Huawei used its annual shindig HUAWEI CONNECT in Shanghai last week to show off its latest all-flash storage array, the OceanStor Dorado V6. And it’s a monster – with the biggest capacity that Blocks and Files has seen in a storage array.
The company has not announced availability but prior to launch it has pumped out some big numbers via a press release and a marketing page.
The Dorado V6 performs up to 20 million I/O operations per second (IOPS) – twice as much as the next-best player according to Huawei, which did not name the rival. We think it is referring to Dell EMC’s PowerMax 8000 which delivers up to 10 million IOPS. Read latency for the Huawei system is down to 0.1 ms.
Speeds and feeds
The five Dorado V6 models support Huawei’s Hi1812E NVMe SSDs and NVMe-oF access. They scale by IO port counts and the maximum number of SSDs supported, as the table below shows.
Basic maths says the maximum raw capacities are:
3000 – 36.8 PB
5000 – 49.2 PB
6000 – 73.73 PB
8000 – 98.3 PB
18000 – 196.61 PB
As it is 2019, the Dorado V6 of course has an AI processor – a first for a storage array, Huawei claims. The system uses this for performance tuning and management. The Dorado V6 is also packed with fault-tolerance features and Huawei claims a one-second switchover with uninterrupted links in the event of controller failure.
The array features inline deduplication and compression, thin provisioning, remote replication, continuous data protection, quality of service, cloning and snapshots and cloud backup.
Huawei has not revealed pricing yet but the V6 will probably cost $2.5m or more, judging by SPC-1 benchmark results for Dorado V3 and V5 arrays. You can apply for pricing on Huawei’s website by providing project details and a budget range into a general pricing inquiry pop-up.
Samsung has started shipping smart PM1733 and 1735 SSDs with up to 30-plus TB capacity, gen 4 PCIe for faster data transfer and features that enable them to operate for longer and with more users.
Speeds and feeds first and then we’ll look at the extra features.
Both SSDs use NVMe over PCIe gen 4. They come in U.2 (2,5-inch) and half-height, half-length Add-in-Card formats. The capacity ranges are;
PM1733 U.2 – 0.96TB – 30.72TB
PM1733 AIC – 1.92TB – 15.36TB
PM1735 U.2 – 0.8TB – 12.8TB
PM1735 AIC – 1.6TB – 12.8TB
The PM1733 is optimised for read performance with endurance of a single drive write per day (DWPD) for fiv years. The PM1735 is optimised for mixed read/write use and endurance is three DWPDs for five years.
Samsung PM1733 exploded view
Bandwidth is up to 3.8GB/sec sequential write for both SSDs in both formats. Bandwidth up to 8GB/sec sequential write in AIC format and 6.4GB/sec in U.2 format. These speeds are at least twice as fast as current PCIe gen 3 drives.
Random read IOPS are up to 260,000 for the PM1735 U.2 with an amazing 1.45m random read IOPS. Samsung has not supplied IOPS numbers for the PM1735 AIC or the PM1733 in U,2 or AIC formats.
Feature fun
The performance numbers are great but the standouts are three extra controller features.
Fail-in-place. If a fault is identified in any NAND chip inside these drives – there are 512 inside the 30.72TB model – FIP software activates error-handling algorithms automatically to effectively bypass the chip. This is conceptually similar to disk bad block handling and enables the SSD to keep working if NAND chips inside it fail.
Virtual SSDs. The SSD is presented as up to 64 virtual and smaller SSDs, providing independent, virtual workspaces for multiple users. Server CPU tasks such as Single-Root I/O Virtualization (SR-IOV) can be done by the SSD controller, offloading the server host.
V-NAND machine learning technology. This detects any variation among circuit patterns through big data analytics. It looks at and verifies cell characteristics and predicts how cell behaviour will develop. As we understand it, the SSD controller monitors telemetry from the chips and runs it through machine learning models to track cell performance.
Samsung said this means its SSDs have higher levels of performance, capacity and reliability because they can sweat their cell assets better. As SSD chips move from TLC to QLC with more voltage levels in the cells, the precision management of cell characteristics becomes more important.
Western Digital’s zoning concept is another example of extra drive features. Blocks & Files expects this approach to spread quickly to other enterprise SSD suppliers.We are heading towards an era of PCIe 4 speed-accelerated and much smarter SSDs.
