be effectively deployed in these applications using optical transport technology. Topics to be covered include:
Optical Transport 101: A brief survey of optical networking (DWDM) technology;
Fibre Channel transport via OTN;
Robustness: Optical transport protection options: Resiliency against fiber outages;
Security: Optical transport OTN encryption: No-overhead layer 1 encryption that can be combined with IPSec or other higher-layer security protocols.
Remember crashed Nyriad and its GPU-based storage controller? We were sent this message from Derek Dicker, Nyriad CEO: “I’m reaching out to provide you with a final update on the Nyriad sale process. With the assistance of a third party, the Nyriad Board has completed three transactions for the sale of the majority of the company’s assets:
The first transaction involved the sale of Nyriad’s patents;
The second transaction entailed the sale of the vast majority of Nyriad’s remaining system inventory; and
The third transaction secured support from a third party for Nyriad’s UltraIO customers, who requested to maintain systems through the balance of their support contracts.
“We are now in the process of paying all of the company’s creditors. The management of the company will soon be transitioned to a third party who will complete a managed wind-down process.”
…
Red Hat announced GA of its Kubernetes-powered Red Hat OpenShift 4.16 with:
Metro disaster recovery provides regional DR for VMs that use storage deployed on OpenShift Data Foundation in conjunction with Advanced Cluster Management for Kubernetes for management.
Hot-add CPU provides users the ability to add additional vCPU resources to a running VM in a declarative manner for improved memory density with safe memory overcommit, and enables users to more easily scale up VMs with CPU hotplug.
Multi-cluster virtualization monitoring with Red Hat Advanced Cluster Management enables users to view all VMs across multiple Red Hat OpenShift clusters as well as collect and more quickly build reports for the VMs.
Image-based updates (IBU) for single node OpenShift. Single node OpenShift users can now shift a large portion of the update process to a pre-production environment, which reduces the time spent updating at the production site. Additionally, if an update fails or the application doesn’t return to a functioning state, it can be rolled back to the pre-update state.
The OpenShift-based Appliance Builder is now available as a technology preview to Red Hat partners seeking to build turnkey, customized appliances with self-contained Red Hat OpenShift instances.
…
Red Hat released its annual State of Kubernetes Report. The report revealed that nearly half (46 percent) of all organizations worldwide lost revenue or customers last year as a result of a container or Kubernetes security incident. Other findings include:
Almost all organizations (89 percent) had at least one container or Kubernetes related incident. Despite this, just under half of those surveyed (42 percent) said that their company does not invest enough in container security.
When it comes to risks, IT specialists are most worried about a vulnerability in the environments (33 percent), misconfigurations (27 percent) and external attacks (24 percent).
…
SQream announced TPC-DS benchmark test results, showing linear scalability of its GPU-accelerated SQL engine. These results show perfect linear scalability, when running both 1TB and 10TB using the same machine and maintaining the same ratio in runtime throughout. Conducted using infrastructure with 4 Tesla V100 GPU processors and 4 Intel Xeon Gold 5220 (18 cores each), the tests included 1TB and 10TB datasets.
Loading Time: 1TB in 892 seconds (14:52 minutes) and 10TB in 8260 seconds (2:17:40 hours).
Query Time: 1TB in 1374 seconds (22:54 minutes) and 10TB in 13083 seconds (3:38:03 hours).
…
Data warehouser Teradata announced the integration of the DataRobot AI Platform with Teradata VantageCloud and ClearScape Analytics to help enterprises maximize their AI potential by providing greater optionality and flexibility for building and scaling safe and effective AI models. The new functionality is available through ClearScape Analytics’s Bring Your Own Model (BYOM) capability. The features of this integration include:
DataRobot is a full AI lifecycle platform built for predictive and generative AI models with choice and flexibility.
Teradata VantageCloud allows enterprises to deploy DataRobot AI models at scale, in the same environment where the data resides. Deployment can be done across all cloud providers and on premises.
Using BYOM, users of VantageCloud and DataRobot have access to models built in their DataRobot AI Platform, from within VantageCloud. This includes importing DataRobot models into VantageCloud for inferencing at scale using ClearScape Analytics’s model scoring functions. As a result, DataRobot models can be scaled easily and cost effectively for Teradata VantageCloud users as there is no additional license cost.
