Wallaroo.AI Investor Update

Key Highlights & Details

• AI Inference is the next frontier for Enterprises… AI deployment is the roadblock. 90% of AI initiatives don’t generate ROI due to the friction in taking AI from prototype to production then scaling and maintaining it. As we scale up our customer outreach and GTM, it’s important to get into the details of the enterprise AI roadblocks and how Wallaroo.AI powers the next wave of enterprise AI by eliminating the roadblocks. We’ve identified 7 key challenges including infrastructure, deployment environments, to Privacy and security that I expand on below.

• IBM GTM: A key strategy involves building a strong Go-to-Market (GTM) motion with IBM, using them as a core distribution channel. We have formally kicked off two main areas in the last month:
  •  On-Prem Sales Motion: Partnering with IBM Infrastructure to reposition their POWER platform for AI inference with Wallaroo. This involves a global enablement program for their sales field, including training, certification, collateral, distribution SKUs, and setting quotas with exec leadership.
  • Cloud Sales Motion: Partnering with IBM Cloud sales with the goal of formalizing an offering allowing customers to use IBM’s Cloud credits (for Power VS) to deploy Wallaroo for inference. This could potentially generate hundreds of free customer trials, leading to Wallaroo AI Starter Kit sales and future enterprise upsell opportunities.

• Qualcomm: We’ve finished integration of our software with Qualcomm’s inference tooling and plan to launch with them in the current quarter, initially in a small neocloud – Cirrascale. This is a big milestone as we also are developing a GTM strategy with Qualcomm to roll the joint solution out to AI developers and expand to alternative datacenters such as Core42 and Humain that have purchased Qualcomm AI chips.

• Mara: After two months of discussion in 4Q 2025, Wallaroo is now deeply engaged with MARA Holdings, a successful Bitcoin mining company that is pivoting its business to capitalize on the AI Datacenter market opportunity using its existing energy, datacenter, and GPU assets. MARA needs a software platform for this pivot, and Wallaroo is set to begin a detailed, three-week joint pilot and technical assessment. Wallaroo estimates that deploying its platform could accelerate MARA’s time-to-market by 12-18 months. Separately, MARA’s Biz Dev team has provisionally approved an investment into Wallaroo, pending technical sponsorship from the assessment.

• Nvidia vs. Everybody Else: Competition Mounts but Where’s the Software? AI workloads need increasing amounts of compute, and for that reason not only are we seeing all sorts of new AI chips in the market beyond Nvidia GPUs – from AMD, IBM, Qualcomm and others – but investment in AI clouds and data centers beyond just the hyperscalers to address capacity and privacy/sovereignty needs. However, this leads to infrastructure silos that I’ve written about before. As this infrastructure comes on line this year, Wallaroo is well positioned to provide the software that not only allows for its effective consumption by enterprises but also allows for portability of AI workloads. Our IBM and Qualcomm partnerships back this up, but also newer engagements with Mara that is building sovereign AI data centers. More below.

• Investors are starting to make big moves – VCs are now paying attention to AI inference software and SaaS, as indicated by the valuations and funding that Fireworks.AI, Baseten, Modular, and Inferact. I’ll explain where these startups have their biggest value proposition compared to Wallaroo.AI, and why I think Wallaroo can quickly follow in their footsteps while providing a unique solution to enterprises, AI developers, AI silicon OEMs, and new AI clouds and data centers.

• Fundraising: Apart from Mara, we have progressed several conversations to raise an additional $2M this quarter via a convertible note ahead of a series B later in the year. There is still an opportunity for smaller investors to get in now – we just closed $200K in January. If you are interested, please get in touch with me or Pete.

• Air Force: We successfully completed a key space project with the USAF Research Lab on November 25, 2025. The readout highlighted our Wallaroo.AI and Loft Orbital partnership to bring enterprise-grade AI inference capabilities to satellites in Low Earth Orbit (LEO). This transitions AI work from labs to real-time execution, focusing on deploying image processing models to detect objects and return condensed results, significantly reducing data latency. This solution has immense defense and commercial applications. Great feedback will allow us to expand this joint solution, with AFRL endorsement, to other projects. For example, we plan on obtaining funding for a US Space Force project to use it for range safety and clearing. 