Western Digital is abandoning the storage systems business and is selling the IntelliFlash array unit to DDN. It has also put its ActiveScale archival storage array business up for sale.
This is an abrupt and unexpected about turn for Western Digital which acquired IntelliFlash when it bought Tegile in August 2017 for an undisclosed sum. As recently as July this year WD extended IntelliFlash capabilities with entry-level NVMe models, a higher-capacity SAS array, live dataset migration and an S3 connector.
The sale to DDN suggests to Blocks & Files that the Tegile acquisition was a WD mistake. This excursion into enterprise storage arrays and archive systems reflects poorly on Western Digital leadership.
The conclusion many will draw is that selling data centre storage systems to the highly competitive enterprise market was a step too far for WD. This business gets more than 80 per cent of its revenues from selling disk drives and SSDs to OEMs and consumers, where it faces limited competition.
A WD statement described: “Western Digital’s strategic intention to exit Storage Systems, which consists of the IntelliFlash and ActiveScale businesses. The company is exploring strategic options for ActiveScale. These actions will allow Western Digital to optimize its Data Center Systems portfolio around its core Storage Platforms business, which includes the OpenFlex platform and fabric-attached storage technologies.”
Mike Cordano, WD COO, said in a prepared statement; “Scaling and accelerating growth opportunities for IntelliFlash and ActiveScale will require additional management focus and investment to ensure long-term success.”
Alex Bouzari, CEO and co-founder of DDN, said in a canned quote: “We are delighted to add Western Digital’s high-performance enterprise hybrid, all flash and NVMe solutions to DDN’s… data management at scale product portfolio.”
IntelliFlash inside DDN
The joint DDN-WD announcement said IntelliFlash customer will benefit from DDN’s focus on storage and data management challenges, deep expertise in service and support and a rich, broad technology portfolio. DDN has a set of capabilities that WD lacks as well as a willingness to invest.
IntelliFlash staff will join DDN, which now has more than 10,000 customers and 500 partners worldwide. WD and DDN will work to deliver a seamless transition for customers and partners with ongoing product availability and support continuity. DDN is to invest in an accelerated roadmap of the IntelliFlash line.
The deal includes a mutual global sourcing agreement in which Western Digital will become a customer of IntelliFlash from DDN and a preferred HDD and SSD supplier to DDN.
DDN bought the crashed Tintri business for $60m in September 2018 and Nexenta for an undisclosed sum in May this year. DDN now has three newly-acquired product lines to integrate and locate in its product space and marketing messages; Tintri, Nexenta and Tegile. This is already starting to happen with Nexenta file capabilities added last month to Tintri systems.
Broadly speaking, Tintri and IntelliFlash compete. IntelliFlash is an enterprise array as is Tintri. It has hybrid and all-flash models but lacks Tintri’s capabilities in virtualized server integration. This software capability could be grafted onto the IntelliFlash OS and we might also expect Tintri and IntelliFlash to evolve towards a common hardware chassis.
The two lines should probably merge – unless DDN can convincingly differentiate them.
And let’s not forget ActiveScale
ActiveScale is an archival storage array, and WD got into the archive vault business when it bought HGST. As well as making disk drives HGST sold the ActiveArchive archive system. The basis for this was HGST’s acquisition of Amplidata in 2015.
ActiveScale arrived in late 2016 and it became the lead archival product. Now it is an unwanted product. It is not an acquisition target for DDN which has its own WOS object storage line and Nexenta object storage software. It does not need a third object storage technology.
The DDN – WD transaction is expected to close later this year subject to closing conditions. WD’s storage systems business exit is expected to generate an annual non-GAAP EPS benefit for WD of at least $0.20, starting in the fiscal 2020 third quarter ending April 3, 2020. It will incur as yet unquantified restructuring and other charges.
Liqid’s composable systems can now compose up to 16TB of Liqid Memory, enabling larger applications to reduce IO and execute much faster.
A composable system dynamically sets up a virtual server to run application workloads by pulling compute, memory, FPGA, storage and networking capabilities from a resource pool. When the server’s job is finished component resources are returned to the pool for re-use.
Liqid Memory is a DRAM and multiple NVMe SSD combination turned into virtual memory by ScaleMP’s vSMP MemoryONE software.
