Print Friendly, PDF & Email

The 3rd AI Hardware Summit took place virtually earlier this month and it was exciting to see how quickly the ecosystem has evolved and to learn of the challenges the industry has to solve in scaling artificial intelligence (AI) infrastructure. I would like to share highlights of the Summit, along with other notable observations from the industry in the area of accelerated computing.

The proliferation of AI has emerged as a disruptive force, enhancing applications such as image and speech recognition, security, real-time text translation, autonomous driving, and predictive analytics. AI is driving the need for specialized solutions at the chip and system level in the form of accelerated compute servers optimized for training and inference workloads at the data center and the edge.

The Tier 1 Cloud service providers in the US and China lead the way in the deployment of these accelerated compute servers. While the deployment of these accelerated compute servers still occupies a fraction of the Cloud service providers’ overall server footprint, this market is projected to grow at a double-digit compound annual growth rate over the next five years. Most accelerated server platforms shipped today are based on GPUs and FPGAs from Intel, Nvidia, and Xilinx, the number of new entrants, especially for the edge AI market, is growing.

However, these Cloud service providers, or enterprises deploying AI applications, simply cannot increase the number of these accelerated compute servers without addressing bottlenecks at the system and data center level. I have identified some notable technology developments that need to be addressed to advance the proliferation of AI:

    • Rack Architecture: We have observed a trend of these accelerated processors shifting from a distributed model (i.e., one GPU in each server), to a centralized model consisting of an accelerated compute server with multiple GPUs or accelerated processors. These accelerated compute servers have demanding thermal dissipation requirements, oftentimes requiring unique solutions in form-factor, power distribution, and cooling. Some of these systems are liquid-cooled at the chip level, as we have seen with the Google TPU, while more innovative solutions such as liquid immersion cooling of entire systems are being explored. As these accelerated compute servers are becoming more centralized, resources are pooled and shared among many users through virtualization. NVIDIA’s recently launched A100 Ampere takes virtualizing to the next step with the ability to allow up to seven GPU instances with a single A100 GPU.
    • CPU: The GPU and other accelerated processors are complementary and are not intended to replace the CPU for AI applications. The CPU can be viewed as the taskmaster of the entire system, managing a wide range of general-purpose computing tasks, with the GPU and other accelerated processors performing a narrower range of more specialized tasks. The number of CPUs also needs to be balanced with the number of GPUs in the system; adequate CPU cycles are needed to run the AI application, while sufficient GPU cores are needed to parallel process large training models. Successive CPU platform refreshes, either from Intel or AMD, are better optimized with processing inference frameworks and libraries, and support higher I/O bandwidth within and out of the server.
    • Memory: My favorite session from the AI Hardware Summit was the panel discussion on memory and interconnects. During that session, experts from Google, Marvell, and Rambus shared their views on how memory performance can limit the scaling of large AI training models. Apparently, the abundance of data that needs to be processed in memory for large training models on these accelerated compute servers is demanding greater amounts of memory. More memory capacity means more modules and interfaces, which ultimately degrades chip-to-chip latencies. One proposed solution that was put forth is the use of 3D stacking to package chips closer together. High Bandwidth Memory (HBM) also helps to minimize the trade-off between memory bandwidth and capacity, but at a premium cost. Ultimately, the panel agreed that there needs to be an optimal balance between memory bandwidth and capacity within the system, while adequately addressing thermal dissipation challenges.
    • Network Connectivity: As these accelerated compute nodes become more centralized, a high-speed fabric is needed to ensure the flow of huge amounts of unstructured AI data over the network to accelerated compute servers for in-memory processing and training. These connections can be server-to-server as part of a large cluster, using NVIDIA’s NVlink and InfiniBand (which NVIDIA acquired with Mellanox). Ethernet, now available up to 400 Gbps, is an ideal choice for connecting storage and compute nodes within the network fabric. I believe that these accelerated compute servers will be the most bandwidth-hungry nodes within the data center, and will drive the implementation of next-generation Ethernet. Innovations, such as Smart NICs, could also be used to minimize packet loss, optimize network traffic for AI workloads, and enable the scaling of storage devices within the network using NVMe over Fabrics

I anticipate that specialized solutions in the form of accelerated computing servers will scale with the increasing demands of AI, and will comprise a growing portion of the data center capital expenditures. Data centers could benefit from the deployment of accelerated computing, and would be able to process AI workloads more efficiently with fewer, but more powerful and denser accelerated servers. For more insights and information on technology drivers shaping the server and data center infrastructure market, take a look at our Data Center Capex report.