• Product: Wallaroo has launched several key product and R&D initiatives:
  • Wallaroo AI Starter Kit for IBM Power: A turnkey solution for optimized enterprise AI deployment on IBM Power, offering simple, container-based deployment in on-prem/hybrid environments.
  • Qualcomm (Re)Imagined Playground Integration: Wallaroo now integrates with Qualcomm Imagine Playground, allowing developers to easily migrate and deploy AI apps/agents onto Qualcomm’s AI 100/200 chipsets for portable, efficient, and optimized enterprise-grade Agentic AI.
  • AMD GPU Support: Upcoming full support for AMD MI350/MI400 GPUs will offer a cost-effective NVIDIA alternative with high-performance LLM runtime and production inference, simplifying heterogeneous infrastructure migrations.
  • R&D and Innovation: The team focuses on optimizing the AI inference stack via heterogeneous prefill-decode disaggregation for LLM inference to improve AI hardware utilization, and researching LLM cache & context optimizations to reduce token costs and accelerate inference, especially for enterprises with known/finite datasets.

Wallaroo.AI announces Pete Agresta as Chief Revenue Officer leading GTM

I’m really excited to have Pete join Wallaroo.AI to help turbocharge our growth at this pivotal moment in the market. Pete brings decades of experience in sales and leadership, building and leading high performance teams globally. Most recently, he served as CRO at Nasuni, a provider of cloud storage solutions leading to a successful exit. His successful track record fostering customer partnerships, growing market share and creating global scale includes leading enterprise sales at Pure Storage, senior sales leadership at Cisco, Alliance Bernstein as well as bringing new solutions to market at growth stage technology companies. 

Read the Announcement.

Enterprise AI deployment is the roadblock. Wallaroo solves that.

We briefly discussed this in the last investor update, but let’s dig deeper into the seven critical issues facing enterprises getting in the way of AI ROI. Not only have we had dozens of conversations with Enterprise AI leaders over the last year, but numerous surveys from Gartner, Mckinsey, and MIT have reinforced our direct experience.

These challenges cause enterprises to abandon initiatives or take an inordinate amount of time and engineering to get into production, eliminate actual ROI due to cost and maintainability, or create unacceptable financial, reputational or regulatory risk. 

Another way to think about these challenges is to look at the stage that an enterprise is in as they seek to operationalize AI and get to business impact. Due to talent, complexity, or cost issues an enterprise may struggle to transition from one stage (e.g. pilot) to another (AI in production).

Wallaroo.AI removes these roadblocks via a combination of software and AI enablement.

Pete Agresta, our new CRO has a nice blog that expands on this. Read More.

IBM GTM

Building an effective GTM motion with IBM is one of our key strategies. – IBM represents a powerful distribution channel but we don’t underestimate the challenges of making this effective in practice.    Our 2 key activities: Enable their On-Prem sales team and establish a partnership with their Cloud team.

On-Prem Sales Motion: 

• Recall our current partnership is with the IBM Infrastructure division (~22% of IBM by Rev) where their strategic goal is to re-position their POWER platform as an effective platform to execute AI inference workloads – exactly what we enable.
• While the POWER platform + Wallaroo is very effective, the size and use cases are constrained by the underlying capability of the silicon making effective use-case positioning extremely important. 
• We are executing a structured enablement program for their global field organization:
  • Hands-on training for technical champions in each market, certification for their sellers, use-case collateral, partner enablement, distribution SKU’s, Use case identification and proposal templates etc. 
• We have identified exec sales leadership sponsors in each geography and are establishing 1st half goals and quotas for the sales teams. 

Cloud Sales Motion: 

• We are building a similar partnership with the IBM cloud sales teams. 
• IBM have been executing a Cloud seeding strategy already
  • IBM cloud runs Power VS (virtual servers) along with GPU’s from Nvidia and AMD. For each on-prem customer sale of IBM Power, the customer gets Cloud credits which they can consume on Power VS.  
• We are hoping to formalize an offering allowing their customers to utilize those credits to deploy Inference using Wallaroo.  This could get 100’s of customers trying Wallaroo at no initial cost to them and generate Starter Kit sales for us with lots of future upsell opportunities.  