This produces software-defined memory that can transparently replace or expand DRAM for memory-intensive applications,, according to ScaleMP. The company has written its own memory management unit to map application and system DRAM accesses to the virtual memory space which includes capacity from the NVME SSDs. The technology works without operating system or application modification.
The upshot is that more of an application’s working set executes in virtual memory and runs faster than when operating in a smaller pure DRAM memory pool with data fetched from SSDs or disk. Prefetch algorithms help achieve near-DRAM performance.
Liqid said Liqid Memory is suited for in-memory databases, including Oracle TimesTen and In-Memory Column Store, SAP HANA, DB2 BLU, Apache Spark, Aerospike DBS, and other memory-intensive applications.
Liqid Memory is available as a PCIe Add-In Card (Element LQD3900), U.2 (Element LQD3925x) drive, or within a Dell appliance with two Xeon Scalable processors and up to 12TB system memory.
Element LQD3900 PCIe add-in-card.
The LQD3900 uses an Optane SSD – as detailed in the spec sheet.
High-end, multi-socket servers can support 12TB or so of memory but low-end and mid-range ones do not. Liqid Memory is a way of giving them access to a larger pool of memory.
Pure Storage has teamed up with Cloudian, Komprise and Veritas to offer customers ways of protecting FlashArray data and migrating data to/from FlashBlade.
FlashArray is Pure’s block access flash array for primary and structured data. FlashBlade is its file/object array for unstructured data.
Cloudian
Cloudian has integrated its HyperStore object storage system with the CloudSnap feature in FlashArray’s Purity OS. CloudSnap takes a FlashArray snapshot and sends it to AWS S3-compliant storage; HyperStore in this case.
CloudSnap connects with HyperStore as a backup, archive or DR target via an S3 API, providing policy-based, data transfer, including moving portable snapshots to and from HyperStore.
From there Cloudian can send it on to the public cloud for longer term and cheaper retention.
A FlashArray would be linked to Cloudian for backing up snapshot data where the fast restore feature from Pure’s FlashBlade file/object storage flash array is not needed.
Komprise
Komprise’s file information lifecycle management software can be used to ingest data into FlashBlade. The software is included in Pure’s FastStart, a customer on-boarding program for FlashBlade. This is available for a set period after which FlashBlade customers could become direct Komprise customers and also continue using the product.
Krishna Subramanian, Komprise co-founder and COO, told Blocks & Files this week at the Pure Accelerate event in Austin TX, that this is up to twice as fast as other ways of achieving FlashBlade data ingest such as RoboCopy. That’s because Komprise has optimised its parallel IO software to take advantage of FlashBlade’s parallel IO.
Komprise supports NFS and CIFS/SMB file sources with S3 object source support coming, possibly in 2020 and in time for Amazon’s ReInvent shindig.
The Komprise software could also tier data off FlashBlade to cheap-and-deep backend targets for long-term retention.
Veritas
There is an existing though limited partnership between Veritas and Pure. Veritas’s NetBackup supports snapshot-based protection for FlashArray and FlashBlade in AI Data Hub form. FlashBlade can also be a storage target for Veritas software.
Veritas has added three more integrations:
InfoScale for mission-critical levels of performance and 24-hour availability
APTARE IT Analytics to optimize capacity, utilisation, performance and compliance with data regulations
The Veritas Access Appliance can be a long-term retention storage target for FlashArray data snapshots
Altogether these integrations means the entire Veritas Enterprise Data Services platform links to FlashArray and FlashBlade.
Gartner’s first combined hybrid and all flash primary array magic quadrant gives higher then expected rankings to NetApp, Pure and Infinidat, including them alongside traditional leaders Dell EMC, HPE and IBM.
The Blocks & Files standard MQ explainer says the magic quadrant is defined by axes labelled ‘ability to execute’ and ‘completeness of vision’, and split into four squares tagged ‘visionaries’, ‘niche players’, ‘challengers’ and ‘leaders’.
Here’s the quick look primary array diagram every MQ fan wants to see:
And it’s a doozy. We have added a diagonal green line indicating balanced progress to the top right high-point for ability to execute and completeness of vision. We’ve also indicated two separate groups in the leaders’ box.
Group 1 includes top-ranked NetApp, then Dell EMC, with Pure unexpectedly in this group and in, we think, third place, with a strong vision component alongside its ability to execute. HPE is in fourth place.