…
TrendForce Japanrevealed a NAND supplier market share chart for Q1 2024:
…
Datacenter virtualizer VergeIO announced a new VAR program with a superior product, seamless migration capabilities, financial incentives, technical assurance, and an optimized sales process. It says this strategic initiative positions VARs to effectively guide their customers through the post-VMware landscape.
…
Nvidia recently unveiled a new reference architecture for AI cloud providers, in order to provide a “proven blueprint for cloud providers to stand up and manage high-performance scalable infrastructure for AI datacenters.” The RA enables Cloud Partners within Nvidia’s Partner Network to reduce the time and cost of deploying AI systems while ensuring compatibility and interoperability among various hardware and software components. VAST Data announced its Data Platform is one of the first validated solutions for Nvidia Partner Network cloud partners, simplifying AI infrastructure-at-scale with a trusted reference architecture for service providers to build high-performance, scalable and secure datacenters that can handle GenAI and LLMs.
…
YMTC is suing Micron, alleging infringement of 11 of its patents and asking for an injunction on Micron and/or royalties related to Micron’s claimed breaches. The case is docket 24-cv-04223 in the Northern District of California. Wedbush analysts note MTC’s technology involves bonding a separate CMOS wafer containing the circuit layer with the memory die, whereas Micron’s circuit layer is integrated under the memory die. It wonders if the case could affect Kioxia and WD with their BiCS8 technology which also involves bonding separate memory and circuit dies together. See Bloomberg Law. There are previous legal disputes between YMTC and Micron against a background of US technology export restrictions.
SaaS data protector Druva has launched a Threat Hunter service to scan customers’ global data estates for malware signs.
It’s also announcing expanded global availability of its Managed Data Detection and Response (Managed DDR) to monitor customer backups for faster detection of threats and responses to them. Druva says it’s threat hunting and monitoring backups inside a gap between a customer’s security perimeter and their production environment. Because of this, “customers can accelerate incident response, minimize downtime, and prevent data loss.”
Jaspreet Singh.
Jaspreet Singh, CEO and co-founder of Druva, said in a statement: “With today’s more advanced and persistent threats,we need to go beyond perimeter-based security. Cyber security needs to be complemented with the power of data to handle these risks. Druva’s 100 percent SaaS approach seamlessly consolidates and contextualizes data across all workloads, enabling customers to bolster cyber resilience and accelerate incident response.”
The threat hunting service looks for so-called indicators of compromise (IOCs) – such as specific file extensions or file patterns – and provides contextual data insights throughout incident response (IR) workflows to understand, remediate, and recover from critical incidents. A granular log of data changes and audit trails helps IR teams to analyze incidents. Users can perform analysis to understand if sensitive data has been compromised, and if compliance regulations have been violated.
Its Managed DDR process scans multiple backups to create a curated, clean snapshot and identify the most recent, clean version of each file – minimizing data loss, ensuring secure recovery, and accelerating the recovery process. Druva says that its Managed DDR offering provides:
24x7x365 monitoring of backups for early threat detection;
Expert analysis by Druva incident response to provide data insights for anomalous behavior;
Pre-built response runbooks and automatic lockdown of backups to safeguard data;
Expedited support and expert assistance to customer IR teams during cyber recovery.
A threat hunter blog and a separate Managed DDR blog provide background information.
Competitor Rubrik announced its threat hunting capability in December 2021. This scanned backups, not a customer’s data estate, looking for mlalware attack footprints. Cohesity has similar functionality with its CyberScan application on the Cohesity Marketplace, which can uncover cyber exposures and blind spots in a Cohesity production environment by running on-demand and automated scans of backup snapshots against known vulnerabilities.
Storage software biz DataCore has loaded up its AI development coffers with $60 million to fuel the integration of AI technologies.
Dave Zabrowski.
The company, founded in 1998, is historically averse to VC funding. Dave Zabrowski became its CEO in 2018 and raised a modest $9.7 million venture round in 2021, the same year as DataCore bought Swarm object storage software supplier Caringo to add to its SANsymphony block storage and vFilo file storage products. It bought MayaData and its OpenEBS Kubernetes (Bolt) storage in November the same year.