Print Friendly, PDF & Email

 

Going into 2020 we outlined ten 5G predictions. Since we now have data for 1H20 and will soon start collecting data for 3Q20, the timing is right for a quick status check on the 5G predictions we outlined going into 2020.

1)  5G RAN+Core Infrastructure Market to More than Double

Prediction: 5G NR continues to accelerate at an extraordinary pace, much faster than expected four or five years ago or even just six or three months ago, underpinned by large-scale deployments in China, Korea, and the US. These trends are expected to extend into 2020. The upside in 5G NR will be more than enough to offset declining LTE investments, propelling the overall RAN (2G-5G) market for a third consecutive year of healthy growth.

Status: This is for the most part playing out as we outlined, though we have adjusted the 5G NR and Core outlook upward to reflect larger than expected deployments in particularly China, which is more than enough to offset LTE capex on pace to decline more than 20% in 2020.

 

 

2) Early Adopters to Embrace 5G SA

Prediction: The path toward 5G has become more straightforward since the 2Q19 quarter with only two options now—Option 3 which is 5G NSA, utilizing the EPC, and Option 2 which is 5G SA utilizing the 5G Core. As we move into 2020, we will see the emergence of the first 5G Standalone (5G SA) networks. We expect service providers in China, Korea, the Middle East, and the US to launch 5G SA sometime in 2020.

Status: 5G SA and 5G core is now a reality. According to the 2Q20 MCN report, 5G Core revenues accounted for nearly 15% of overall Mobile Core Network Revenues, underpinned by strong developments in the APAC region. “The ramp-up in the second half of 2020 has already begun, with T-Mobile commercially launching its 5G Standalone network in August, recognizing Cisco and Nokia, as their 5GC suppliers,” said Dave Bolan, Dell’Oro 5G Core, and MEC Expert.

 

3) More than 100 M Transceivers

Prediction: The Massive MIMO business case has changed rather significantly over the past two to three years with the technology now considered to be a foundational building block for mid-band NR deployments. We recently revised the 2020 Massive MIMO outlook upward, driven by surging year-to-date shipments and improved market sentiment for 2020. The overall 5G NR transceiver installed base – Massive MIMO plus Non-Massive MIMO for sub 6 GHz and Millimeter (mmW) macros and small cells – is projected to eclipse 0.1 B by 2020.

Status: Preliminary estimates suggest combined macro and small cell transceiver shipments for non-Massive MIMO and Massive MIMO configured systems was already in the 50 M to 100 M range for the 1H20 period. The forecast remains on track with total transceiver shipments projected to surpass 0.1 B in 2020.

 

4) Dynamic Spectrum Sharing Takes Off

Prediction: The attitude towards spectrum sharing is on the upswing, with both suppliers and operators discussing their spectrum sharing roadmaps. In addition to the spectral efficiency gains of 15% to 20%, operators are considering the benefits from a marketing perspective. Operators also see the extended 5G NR coverage with a lower band spectrum as a key enabler for 5G SA and network slicing. The technology is expected to play a pivotal role in upgrading existing low-band LTE sites to NR in the year 2020.

Status: Operators are clearly warming up to the idea that the upside with DSS outweighs the downside, with the key benefits including 1) Accelerate nationwide coverage, 2) Spectral efficiency upside comparing NR and LTE, 3) Overall performance upside using CA, and 4) Simplify and accelerate the transition towards 5G SA.

The downside with DSS is that it can impact the net capacity. But at the same time, operators also know that there are only three paths to move from 4G to 5G with the existing LTE spectrum — re-farming, static sharing, and DSS. And unless the goal is to stay on LTE, the reality is that the picture becomes more favorable when comparing DSS with static sharing. According to Ericsson, the relative performance upside with both the LTE and 5G NR traffic when comparing DSS with static sharing at various subscriber penetration rates is material (for a 50% NR penetration, Ericsson estimates the LTE and NR upside could be around 86% and 57%, respectively).

And more importantly, operators are shifting from talking about DSS to deploying the technology. Swisscom and Verizon have already deployed DSS nearly nationwide. AT&T is also using DSS to expand its low-band coverage. Deutsche Telekom said with its 2Q20 update that its low-band 5G network will cover two-thirds of the German population by the end of 2020, relying heavily on DSS.