A sweaker and more closely positioned group of leaders consists of Hitachi Vantara and IBM, then Huawei and, surprise, surprise, startup Infinidat entering alongside publicly-owned and more mature competitors.
The challenger’s box includes Western Digital – another surprise (thank you. Tegile); Fujitsu; DDN – strengthened through the Tintri acquisition; and Lenovo, another unexpected entrant boosted by an OEM deal with NetApp.
The niche players are China’s Inspur. Oracle, NEC, Infortrend and Synology. Playing a solo role in the visionaries’ box is Kaminario, quite close to the leaders’ quadrant. That’s 18 suppliers in total.
Get a copy of the report from Gartner, at a price, or, hopefully, from a vendor licensed to distribute it for no charge.
Jeff Clarke, Dell Technologies’s vice chairman and overall product and operations boss, has stepped away from day-to-day control of the server-to-storage-to-networking-to-protection ISG Group. He passes the baton to EMC veteran Jeff Boudreau, GM for storage inside ISG, who will report to him.
Dan Inbar, SVP in charge of Israel R&D, becomes head of storage and reports to Boudreau. John Roese, CTO in charge of cross-product operation, becomes CTO for all ISG products. Clarke retains his titles and overall responsibilities following the management shuffle.
Clarke spent two years wrestling the conglomeration of Dell servers, switches and storage products and the acquired EMC businesses into a single operation.
He swept away EMC’s notion of a multi-tiered product range with partially overlapping and competing products. And in its place he imposed the Dell way of operating lead brands with co-ordinated product ranges and branding across ISG (Infrastructure Solutions Group.)
The management moves constitute the completion of Dell’s integration with EMC storage operations. The public outcomes so far are PowerEdge servers, and PowerMax, PowerVault, and PowerProtect storage systems along with PowerSwitch network products.
Blocks & Files understands that Dell will have announced the full Power portfolio by or before Dell Technologies World in Spring 2020. This will include the as yet unannounced MidRang .Next storage line which will combine the SC, Unity and Xtremio storage array products.
We might expect the Isilon scale-out filer business to adopt PowerFile or similar branding. Sort out Data Domain branding to Power-something and then it would be a clean Power sweep.
Hitachi is merging the Hitachi Vantara and Hitachi Consulting businesses under the Hitachi Vantara brand, with effect from January 2020.
Hitachi Vantara supplies VSP storage array, servers and business analytics software based around its 2015 Pentaho acquisition and others.
The combined business will “strengthen front line and delivery capabilities to increase alignment and unlock the synergies between Hitachi Vantara and its vertical business units.” The company will also focus on building Hitachi’s Lumada Internet of Things business.
The new entity is to be led by chairman and CEO Toshiaki Tokunaga, who chairs Hitachi Global Digital Holdings, the holding company that oversees Hitachi Vantara and Hitachi Consulting.
Brian Householder, current Hitachi Vantara CEO, and Hicham Abdessamad, CEO of Hitachi Consulting, will remain at Hitachi in executive leadership positions.
This looks like the Japanese parent corporation returning two American-led subsidiaries to Japanese control. Tokunaga is board chairman for Hitachi Global Digital Holdings and it is unlikely he will remain as the CEO exerting day-to-day control of the new Hitachi Vantara. It is also possible that either Brian Householder or Hicham Abdessamad could be the new CEO. Either would signify continuity for north America and EMEA customers. A Japanese CEO could signal a break from that continuity.
Leadership and organisational structure details will be announced in January 2020.
Goldman Sachs has led a $147m funding round in Acronis which values the data recovery and protection vendor at more than one billion dollars and gives it unicorn status.
Acronis will use the money for acquisitions, engineering hires, data centre expansion and to increase public sector revenues in North America.
Founded in 2003, Acronis is headquartered in Singapore and Switzerland and has more than 1,400 employees. The company claims more than 500,000 business customers, five million consumer customers, 50,000 channel and service provider partners in 150 countries.
Serguei Beloussov, Acronis co-founder, chairman and CEO, said in a statement announcing the new funding that the company grew 20 per cent in 2018, and this year is gunning for 30 per cent growth. He expects Acronis Cyber Cloud business to grow more than 100 per cent this year.
Backup bonanza
The backup world is embracing cyber protection and cloud-based services. As customers adopt the public cloud there is an opportunity for new backup providers to jump in and offer cloud-based services complementing existing on-premises backup arrangements. It’s a gold rush or in Acronis’s case, a Goldman Sachs rush.