As a result of a deal with shared storage supplier Symply in 2022, it released Perifery, a media archiving appliance using Swarm as its object storage base. Perifery became DataCore’s edge division with AI+ services to provide preprocessing tasks at the edge of media and entertainment company workflows. Then it bought entertainment and media-focussed object storage software supplier Object Matrix in January 2023.
None of these events needed more funding as DataCore generated enough cash to sustain and develop its business. But GEN AI has changed all that, and DataCore now needs the cash to climb aboard the Gen AI and cyber-resilience trains.
Zabrowski said in statement: “This pivotal funding marks a new chapter” and “we are poised to accelerate the development of intelligent and resilient solutions in the areas of new-age data infrastructures that not only address today’s data challenges but also pave the way for future advancements. Our goal is to empower businesses with the tools they need to thrive in an increasingly complex digital landscape, driving long-term success and creating sustainable impact.”
DataCore is focused on fueling product innovation with transformative AI and cyber resilience solutions, pushing the boundaries of what is possible in data management. We asked Chief Product Officer Abhi Dey a few questions to find out more.
Abhi Dey.
Blocks & Files: Will the Perifery business move to add GEN AI LLM capabilities?
Abhi Dey: Yes, it is already moving to Gen AI as part of its Intelligent Content Engine (ICE) offering.
Blocks & Files: Will Swarm look to supply data to LLMs, possibly via retrieval-augmented generation (RAG)?
Abhi Dey: Yes, storage solutions in Swarm and Object Matrix will serve as storage endpoints for the data that is then provided to the LLMs to generate more intelligence using AI algorithms.
Blocks & Files: Will Swarm, SANsymphony, Bolt, vFilo, Object Matrix, and Perifery get cyber-resilience features added?
Abhi Dey: Yes: SANsymphony’s third mirror already provides cyber-resilience today. We are extending those capabilities to support NIS2 and KRITIS. Swarm and Object Matrix, by definition, have object immutability which implies WORM-compliant data needed for NIS2. Additionally, the SNS (Single Node Swarm, an edge-based Swarm variant for ROBO and M&E) will add MFA through its existing SAML2 integration. Bolt, now called OpenEBS Pro, has a container-level security stack integrated through the DPDK API. We are extending cyber resilience support to this solution.
Blocks & Files: Will DataCore provide its own LLM chatbot to help its users manage and optimize DataCore products?
Abhi Dey: We will leverage best-of-breed LLMs that are fit for purpose depending on the use case to help users manage their data or content.
***
The funding round was led by Vistara Growth and this is DataCore’s largest ever funding round.
Bootnote
NIS2 is a European Union cyber-security directive. The BSI Kritis Regulation defines German national facilities, installations or parts which are considered critical infrastructures and need safeguarding.
Privately owned ExaGrid says it added 137 new customers for its deduping backup target appliances in the latest quarter, claiming it added 64 six- and seven-figure deals to break its own sales records. The corporation did not reveal revenues and is not obliged to do so as it remains unlisted.
The company’s systems ingest backup data to a disk cache landing zone, from which restoration is fast, with post-ingest deduplication to a repository tier providing efficient capacity usage with restores taking longer due to rebuilding – rehydrating – the deduplicated data.
ExaGrid claims it has been Free Cash Flow (FCF) positive, P&L positive, and EBITDA positive for 14 consecutive quarters, and claims revenues grew 18 percent year-on-year in the second calendar quarter of 2024. It now has more than 4,300 active mid-market to large enterprise customers for tiered backup storage, it says.
Bill Andrews
Bill Andrews, President and CEO of ExaGrid, claimed in a statement: “ExaGrid continues to have an over 70 percent competitive win rate replacing primary storage behind the backup application as well as deduplication appliances such as Dell Data Domain (PoiwerProtect), HPE StoreOnce and NetBackup Storage Appliances.