Ericsson said earlier in the year that 80% of customers that are testing DSS have plans to deploy the technology over the next year.

 

5) 5G NR Indoor Small Cell Market to Surpass LTE

Prediction: With more data points suggesting the beamforming gains with Massive MIMO radios delivering comparable outdoor coverage in the C-band relative to 2 GHz LTE deployments, preliminary data from the field also suggests indoor performance will be a challenge and operators are already migrating the indoor capex from 4G to 5G. These trends are expected to intensify in 2020.

Status: We have not made any material changes to the overall outlook and still expect full-year 2020 5G NR indoor revenues to surpass LTE pico investments. While shipments were impacted negatively sequentially between 4Q19 and 1Q20, partly due to COVID-19, market conditions improved significantly during 2Q20 propelling the overall 1H20 indoor 5G NR revenues to advance more than 10x year-over-year.

 

6) Millimeter Wave (mmW) to Approach 10% of 5G NR Small Cell Installed Base

Prediction: Even though deploying 5G NR in the mid-band using the existing macro grid will deliver the best ROI for some time for operators seeking to optimize cost per GB and average speeds, 5G NR mmW shipments and revenues increased substantially in the third quarter of 2019, with the overall mmW NR market trending ahead of expectations. We recently adjusted our near-term mmW outlook upward to take into consideration the state of the market and improved visibility about the underlying fundamentals in the US, Korea, and Japan.

Status: As we discussed in the recently published 5-year RAN Forecast, the outdoor mmW market has surprised on the upside while the indoor mmW market has taken a bit longer to ramp than we initially expected. Even with the recent upward adjustment for the outdoor segment and downward revisions for the indoor mmW forecast, the total mmW market remains on track to surpass 0.1 M units and is expected to account for 5% to 10% of the overall 5G NR small cell installed base by the end of 2020.

 

7) 5G MBB to Account for More than 99% of the 5G NR Market

Prediction: We remain optimistic about the IoT upside for Industrial IoT/Industry 4.0, reflecting a confluence of factors including 1) Suppliers are reporting healthy traction with the vertical segments, 2) More countries are exploring how to allocate spectrum for verticals, 3) Ecosystem of industrial devices is proliferating rapidly, and 4) New use cases that require cellular QoS are starting to emerge. At the same time, the LTE platform is expected to suffice for the majority of the near-term vertical requirements implying it is unlikely 5G NR IoT-related capex will move above the noise in 2020.

Status: MBB continues to dominate the 5G capex. LTE IoT is picking up the pace and remains on track to comprise a low-single-digit share of the 2020 LTE RAN market. 5G IoT is moving in the right direction with interesting use cases starting to emerge. But it will take some time before the 5G IoT NR capex will move above the noise. At the same time, we adjusted the overall FWA outlook upward with the recently published 5-Year RAN Forecast, reflecting an improved business case and shifting usage patterns triggered by COVID-19. With this adjustment, we currently estimate 5G MBB remains on track to account for more than 97% of the 5G NR market.

 

8) Virtual RAN 5G NR Revenues to Exceed Open RAN 5G NR Revenues

Prediction: There are multiple ongoing efforts driven both by operators and suppliers with the primary objective of realizing a more flexible architecture that will optimize TCO for both the known and unknown use cases while at the same time improving the ability for the service providers to differentiate their services. Given the current state of these tracks with the incumbents investing more in virtual solutions and the readiness of Open RAN initiatives for existing 5G MBB deployments, we envision Non-Open RAN Virtual 5G NR revenues will be greater than Open RAN (virtual RAN with open interfaces) 5G NR revenues in 2020.

Status: With Open RAN now developing at a faster pace than initially expected, we might need to revisit this assessment.

 

 

9) 5G NR RAN Revenue HHI to Increase > 100 Points

Prediction: Total RAN HHI has been fairly stable over the past three years, reflecting a competitive dynamic that remains fierce, moderately concentrated, and relatively stable. Initial readings suggest the 5G NR HHI for the 4Q18-3Q19 period is trending below the 2018 overall RAN HHI, however, we expect the 5G NR HHI to increase in 2020.

Status: Preliminary estimates suggest the 1H20 5G NR HHI was about 5% to 10% greater than the overall RAN HHI, underpinning projections that large scale 5G NR deployments in China are impacting the 5G RAN landscape.