Acronis’s competitor Veeam took in $500m in January this year. Rubrik and Cohesity, are funded to the tune of $552m and $410m respectively. Druva had a $130m round this year taking total funding to $328m and also gaining unicorn status. And new startup Clumio entered the cloud backup-as-a-service space with $51m in funding in August.
Intel Optane DC persistent memory represents a new class of memory and storage technology designed specifically for data center usage. (Credit: Intel Corporation)
Oracle today announced its Exadata X8M storage servers will use Intel Optane DC Persistent Memory – 3D XPoint implemented in DIMM form.
Exadata systems are integrated hardware and software systems for running the Oracle database.
At Oracle OpenWorld yesterday the company said customers will achieve vastly superior performance for latency-sensitive activities such as high-frequency stock trading, IoT data integration, and applications that require real-time human interactions.
Navin Shenoy, GM of Intel’s data center group, said in a canned quote: “By enabling faster analytics and enhanced response times our customers are experiencing what’s possible with Optane DC persistent memory.”
This is certainly good news for Intel, with Optane support spreading across the storage industry. Pure Storage added it today and Dell EMC announced PowerMax support for Optane last week.
Exadata X8M
The X8M is designed to support Online Transaction Processing (OLTP), analytics, mixed workload database requirements, database consolidation and in-database machine learning.
Exadata X8M uses Xeon SP CPUs, Optane DIMMs and RoCE (Remote Direct Memory Access across Converged Ethernet) over 100GbitE. RoCE enables Oracle’s database to directly access persistent memory, thus bypassing the OS, network, and IO software stack. This reduces IO latency in Exadata X8M tenfold compared to the previous Exadata release. It also increases I/O throughput up to 60 per cent.
The X8M’s immediate predecessor, the Exadata X7, used Skylake CPUs and 25GbitE and had a 250 microsec latency rating. Therefore we presume X8M has 25 microsec latency.
Pure Storage has lined up a series of launches for day one of its Accelerate customer event in Austin, Texas that see the company’s hardware go faster, go larger, go nearline, go to subscriptions and get smarter VM analytics.
It has juiced up FlashArray//X with Optane, doubled FlashBlade capacity, announced a FlashArray//C for nearline storage use and is to make every product available as a service.
DirectMemory Modules
As we wrote a few days ago the FlashArray//X70 and //X90 models get read caching with Direct Memory Modules that carry 750GB of Optane storage-class memory. This is a non-disruptive upgrade for customers with workloads qualified by Pure.
Users could get as much as 50 per cent cut in latency depending on their workload read IO profile but there is no hard and fast rule. Pure said 80 per cent of its installed //70-90 arrays should see 20 per cent lower latency, with 40 per cent seeing 30 to 50 per cent latency reduction.
FlashArray//C
The FlashArray//C provides a slower and capacity-optimised tier of storage than the latency-optimised //X and uses DirectFlash modules fitted with QLC (4bits/cell) flash chips.
FlashArray//C
There are three //C configurations:
366TB raw – 1.3PB effective with 4:1 data reduction
878TB raw – 3.2PB effective
1.4PB raw – 5.2PB effective (FlashArray//C 60)
FlashArray//C 60. with 20 modules in a row in the expansion shelves
The module capacity is 18.3TB and there are 20 modules in the base configuration. Averaged end-to-end, read and write latency is 2 to 4 msecs – faster than disk arrays. Pure said //C systems could replace many racks of nearline disk arrays and open up new use cases for all-flash arrays.
Pure is not revealing the specific endurance of these QLC modules. It has written flash management software to handle QLC flash read retention and read disturb problems. This has reduced the write burden enough for the company to offer the same Evergreen device fail/replacement guarantees as for the FlashBlade and FlashArray//X.
Alex McMullan, international CTO, said in a interview, that Pure will ship //C arrays with TLC (3bits/cell) chips until QLC chip availability is mainstream. The array software and module firmware can cope with this and, as a byproduct, it extends the TLC module endurance. He said customer reaction to //C arrays were so positive that Pure has brought forward the launch, originally slated for next year.
A source close to Pure said QLC will be available next year, in 24TB modules. That will increase capacity presumably.
