Ampere Wants To Help Channel Rein In AI Server Costs With Systems Builders Program

In an interview with CRN, Ampere Chief Product Officer Jeff Wittich says the company hopes to grow its Arm CPU business by enticing more channel partners through its Systems Builder program, which seeks to create more flexibility and lower costs for AI computing infrastructure.

Ampere Computing is hoping to grow its Arm CPU business by enticing more channel partners to work with the chip designer through its new Systems Builders program, which seeks to create more flexibility and lower costs for AI and cloud computing infrastructure.

Unveiled Thursday, the Ampere Systems Builders program is bringing together several IT infrastructure players, including U.S.-based Supermicro, to speed up the development and delivery of server platforms for AI and cloud-native computing, with a focus on using standards such as the Open Compute Project’s data center modular hardware system (DC-MHS) to drive “interoperability, customization and faster time to market.”

The other companies participating in Ampere’s program include Giga Computing, the server arm of Taiwanese electronics giant Gigabyte, as well as Pegatron-owned ASRock Rack in Taiwan and U.S.-based Jabil. Ampere is also working with other chip designers, namely U.S.-based Broadcom and South Korea-based Rebellions, whose components can be incorporated into systems developed as part of the program.

These companies are using “shared engineering resources, open specifications and coordinated platform validation” to co-develop solutions, which includes “turnkey solutions that integrate hardware, software and services tailored” for AI inference, according to Ampere.

“Through the Ampere Systems Builders program, Supermicro will support customers with MegaDC modular systems integrating DC-MHS-based Ampere motherboards, supporting both CPU-only and CPU-plus-accelerator configurations,” said Michael Clegg, vice president and general manager of edge at Supermicro, in a statement.

““These systems are specifically designed to meet the needs of high-volume AI inference deployments across cloud and enterprise markets, where flexibility, performance and efficiency are critical,” he added.

The announcement was made nearly two months after Japanese investment giant SoftBank Group unveiled a deal to acquire Santa, Clara, Calif.-based Ampere for $6.5 billion in cash as part of a push to fuel a new wave of AI infrastructure. The deal is expected to close in the second half of this year.

Founded in 2017 by former Intel executive Renée James, Ampere is among a growing number of companies challenging Intel and AMD’s x86 duopoly with CPUs that are based on or compatible with the Arm instruction set architecture.

Ampere Hopes To Grow In The Channel ‘Very Fast’

In an interview with CRN, Ampere Chief Product Officer Jeff Wittich (pictured above) said the company’s Systems Builders program has two goals.

The first goal is meant to make its products, including the AI-optimized AmpereOne M processors it released last year, more accessible in the channel. While the company has been working with some systems integrators and other kinds of channel partners, it was largely focused on selling to hyperscalers and large OEMs in the past several years.

While channel partners currently don’t represent a large percentage of Ampere’s sales volume now, Wittich said, he considers them “strategically” important and said the company hopes to grow its channel business “very fast.”

“What we hope is the next step is that we get a lot of attention and interest out of those channel partners, and we can start to pull them into the program,” he said.

For the channel partners Ampere works with, it provides resources such as MDF and volume rebates, but they are done in a “more bespoke manner,” according to Wittich.

“We have to tailor these to the channel providers right now versus creating a completely one-size-fits-all solution. Otherwise, I fear that it won’t end up actually driving unit sales,” he said.

However, Wittich said, Ampere hopes to eventually make its engagements with channel partners more programmatic over time like it has done with other programs.

“This is what we’ve done time and time again,” he said.

An Alternative To Expensive, Vertical AI Hardware Solutions

The other goal of Ampere’s Systems Builders program is to provide channel partners and customers with more flexibility and choice for AI computing infrastructure.

Wittich said this is important because it has become “more difficult” for original device manufacturers, systems integrators and distributors to innovate in the AI space due to major infrastructure providers like Nvidia focusing on vertical hardware solution stacks, where system designs are largely predetermined before reaching partners.

This contrasts with general-purpose computing infrastructure, where “it was relatively easy to go grab one new [server chassis], configure it the way you wanted, add a bunch of components [and] really customize it,” according to the executive.

“There’s really not a lot of configurability, and I get it. Part of it comes from these are complex platforms, and so while they’re new, it can be easier to spec everything out and kind of limit the choice. On the other hand, it doesn’t necessarily help this part of the market where you’re looking for the ability to differentiate in some way,” he said.

Wittich said preconfigured hardware platforms like the ones from Nvidia were likely required to train the massive, foundational AI models that are often the starting point for innovations throughout the industry.

But not all AI workloads require super-expensive hardware platforms that require an unprecedented amount of power, especially when it comes to inference, he added.

“If we get the wrong solution, we’re going to consume potentially multiple times more power [and] a ton more money, which at some point it just means that the business model becomes unsustainable but in a way that wasn’t necessary, and it also makes it really difficult for other players to participate in the market,” Wittich said.

Ampere’s CPUs, for instance, can deliver “more than enough tokens output” on AI models with up to 15 million parameters, with an AmpereOne M processor providing “well over 400 tokens per second” on a model like the 8-billion-parameter Llama 3.1 model, he added.

With the Systems Builders program, channel partners and customers will have options to configure CPU-only systems or those that also use PCIe-based accelerator chips when higher performance is required. These systems support PCIe-based solutions from member companies like Rebellions as well as non-member companies like Nvidia.

“It’s a big piece of how we deliver a full set of solutions out in the market,” Wittich said.

Systems Builders Program Sets The Stage For New Ampere Chips

Regarding SoftBank’s pending acquisition of Ampere, Wittich said he couldn’t speak “a lot about what may or may not happen” after the deal closes but noted that it will give the company the “ability to scale” and go after a “bigger market over time.” There are “no changes” to its product road maps, support and customer engagements, he added.

The executive said the acquisition will also help Ampere “realize our ambitions,” such as the AmpereOne Aurora chip it unveiled last year that will scale up to 512 custom, Arm-compatible cores and feature integrated AI acceleration for the first time.

“You can see how that comes with a lot of complexity. Not only is that more [intellectual property] that we’re developing, more products that we're developing, but it’s also a lot of work on the system side because now we have a variety of form factors that we’ll be supporting, not just a socket and CPU,” Wittich said.

With the Systems Builders program, Ampere is starting to “lay the groundwork for some of that stuff,” according to the executive.

“[The acquisition] gives [us] more tools to do that, and it gives us more ability to realize our ambitions faster than we otherwise could have,” he said.