 

10) 5G NR Subscriptions to Approach 0.2 B

Prediction: Preliminary estimates suggest the shift from LTE to NR is roughly two to three years faster than the 3G to 4G migration from a RAN infrastructure and subscription adoption perspective. The end-user ecosystem is developing at a rapid pace with multiple chipsets, devices, and phones supporting both NSA and SA for the low, mid-, and mmW spectrum now commercially available. While TDD has dominated mid-band and mmW deployments to date, FDD based 5G NR phones became a reality in 2H19 and will proliferate in 2020. End-user device adoption is projected to accelerate rapidly in 2020, with 5G NR approaching 0.2 B subscriptions, bolstered by healthy NR subscriber adoption in China, Korea, and the US.

Status: According to the GSA, 5G subscriptions approached 138 M in the second quarter. It is possible our 2020 forecast is low with total 5G subscriptions not only approaching but even surpassing 0.2 B this year.

 

For more information about our 5G RAN, Mobile Core Network, Open RAN, and MEC programs, please contact us at dgsales@delloro.com.

Print Friendly, PDF & Email

 

At this year’s virtual Cable-Tec Expo, four prominent themes have emerged throughout the online panels and technical presentations:

    1. Cable broadband networks have performed incredibly well during the COVID-19 pandemic, with minimal outages and minimal complaints from customers.
    2. Despite the reliability, there is a clear and pressing need to dramatically improve upstream bandwidth.
    3. Cable operators’ future is one of business, infrastructure, and service convergence, with their DOCSIS networks serving as the platform for fixed-mobile convergence on a large scale.
    4. Convergence at all levels will be driven in part by the evolution of a common control and management plane across all networks and services.

I’ve dealt with the first two topics earlier this year in multiple blog posts and articles. Those two themes will certainly continue to evolve and have an impact on cable operator spending and strategic priorities for their access networks for the next year.

With this blog, I do want to spend some time considering the overall impact of convergence on cable operators’ long-term strategic plans, especially when it comes to their desire to become both fixed and mobile network operators.

The FCC’s auction of 3.5GHz CBRS licenses, which concluded in August, yielded few surprises when it came to the leading purchasers of the spectrum. Verizon and Dish Network led all bidders in terms of money spent, with Comcast, Charter, and Cox rounding out the top 5. Other major cable bidders included Mediacom, Midcontinent Communications, Shentel, and Cable One.

Comcast and Charter have been signaling for some time that they intend to build CBRS-based mobile networks in their existing cable footprints in an effort to reduce the amount of money they pay Verizon and other MVNO partners to use their networks. Their MVNO operations were always intended as a way to build a subscriber base and a brand in advance of owning their own wireless networks, even if that meant consistent EBITDA losses.

Cox, which had entered the wireless space a decade ago, only to exit after disappointing results, has signaled its intention to re-enter the wireless market through the purchase of a significant number of CBRS licenses across its cable footprint.

Finally, Cable One has taken an interesting approach, acquiring CBRS licenses but also making investments in two fixed wireless ISPs (WISPs) to provide coverage in rural and less dense areas surrounding its cable footprint.

Though they have no intention, at this point, of becoming national carriers, cable operators can certainly become competitive in their current markets, offering bundles of fixed and mobile services with the goal of reducing churn and stealing away some market share from their telco rivals.

Let’s not forget that the largest cable operators already have a very dense network of millions of Wi-Fi hotspots either through their own doing (Comcast’s Xfinity) or through their CableWiFi Alliance. Additionally, most cable operators have been deploying advanced Wi-Fi gateways in residential and small enterprise locations that typically reserve a single SSID for either open CableWiFi or Xfinity Wi-Fi subscriber access. These hotspots can very easily be turned into 5G small cells, expanding and amplifying mobile network access for their subscribers.

The dense network of hotspots and access points that the largest MSOs already have in place combined with the licensed CBRS spectrum that they have acquired should give them access to 150MHz of spectrum that they can reuse across a larger number of subscribers per individual access point.