Read More: The AI Lifecycle: Turning Vision Into Value With Wallaroo and IBM Power

MARA

We are deeply engaged with MARA Holdings (Nasdaq; MARA) which has high strategic potential as a customer, strategic partner, and/or potential investor.  MARA was formerly Marathon Digital Holdings and one of the earliest and most successful Bitcoin mining companies.  They built their business by vertically integrating and securing long-term, cost-effective energy contracts where they built data-centers running GPU’s and ASICs to mine bitcoin.  As other bitcoin miners have recognized (ex CoreWeave; Nasdaq CRWV) – MARA’s assets  – energy, datacenters, GPUS – are exactly what’s needed to build AI Datacenters and capitalize on the market opportunity.  As such, they are investing heavily in pivoting their business model.  

• MARA recognizes they need to either build or acquire a software platform to enable their business strategy.
• We have planned and are about to jointly execute a detailed POV and technical assessment of our Platform for current and future needs that will take place over the next ~3 weeks.
• We conservatively estimate deploying Wallaroo would accelerate MARA’s time-to-market by 12-18 months and are jointly building a business case to document this outcome.
• In parallel to the technical assessment, MARA Biz Dev team have conducted due-dilligence of an investment into Wallaroo.  They have shared that it was provisionally approved through their investment committee pending technical sponsorship.

Air Force

We completed our first TacFi project with AFRL on November 25, 2025. The final readout to AFRL stakeholders focused on how the partnership between Wallaroo.AI and Loft Orbital designed to bring advanced AI inference capabilities directly into space. 

The solution focuses on deploying enterprise-grade AI inference capabilities in-orbit. It transitions AI work from a laboratory setting to real-time execution on satellites currently in Low Earth Orbit (LEO). The primary goal is to deploy image processing models that can detect objects and return condensed results, thereby reducing the latency typically associated with sending massive amounts of raw data back to Earth.

The solution leverages several cutting-edge technologies to bridge the gap between AI development and space infrastructure:

• Wallaroo AI Orchestration Platform: A unified platform that provides a frictionless way to ship, run, and orchestrate AI inference across diverse infrastructure, from cloud (on-the-ground testing) to the edge (satellites in space).
• Loft Orbital’s “Virtual Missions”: A service that allows users to deploy software applications to a fleet of LEO satellites without needing to develop or launch their own hardware.
• Onboard Edge Compute: Satellites like the YAC-4 constellation are equipped with powerful onboard processors (Nvidia and AMD) and diverse sensors (Optical, Thermal, Hyperspectral) to support real-time data fusion and AI execution.

Key Benefits of this solution include:

• Actionable Intelligence: Enables immediate detection and response to environmental changes, vessel movements, and national security threats directly from space.
• Cost and Resource Efficiency: Provides up to 80% less infrastructure overhead for AI inference and insights downlink for application quality analysis.
• Improved Reliability: By leveraging existing space assets with flight heritage, organizations can complete missions without the technical and schedule risks of launching new hardware.
• Scalability: Supports a wide range of models and hardware architectures, ensuring that AI initiatives can scale across constellations.

The general feedback on this collaboration has been extremely positive as we look for further opportunities to expand to newer use cases and infrastructure. 

As a result of this initial TacFi, we are applying the Wallaroo<>Loft joint solution, with endorsement from AFRL, to other use cases and project collaborations. For instance, the TacFi with the USSF will feature this joint solution to complete a critical project focused on range safety and clearing:

• Context
  • “Range Clearing” is the act of making sure that the “Range” is cleared for spacecraft takeoff – clear of other vehicles, debris, weather
  • The US has two ranges (Cape Canaveral in FL and Vandenberg in CA) with extensive infrastructure to make this possible
  • Goal of a next gen range safety system would be to clear ranges from orbit in order to make lighter-weight ranges possible.
• Scope:
  • Deployment of  a suite of AI models to an actual satellite focused on Range Characterization, a subset of Range Safety
  • Wallaroo.AI and Loft Orbital are partnering to:
    • Deploy Wallaroo’s AI inference capabilities in-orbit
    • Deploy a wider range of image processing AI models tuned for range characterization
    • Proactively update models on orbit using deployment insights.