“ExaGrid’s customer retention rate is over 95 percent, and 99 percent are on maintenance and support. These are two industry-leading customer success metrics. Our business is strong in North America, Latin America, Europe and the Middle East and we are putting an increased focus on the Asia Pacific region.”
ExaGrid has physical sales and pre-sales systems engineers in the following countries: Argentina, Australia, Benelux, Brazil, Canada, Chile, CIS, Colombia, Czech Republic, France, Germany, Hong Kong, India, Israel, Italy, Japan, Mexico, Nordics, Poland, Portugal, Qatar, Saudi Arabia, Singapore, South Africa, South Korea, Spain, Turkey, United Arab Emirates, United Kingdom, United States, and other regions.
Andrews told us: “In all of our new logo customers, we take out low cost or primary storage disk behind the backup app about 60 percent of the time, Dell Data Domain about 30 percent of the time, HPE StoreOnce (much smaller base than Data Domain) about 10 percent of the time. The rest is small.”
Some competing vendors are upgrading their products. For example, Dell added faster PowerProtect appliances, enabled PowerProtect to back up all-flash PowerMax arrays, and introduced an AI copilot to help customers use its APEX Backup Services, all in April.
In the same month Quantum added two all-flash appliances to its DXi deduping backup target product set.
Despite this, Andrews claimed: “None of the big guys are doing any special development these days to improve backup storage:
Ingest performance
Restore performance
Scalability
Integrations with backup apps
Comprehensive security
Tiered air gap, etc. for Ransomware recovery – our retention time-lock story
Strong disaster recovery store
“Etc. etc. We have a very exciting road map for the next six quarters as we are focused only on Backup Storage.”
Italy-based Cubbit has gained a $12.5 million cash boost to extend its decentralized storage business in Europe, with data sovereignty and sustainability goals.
Decentralized or Web 3.0 storage – the Airbnb spare room rental approach to aggregating spare datacenter capacity and selling it as a storage cloud – is a lower cost alternative to classic AWS, Azure, GCP, Wasabi, etc., S3-style public cloud storage. Data is shared across storage drives in geo-distributed datacenters, with erasure coding or similar technology used to protect it and the Cubbit’s DS3 and Composer software managing the data’s distribution and access, customer and storage provider payments and so forth. There are more than 350 business customers, and Cubbit is supported by several international partners, including HPE, Equinix and Exclusive Networks.
Stefano Onofri
Stefano Onofri, co-CEO and co-founder of Cubbit, said in a statement: “Having top-tier international VCs invest in Cubbit’s geo-distributed technology is a major endorsement. Over the past few years, we have grown massively and closed key partnerships and agreements with international players such as Leonardo, HPE, and Equinix – now it’s time to take our expansion to the next level.”
Cubbit was founded in 2016, took in a $2.3 million grant in 2019 and enjoyed an $8.3 million cash raise in 2021. It has been developing its business and management software since then, and says that, with its DS3 Composer software, it combines a proprietary data orchestration platform with data fragmentation and geo-distribution technologies. The claimed result is a sovereign, hyper-resilient, flexible, and highly cost-efficient cloud storage service that protects European data.
As Cubbit sees it, new funding will enable it to grow capex-free in Europe, piggybacking so to speak on existing datacenters – initially focusing on the DACH, and French-speaking regions, as well as the United Kingdom. Part of the funding will go towards further consolidation of the application ecosystem around Cubbit’s enabling technology to support vertical projects in various industries, including aerospace, defense, cyber security, healthcare and public administration.
There were multiple contributors to this funding round: eight new investors, three existing ones and two named individual investors, making 13 investing sources – quite a contrast with Silicon Valley VCs where you might expect fewer and larger funding sources in what is a relatively small funding event.
The round was co-led by LocalGlobe, EMEA’s number one VC investor according to the 2023 Dealroom report, and ETF Partners, said to be Europe’s original sustainability investor. New investors are Verve Ventures, 2100 Ventures, Hydra (holding of Datalogic), Growth Engine, Eurenergia and Moonstone.
Returning investors are Azimut Libera Impresa SGR, CDP Venture Capital SGR through its Fondo Evoluzione, and Primo Ventures. Individual investors include Fabio Fregi, former Italy Country Manager of Google Cloud, and Joe Zadeh, former VP product at Airbnb.