The //C system is pitched as a storage tier for test and dev work using production data from //X, to hold snaps from //X and as a disaster recovery resource for //X. Pure does not expect it to cannibalise low-end //X sales.
Unifying file and block
Pure’s //X and //C arrays offer block access. The company bought scale-out NAS software supplier Compuverde in April this year and is integrating its technology into the Purity software that runs the //X and //C arrays.
The company thinks large customers want separate file and block arrays whereas mid-range and smaller customers want a unified file+block system. A Compuverde-integrated system is in the works and should be ready in 2020.
On the file/object front Pure is doubling the capacity of FlashBlade arrays, from 75 blades to 150, giving 8PB effective capacity. The AIRI system, which uses FlashBlade as a component, is being extended into an A’I Data Hub’. This will include Kubernetes and Amazon S3 support.
Pure-as-a-Service
Pure Chairman and CEO Charles Giancarlo
Pure CEO Charles Giancarlo said in a press briefing that Pure’s ES2 Evergreen storage-as-a-service branding was confusing. The company has renamed it Pure-as-a-Service and will make all Pure products available on subscription.
This parallels what HPE is doing with its GreenLake service. Max Kixmoeller, Pure’s VP for strategy, said US accounting rule changes meant leases now had to be treated as CAPEX. The PaaS offering is OPEX.
A customer’s capacity licence can be transferred between on-premises Pure arrays and a Pure array running on AWS. But, unexpectedly, cloud capacity costs more, by an unspecified amount.
Cloud matters
CloudSnap, service with //X snapshots stored in AWS, is now available for Azure. Cloud Block Store, which runs on Amazon, is now generally available. It will also be ported to Azure. Pure will offer lower cost options, for instance, one with lower availability.
These moves position Pure as a multi-cloud service supplier. Over time we can expect the company to support the Google cloud.
Pure1VM Analytics, introduced earlier this year, has a Tier 1 VM Analytics Pro option offering smarter analytics. Pure believes that a transition is underway in its customer base from manual array management to API-based management, and AI tools are required to automate storage array management.
Availability
FlashArray//C, the Direct Memory Modules, Cloud Block Store for AWS and CloudSnap for Azure are all generally available. The expanded FlashBlade is on limited or directed availability as is Pure1 VM Analytics.
Cloudian is building ruggedised object storage devices for the IoT edge market in Japan via its EDGEMATRIX subsidiary.
The idea edge device sensors such as video cameras generate enormous data. This can be used in applications such as city traffic and pedestrian flow management, environmental and operations control in smart buildings and factories, and customising roadside advertising displays for oncoming vehicles.
Decisions about incoming data content are complex and need to be made in realtime, necessitating artificial intelligence-type software models and local data analytics processing by hardware devices running application and storage software. GPUs will enable machine learning models to be processed.
Cloudian has set up EDGEMATRIX to build and sell deployment-ready edge devices for these applications.
The system software can send subsets of data analysed at the edge to a central storage repository using the S3 API. Customers can manage this S3-accessible data across on-premises and public cloud environments using Cloudian facilities.
EDGEMATRIX indoor (left) snd outdoor (right) devices. Not to scale.
Speeds and feeds
There are indoor and outdoor models with the outdoor product able to withstand 30 minutes under water. The basic specs are;
Networking: LTE, Wifi, Gbit Ethernet x 2, HDMI, USB, Power over Ethernet (for cameras, etc.)
The systems will come in Light (Jetson Nano), Standard (Jetson TX-2) and Advanced (Intel Core i7, Tesla T-4 GPU) configurations.
The standard and advanced boxes are available in the fourth quarter. The light box should be released by the end of the year.
Business set-up
EDGEMATRIX is majority-owned by Cloudian and run by co-founder Hiroshi Ohta.
The company has raised $9m from a group of strategic investors that include NTT DOCOMO. SHIMIZU Corp. and Japan Post Capital, as well as Cloudian CEO and co-founder Michael Tso and board director Jonathan Epstein. SHIMIZU constructs smart buildings and NTT DOCOMO is a mobile telco with developing 5G Interests.
In an announcement Ohta said: “In just a short time, we have built a compelling product, an initial base of top-tier customers and a range of strategic partnerships.”
Cloudian CEO Michael Tso contributed this quote: “By establishing the EDGEMATRIX subsidiary, we can align more closely with our strategic partners in AI.”