More reading: Wallaroo is enabling “on-orbit” AI.

Product

• Wallaroo AI Starter kit for IBM Power: We launched a turnkey solution designed to bridge the gap between AI development and enterprise deployment. It specifically targets organizations using IBM Power (including Power10) looking to move AI applications into production and realize AI ROI without complex infrastructure re-engineering. Core capabilities include:
  • Optimized Performance: Specifically engineered to leverage IBM Power’s unique hardware features, such as MMA (Matrix Multiply Assist) and Spyre Technology, to achieve high-performance AI inferencing on CPUs & accelerators that can rival traditional GPU setups.
  • Turnkey Deployment: Includes pre-built AI APIs and an optimized AI runtime. It allows users to deploy models (including Computer Vision and LLMs) into an LPAR (Logical Partition) in minutes using container tools like Podman.
  • Flexible Footprint: Supports deployments across the hybrid cloud spectrum—whether on-premises, in the IBM Cloud (PowerVS), or at the edge.

• The Qualcomm (Re)Imagine(d) Playground – Powered by Wallaroo: We have completed the development of the Wallaroo integration onto the Qualcomm Imagine Playground offering. The Qualcomm Imagine playground key features powered by Wallaroo enable AI developers to easily migrate AI applications & agents from common environments and silicon (e.g AWS running NVIDIA) to Qualcomm’s AI 100 & 200 chipsets running in secure environments. As a result, AI developers and enterprises can take advantage of Wallaroo’s automated deployment, ease of use and production-grade AI inference capabilities to unlock seamless AI portability onto Qualcomm’s AI silicon, enabling enterprises ultimately with  power efficiency, Improved TCO and optimized enterprise-grade Agentic AI applications with their infrastructure. The joint solution will be announced in Q1 2026. Stay tuned for more updates!

• AMD GPU support: We are gearing up to launch the full support of the latest AMD GPU chipsets (MI350 and MI400) with Wallaroo’s state-of-the-art AI inference stack including portability from NVIDIA across environments (cloud and Edge), a high-performance LLM runtime (SGLang) and production-grade AI inference (scaling, load-balancing, telemetry, logging and secure encryption) . This is a major product roadmap milestone for Wallaroo, given the recent demand for AMD GPUs, as it enables enterprises and infrastructure providers with a viable cost-effective alternative to NVIDIA unlocking simpler, faster and AI migrations across heterogeneous AI infrastructure deployments.

• R&D and Innovation : As we unlock further functionality for AI deployments across a wider range of silicon options, we are investing additional R&D time exploring further optimizations within the Wallaroo AI inference stack. With the recent push from NVIDIA for prefill-decode disaggregation for LLM inference, we are implementing new enhancements in the following areas:
  • Heterogeneous prefill-decode disaggregation with a simpler AI developer experience, taking advantage of Wallaroo’s built-in compute node orchestration capabilities which are leveraged today to route AI inference workloads across CPUs and GPUs for our cloud and on-prem enterprise customer base. This is relevant for enterprises and infrastructure providers looking to optimize GPU utilization costs across high-end AI silicon such as NVIDIA for prefill and emerging AI silicon such as Qualcomm and IBM Spyre for decode. 
  • LLM cache & context optimizations: Given that Prefill-decode disaggregation will incur additional AI silicon deployments, and as a safeguard to unnecessary costs resulting from such deployments, we are investing additional effort researching further cache optimization techniques. This is super important for enterprises working on known/finite datasets as these caching techniques enable storing and reusing previously generated responses across similar search/prompt contexts to accelerate LLM inference times and drastically reduce token costs, and therefore AI silicon utilization.