Alessandro Cillario
A statement from Alessandro Cillario, Cubbit’s co-CEO and co-founder, said: “Enterprises worldwide are facing the daunting challenge of orchestrating massive amounts of data – or they will soon. They don’t need just another cloud provider – they require a cloud enabler that allows them to implement the custom IT infrastructure strategy that they are looking for. Organizations need to keep full control over their data in order to simplify their workflows and reduce costs. Cubbit is here to help them achieve what was not possible before.”
Ferdinando Sigona, partner of LocalGlobe, said, “Data generation is already one of today’s steepest exponential curves, and the generative AI wave is only poised to accelerate this further. As a result, companies of all types and sizes are facing escalating complexity and cost. At the same time, the geopolitical environment is driving investment into AI sovereignty, and we actually think that need for control will extend down the stack, all the way to data storage.
“Cubbit’s cloud storage software elegantly addresses all of these needs, and we’re excited to back Stefano, Alessandro and the whole team as they respond to the intense pull they’re experiencing from the market.”
Archival glass storage system specialist Cerabyte is bringing its technology to the USA.
Cerabyte’s tech stores data imprinted as femtosecond laser nanoscale holes in a ceramic medium layered on a glass substrate tablet that can hold data in a usable state for thousands of years. Data is written at a rate of 2 million bits per laser pulse with a 1 GB capacity on each of the tablet’s surfaces. Data is read using scanning microscopes. Tablets are stored offline in a robotic library and a prototype system has been built using commercially available components. In January, Cerabyte, based in Munich, said it was examining possibilities for VC funding or commercial partnerships to productize its technology and bring it to market.
Christian Pflaum
Christian Pflaum, co-founder and CEO of Cerabyte, said in a statement today: “A data tsunami is on the horizon – and new, trail-blazing approaches to data storage are needed to meet the looming scalability and economic requirements. Cerabyte is prepared to transform how data is stored and address the urgent cost and sustainability demands of datacenters. Our vision is to achieve $1 per petabyte per month, a cost reduction of 1000x, within the next two decades.”
Discussing the archival storage’s energy efficiency, Fred Moore, founder of Horison Information Strategies, said: “A primary objective for many data centers today is that ‘if data isn’t used, it shouldn’t consume energy.’ A staggering 60 to 80 percent of all data is archival/cold, much of which is stored on energy-inefficient HDDs. By 2025, archival/cold data will amount to 4.5 to 6 zettabytes, making it the largest storage classification category. Cerabyte is poised to be the first storage solution to address all requirements effectively.”
Cerabyte claims that, due to its media longevity and rapid access, it solves various long-term archival data storage problems, enabling the implementation of fast-retrieval active archive systems and eliminating the need to periodically migrate data from one media to another.
Steffen Hellmold
In general, the largest archival product providers and many of the largest consumers, are located in the US. Cerabyte, which appointed US-based director Steffen Hellmold to aid product commercialization, has now opened offices in Santa Clara, California, and Boulder, Colorado.
Silicon Valley is home to many VCs experienced in data storage, management and usage matters. Boulder has a long connection with long-term storage operations centered on tape and systems software including Quantum and Spectra Logic.
Cerabyte’s technology is available as a data storage system prototype and it would seem the group is priming itself for commercialization. It says it has demonstrated end-to-end functionality in target environments. Cerabyte’s prototype and technology are here.
Nutanix‘s growth is getting an assist from the move to hybrid cloud and Gen AI adoption, and it will support Nvidia’s fast GPU Direct data feed protocol.
An interview with Nutanix CEO Rajiv Ramaswami revealed that he thinks its double-digit growth rate can continue for some time – because of its superior hybrid cloud software, and the fact that generative AI provides another growth driver.
Rajiv Ramaswami.
Nutanix says that it has migrated its on-premises enterprise infrastructure capabilities for people running either virtual machines or containers to the public clouds, to provide a consistent hybrid cloud experience. We explored aspects of these points.
Blocks & Files: Could you describe Nutanix’s Kubernetes facilities across the on-premises and the public cloud environments?