But that type of spectrum reuse will only be possible with a vast and far-reaching deployment of CBRS small cells. In fact, according to a fascinating paper by Cisco’s John Chapman presented at Cable-Tec Expo, it “can take 200 CBRS small cells to cover an area equivalent to the area covered by one LTE macrocell.”[1]

Though the deployment of such a huge number of small cells seems daunting and costly at first, Chapman goes on in his paper to show that existing and future DOCSIS networks are completely up to the task. Firstly, a large percentage of small cells deployed by cable operators will be strand-mounted, drawing power from the existing HFC plant. Those strand-mount small cells will be deployed in conjunction with small cells located in residences to expand coverage and capacity, such that cable operators could expect to see a small cell count of anywhere from 1 to 80 per optical node, depending on the density of the area being covered, the average span length, and the number of mobile subscribers being served.

Cable operators are very accustomed to thinking about their networks as a shared resource among households and subscribers and then adding capacity when utilization rates remain consistently above 70% for any particular service group. As MSOs have been pushing fiber deeper into their networks, reducing the average number of amplifiers per node, and deploying DAA nodes in an effort to improve MER (Modulation Error Ratios,) they have prepared themselves for an access network that can handle the variable requirements of both fixed and mobile traffic.

Chapman points out in his paper, DAA nodes and CBRS small cells are essentially performing the same function: They are both RF gateways that convert RF traffic to IP over Ethernet. As cable operators continue to add capacity to their networks by pushing fiber deeper and reclaiming spectrum used for broadcast video (which they have been actively doing during the COVID-19 pandemic,) there is more than enough bandwidth to backhaul fixed and mobile broadband traffic over their existing DOCSIS infrastructure. Furthermore, with the introduction of low latency DOCSIS and the new LLX (Low Latency Xhaul) protocol, the overall DOCSIS network can deliver the 2ms of latency mandated by today’s 5g services.

Finally, today’s virtual CCAP platforms are already evolving to provide flexible data and control plane functions across cable operators’ converging fixed and mobile networks. Services like DOCSIS, 1588 and SyncE, BNG, as well as PON, can all be containerized and isolated either physically or logically, depending on the operator’s preference. The virtual CCAP becomes the centerpiece for the control and management of a diverse collection of media gateways located in the outside plant, including DAA nodes, CBRS small cells, PON OLTs, Wi-Fi access points, and cable modems.

Cable’s path to convergence is clearer now than it ever has been, from a business and service perspective to an infrastructure perspective. Chapman summarizes his paper with two (of a number) of points:

  • Today’s cable operators are tomorrow’s mobile operators
  • Behind every great wireless network is a great wireline network

I am in complete agreement with him and would add that the efforts being made by vendors to realize this at the control and management planes suggest that they agree, as well.

[1] John T. Chapman, “Small Cell Traffic Engineering: How Many Small Cells are Needed for Proper Coverage,” SCTE Cable-Tec Expo, October 2020.

Print Friendly, PDF & Email

Unlike many enterprises that have been migrating their IT infrastructure to the Public Cloud in recent years, the financial sector continues to be an important contributor to purchasing and deploying IT, which includes data center infrastructure such as servers, storage, and networking equipment. Last year, we estimated that roughly 24% of the on-premise enterprise server installed base among Fortune 2000 firms to be attributed to the financial sector, with high-frequency trading among a major use-case. We expect the financial sector to gain share in the on-premise enterprise server installed base, as traditional enterprises in the areas of oil & gas, offline retail, travel, and hospitality, and healthcare are projected to reduce spending on IT this year due to weak economic outlook. Growth in those traditional enterprise industries tends to mirror that of the broader economy, and could face a sluggish road to recovery ahead.

On Premise Server Unit Installed Base by Fortune 2000 Industry Segment (2019)

 

Other observations related to data center capex trends in the financial sector:

  • Data Center IT is moving to the Public Cloud (i.e. AWS, Microsoft Azure, and Google Cloud), and the pandemic has accelerated Cloud adoption as companies prefer an operating expenditure-based IT consumption model to conserve capital in uncertain economic times, and the ease in which cloud applications could be provisioned to a remote workforce. However, we expect the financial sector will be one of the last enterprise sectors to move the workload to the Cloud. Applications such as high-frequency trading will continue to be owned and operated by these enterprises due to proximity requirements and demands an ultra-low latency network. Regulations and data sovereignty requirements also dictate that certain data and workloads need to remain in on-premise data centers (and not move to the Cloud). We estimate that over 80% of the workloads for the financial sector have remained on-premise, versus, 20% of the workload outsourced to the Cloud.
  • Some of these major financial institutions such as JP Morgan and Goldman Sachs own and operate their own data centers, which are also complemented by colocation data centers that are operated by Equinix for instance. In the first half of 2020, we estimate a 2% average Y/Y growth in revenue generated by the financial sector among the colocation providers. That is versus an estimate of 5% average Y/Y growth in 2019. While the 1H20 growth rate is conservative, the financial sector is expected to outperform that of other enterprise sectors when it comes to colocation infrastructure spending this year.
  • The large financial institutions are innovators of server and data center architecture. While some have offloaded some requirements to the Public Cloud, they continue to invest in their own data centers. Some of the large firms operate hundreds of thousands of servers and would be able to benefit from economies of scale a Cloud service provider would. The more servers that these firms have, the more transactions and customers they can support. While this segment still purchases IT equipment from branded server and storage system vendors such as HPE, Dell, and IBM, some of the more sophisticated firms have customized servers in their data centers in an effort to streamline their architecture. They have also deployed accelerators such as Smart NICs (network interface cards) to reduce network latency necessary for high-frequency trading and to ensure secure connections. Specialized servers, equipped with accelerators such as GPUs for instance, are also deployed for AI-centric applications pertaining to predictive analysis and risk management computations.

In order to get additional insights and outlook for spending on servers and other data center center equipment by Cloud and Enterprise segments, please check out our Data Center Capex report.

Print Friendly, PDF & Email

At this week’s NVIDIA GPU Technology Conference, the company announced its groundbreaking Bluefield -2X DPU (Data Processing Unit), which combines key elements of the existing Bluefield DPU and NVIDIA’s Ampere GPU. The new DPU enables the use of AI to perform real-time security analytics, and identify abnormal traffic and malicious network activities. The building-blocks NVIDIA needs to enable high-performance and AI workloads in the data center and edge are coming together—this includes the GPU for application-specific workloads, the DPU to facilitate data I/O processing, and finally, the CPU for compute, as NVIDIA seeks to complete its acquisition of ARM.

The DPU, a terminology coined by NVIDIA, falls within Dell’Oro Group’s definition of the Smart NIC market that we track in the Ethernet Controller and Adapter research. Smart NICs are fully programmable network interface cards designed to accelerate key data center security, networking, and storage task functions, offloading valuable CPU cores to run business-oriented applications.

According to Dell’Oro Group’s latest forecast, the Smart NIC market will grow at a 26% compound annual growth rate, from $270M in 2020 to $848M by 2024, vastly outpacing the overall Ethernet controller and adapter market. We believe that this new class of Smart NIC with integrated AI that NVIDIA has introduced, to be potentially disruptive, and could expand the range of applications that have been available to Smart NICs.

As Cloud and Enterprise data centers continue to scale, we believe that Smart NICs could be a solution in achieving high network throughput, low latency, and minimal packet loss demanded by emerging applications such as high-performance computing and AI. However, there are some notable constraints vendors would need to address before we see stronger adoption of Smart NICs:

  • The scalability of Smart NICs depends on key considerations such as price and power consumption. While we are still awaiting details from NVIDIA, I believe that the inclusion of both the ARM processor and GPU in the Bluefield-2X DPU could result in higher unit cost and power consumption compared to that of alternatives. NVIDIA announced that future generations of the Bluefield, such as the Bluefield-4X DPU, will have an integrated ARM and GPU cores, which could result in total cost of ownership improvements.
  • Most Smart NICs shipped deployed are based on the ARM architecture, including that of the NVIDIA Bluefield DPU family. However, other Smart NIC vendors such as Intel, Napatech, and Xilinx have released FPGA-based solutions that have demonstrated benefits in adapting to a wide range of applications in conjunction with AI inferencing applications. I predict that both ARM-based and FPGA-based solutions will coexist and be optimized for different use-cases.
  • Extensive engineering resources and lead-time are required to bring Smart NICs to market. While the major Cloud service providers have the engineering resources devoted to Smart NIC application development, the smaller Cloud service providers and enterprises do not. It is crucial for vendors to provide value-added services and application development toolkits to customers for software implementation. NVIDIA announced the availability of the NVIDIA DOCA SDK, which enables developers to rapidly create applications and services on top of the Bluefield DPU.

Ultimately, customers will need to find a balance between the benefits Smart NICs could bring, along with the aforementioned considerations. However, I am excited that the vendor community is bringing new innovations to the market that could give customers more choices to implement the solutions that fit their requirements. As servers continue to become more commoditized overtime, Smart NICs could shift more control of the data center architecture back to the customers.