More reading: The Wallaroo AI Starter Kit for IBM Power.

Nvidia vs. Everybody Else: Competition Mounts but Where’s the Software?

The December 12 WSJ article “Nvidia vs. Everybody Else: Competition Mounts” was very timely. 

The key thrust is that Nvidia, the dominant player in the AI chip market for the past decade, is encountering increasingly intense competition from multiple directions. Established tech giants such as AMD, Broadcom, Intel, IBM, and Qualcomm, along with hyperscalers like Google (with its TPUs) and Amazon (with its Trainium chips) who are now offering their AI chips to external clients, are directly challenging Nvidia’s flagship GPUs for AI workloads. While Nvidia remains the current market leader with significant sales and a comprehensive software ecosystem, this evolving landscape compels the company to adapt to a rapidly diversifying and competitive market.

What’s missing from the WSJ article for the most part is a lack of appreciation of the role that software plays in helping drive Nvidia’s chip adoption by customers. Beyond CUDA, Nvidia’s low-level software library for AI developers, Nvidia has built or bought an extensive set of AI software capabilities for orchestrating AI and running high-performance inferencing on its hardware, collectively known as Dynamo.

For other chip suppliers to take market share from Nvidia, they must not only deliver compelling AI chips, but provide an integrated Software + Hardware stack so AI developers can both easily deploy AI on new AI silicon, and also migrate appropriate workloads from Nvidia. Without such software, this new AI silicon presents as much complexity as it does opportunity to reduce costs or increase capacity and scale. I’ve written about the challenge of infrastructure silos here.

At Wallaroo, we’ve been building a unified AI orchestration and inference software stack for a while, even working directly with many of the chip suppliers mentioned in the WSJ article. What’s exciting is to see our vision and approach getting validated not only by the likes of the WSJ, but by the market as a whole. For us it’s not about Nvidia vs. other AI chips, but how do enterprises and infrastructure providers make it easy to deploy, scale, and operate AI in Nvidia + other AI chips, Cloud + on-prem and edge world. As $100s of Billions of committed AI infrastructure comes online, AI software will be increasingly important. At Wallaroo.AI, we’ve got the software and we are ready.

More Reading: From AI Infrastructure Silos To AI Infrastructure Freedom

Investors are starting to make big moves in AI inference & orchestration

Even though most AI initiatives are still stuck in the experiment (pre-production) phase, the investment in AI infrastructure has been rapidly shifting from model training and testing to production inference. As with any modern hardware, software is essential to help AI developers, enterprises, ISVs, and others actually deploy, run and use the AI infrastructure. While Nvidia, and the three main hyperscalers provide a set of software tooling to help do that, these solutions are incomplete from both an infrastructure and developer perspective.

This graphic illustrates the challenge facing AI developers in trying to bridge the gap between their pilots/prototypes in a development environment and scalable, enterprise-grade AI products and services deployed onto the underlying infrastructure.

Of course the devil is in the details when it comes to understanding specific software needs and capability. Referring back to the enterprise AI roadblocks, we can organize these needs into Day 1/Day 2/Day 3 drivers.

Startups focused on state-of-the-art inference with recent large funding rounds ($150M+) include Fireworks.AI, Baseten, Modular, and Inferact.

• Fireworks.AI is primarily an inference-as-a-service (IaaS) offering and does not offer truly portable and heterogeneous inference from any cloud to edge, nor centralized orchestration, management of AI workloads in diverse or hybrid environments
• Baseten & Modular, have both IaaS and inference software offerings, but with minimal support for Day 1 and Day 3 tooling (easy packaging and deployment, and centralized orchestration and management of workloads)
• Inferact is a new Seed stage startup focused on commercializing the open-source vLLM engine for generative AI models

Wallaroo is the only software with unified orchestration/operations and fast inference, portable across heterogeneous infrastructure. So unlike Fireworks.AI our software enables enterprises and infrastructure providers to quickly build and deploy their own inference services on any infrastructure, and unlike Baseten, Modular, Inferact, we aim to integrate available technologies in the inference stack to offer a unified orchestration, deployment and operations platform.