Rajiv Ramaswami: We manage Kubernetes in a multi-cloud world. We expect customers to run these applications everywhere. They might be running on native EKS or AKS [and] we have a unified management plane across all of these, on-premises, and the public cloud native, including Amazon’s service, Kubernetes service, our Azure Kubernetes service program, and some of the running might be on our own platform.
Our vision is people can use and run Kubernetes clusters anywhere and we’ll be the manager. We also provide storage for blocks, files and objects as well. … If somebody is building an application on AWS, they will be able to use our storage and it’s completely cloud-native. Customers could be using our storage servers in the public cloud as an alternative to EBS.
The advantage would be twofold. First, the same Nutanizx platform will be available everywhere: cloud, across multiple clouds, on-prem, etc. So you don’t have to redo platform applications if you’re thinking of working this way. Number two is that AOS has a lot of built-in enterprise grade resiliency features. We do disaster recovery across clouds and globally. We do synchronous replication. All of those capabilities now become available to you in the public cloud. Whereas AWS, for example, with EBS, doesn’t provide these services.
What typically happens in those scenarios is that, if somebody is building an application [in the cloud], they have to manage all the resiliency aspects at the application layer. They have to build it into the app. Whereas the typical enterprise relies on the underlying infrastructure. And so we basically provide the same enterprise mission-critical storage in the public cloud [as for enterprise on-prem environments].
Blocks & Files: I believe you want to make it easier for enterprises to run databases in the hybrid cloud as well?
Rajiv Ramaswami: This is about the infrastructure layer, the platform. And that has to do with the other components that most apps need to use database. They use caching. They use messaging or streaming, for example. And this is our grand vision, forward-looking.
Today, if you look at Nutanix, we already provide a database management service. So people can manage range of databases with our platform – Oracle, SQL, Mongo, Postgres. And what we want to do is to expand that, to first of all, make that available also in the public cloud, and also expand the range of services.
So we could be managing Kafka stream or Redis for caching. The notion if you look forward is to say, we will either offer it or partner with people. We have a bunch of knowledge about EDB Postgres to be able to offer a range of what are called platform layer data services, that people can use to build these applications. And once we do that, those services can also be made available everywhere.
Blocks & Files: Will these database services move to include vector databases?
Rajiv Ramaswami: The vision is broad. We don’t have a roadmap of everything. I’m focused right now on transactional databases. And, I know where you’re going, you’re going to AI and GPT in a Box. So, absolutely, yes – but in the long term. We haven’t announced timelines or anything like that. But the vision is about really creating a set of services that are available everywhere.
We are not a [database] engine provider. Yep, we may choose to do that sometime down the road. But right now we partner with other database providers.
Blocks & Files: You store a great deal of information for your customers, wherever they happen to have their data – on-premises or in any of the public clouds. That information is going to be needed by large language models (LLMs), which are helped by retrieval augmented generation (RAG). What are you going to do to help that happen?
Rajiv Ramaswami: I’m not going to announce a new roadmap but the vision is on the right track.
That’s exactly our goal. It is exactly what is what you said, which is that our data is available. We can be the platform to manage the data. We do think data is going to be everywhere – not just in the public cloud, not just on-prem. And this whole launch of GPT in a Box was exactly to try and simplify deployment of AI applications on our platform.
Today, the scope of it is a little bit more limited, which is, we have the platform which is supposed to give us all the storage pieces. On top of that, what we are able to do is to provide automatic workplace connectivity into model repositories. So we can get into Hugging Tree. We can connect to immediate repositories. One click, people can download a model that they want, instantiate on the hardware, associate that to the GPU. … Create an inference endpoint and export that to an API to develop already – so that people will automate it.
Blocks & Files: Okay, that will take care of AI inferencing and also fine tuning. How about the training side of it? Because again, you’ve got an enormous amount of information, which you could feed to GPUs?
Rajiv Ramaswami: GPU Direct is on our roadmap. We will have GPU direct, especially for files – this is really where you need GPU Direct. It’s on the roadmap – we know what’s needed. And the other thing that’s also needed is high bandwidth I/O. We are now supporting 100 gig NICs. Clearly, you need to be able to ingest a lot of data very quickly. And we understand that and we’re enabling that. … We’re doing high bandwidth … and then a machine with large memory also. All of these things play a role.
Blocks & Files: Do you think that AI inferencing has got to deliver accurate, complete, and not fake results?
Rajiv Ramaswami: Absolutely. You’re going to need to have accuracy in the models. Absolutely. And this is one of the things that there’s a lot of work required to make sure you’re getting that accuracy.
A lot of the initial use cases for Gen AI are going to be assisted use cases … that’s what we mean by Copilot in a broad sense. In other words, use it, but verify before you actually do it. So I’ll give you one example of a use case that we’re doing internally. It’s in customer support.
We’ve deployed models using our design documents, using our knowledge base articles. When the support engineer gets an incoming customer request, he or she types it into our GPT engine. And it makes recommendations in terms of “it could be this problem here, and this is what you might want to do.” And the idea is that it speeds up our time to respond and increases our productivity of our support engineer. So it’s better to serve the customer and better for us.
And this is working well. We are just finishing up our pilot for them. But the most important thing we had to do here was to keep training and fine tuning it till we could get reasonable accuracy.
Blocks & Files: My thinking is that in general, the RAG approach absolutely has to work or AI inferencing dies a death.
Retrieval augmented generation (RAG) is not enough for chatbot accuracy – it needs knowledge graphs, according to Illumex, which has revealed more about its knowledge graphs used for context-blind AI large language models.
Israeli startup Illumex wants to bring database item relationships and data in structured, block-based storage to GenAI large language model (LLM) chatbots. Up until now they have been trained using unstructured (file and object) text, audio, image and video data. However much mission-critical business data is stored in relational databases with metadata – data contents are connected and interdependent. Knowledge graphs model how pairs of items (entities) are related, and such relationships can be used to improve chatbot accuracy and relevance. Illumex’s General Semantic Fabric (GSF) aims to be a mechanism for achieving this.
Founder and CEO Inna Tokarev Sela explained in a canned quote: “Our platform is designed to bridge the gap between enterprise data complexity and the transformative power of generative AI. By providing a turnkey solution that automates the complex process of mapping data semantics and resolving terminological inconsistencies across business silos, we’re enabling organizations to achieve true AI and GenAI readiness.”
A Gartner report, “Cool Vendors in Data Management: GenAI Disrupts Traditional Technologies,” suggests data management leaders should: “Deliver customized, context-aware and more accurate GenAI results by deploying RAG, knowledge graphs and other semantic technologies to leverage your organization’s existing data. Context-aware results are especially helpful for providing concrete guidance to users who engage with data via natural language queries.”
Knowledge graphs store and model relationships between entities (events, objects, concepts, or situations), between so-called head and tail entities, with a “triple” referring to a head entity + relationship + tail entity. Such triples can be linked and the relationships are the semantics.
Two simple knowledge graphs showing duplicate block entities that are not the same. Unless a chatbot “knows” the context in which the word “block” is being used it could create an invalid response to users’ requests. B&F diagram
The Gartner report explains: “In our opinion, illumex’s platform directly addresses this need by:
Automating the creation of semantically enriched knowledge graphs
Mapping enterprise data to industry-specific ontologies
Enabling more precise and contextual natural language queries
Providing an alternative to RAG that enables transparent and governed data interactions with LLMs”
LLMs use semantic search based on finding vectors (encoded tokens describing abstracted as aspects of text, audio, image or video data) similar to a chatbot user’s input request. The trick Illumex wants to pull off is getting its separate knowledge graph semantics used as well. But you cannot simply vectorize a knowledge graph describing structured data as you can a piece of text. Instead you abstract its inherent relationships – Illumex’s GSF – and enable an LLM to process this by using a dedicated application.
Diagram from research paper discussing how LLMs and Knowledge Graphs can be used in complementary fashion.
An Illumex spokesperson told us: “Today’s LLMs do not yet use knowledge graphs directly, but unifying the two technologies is an active area of research (see here).
“Currently, LLMs are integrated with knowledge graphs in two ways:
LLMs are used to automate the creation and maintenance of knowledge graphs.
Agent-like applications enable LLMs to retrieve information from knowledge graphs and use it to augment answers (the integration is in the application layer and not directly within the LLMs).
“Illumex uses both of these approaches, integrating custom knowledge graphs with any of the commercially available LLMs.”
We understand that these agent-like applications are like connectors between Illumex-created knowledge graphs and LLMs. Without these connectors Illumex and its knowledge graphs are stranded. These knowledge graph agents will be specific, at least initially, to a particular structured dataset’s knowledge graphs and a particular LLM. They’ll need to be created by Illumex itself or an Illumex customer wanting to use the GSF to improve the output from an LLM it uses or wants to use.
On being queried about this, Illumex told us: “Illumex has actually already created the connector themselves, which they call illumex Omni. Users can access a chat interface within their organization’s Slack. When the user asks a question, Omni interprets the question and maps it to the correct data objects using the knowledge graph, generates SQL and retrieves the relevant data, and then passes the context and data to an LLM to get an answer for the user. Omni supports open source LLMs, as well as commercial LLMs like ChatGPT.”
Our understanding is that Illumex’s progress depends upon demonstrating that its GSF significantly and verifiably improves LLM responses in a RAG and vector-based semantic search environment that produces inadequate and/or wrong responses. Response accuracy and relevance should hopefully shoot up when its GSF is added to the mix.
Bootnote
The “Unifying Large Language Models and Knowledge Graphs: A Roadmap” research paper discuses how knowledge graphs can be used to improve LLMs, and LLMs provide help for knowledge graph creation. It also suggests a three-stage roadmap leading to a unification between LLMs and knowledge graphs.
Cloud storage firm Wasabi and Zero Circle, a sustainable finance marketplace, have hooked up to try to give businesses the data they need to “reduce their environmental impact.”
Wasabi has integrated Zero Circle’s invoice-based carbon footprint calculator to give customers “transparency” and a real-time assessment of their emissions.
Customers can upload their Wasabi invoice to Zero Circle’s carbon calculator to determine their estimated CO2 footprint, based on actual data stored in a Wasabi storage region. This “increases the accuracy of their ESG benchmarking” with their peers, claimed Wasabi.
Wasabi’s “hot” storage services may well generate relatively high CO2 emissions among customers, as they constantly access stored data for use by their operational applications through the cloud.
According to the Wasabi 2024 Global Cloud Storage Index, sustainability ranks among the top three most important considerations for buyers when choosing a cloud storage service. Factors like ESG commitments, energy-efficient architecture design, and native tools for carbon footprint calculation are all increasingly important to modern enterprises as they adopt cloud infrastructure services, maintains the provider.
The inference is that cloud service providers must deliver accurate, reliable tools, metrics, and programs to meet customer expectations, as well as do everything technically possible at their end to store and deliver the data in a sustainable way to their customers on-demand.
According to the hyperscaler’s latest environmental report, the leap was down to the increasing amounts of energy needed by its growing global datacenter footprint, exacerbated by the big growth of energy-sucking AI data workloads.
“Partnering with Zero Circle underscores Wasabi’s dedication to providing top-tier, sustainable cloud storage solutions,” insisted David Boland, vice president of cloud strategy at Wasabi Technologies. “Zero Circle’s carbon footprint calculation has already enabled Wasabi, our partners, and customers to gain a deeper understanding of our environmental impact.”
By integrating this information into its supply chain, Wasabi was “amplifying” the benefits for everyone, he said.
The Wasabi collaboration marked a “significant step” in advancing environmental sustainability within the cloud storage sector,” added Hemanth Setty, founder and CEO of Zero Circle. “This paves the way for innovative and groundbreaking solutions.”
The carbon footprint solution is now available to all Wasabi customers, and the pair are also “exploring” other “eco-friendly” initiatives, such as renewable energy procurement and carbon offset programs to “further reduce” the environmental impact of cloud storage.
Zero Circle is a green finance marketplace that “simplifies access” to green capital for mid-tier businesses. Its AI-powered platform streamlines sustainability assessments and reporting, and automates financial and sustainability KPIs to “incentivise” green financing, it says.
