Fusion AI In-a-Box Shown to Jensen Huang at GTC Berlin [2026] - Tech Insider

Elias Virtanen October 25, 2026 13 min read A Moroccan company walked up to one of the most watched executives in tech this week and showed him a box. On October 22, 2026, ABA Technology presented Fusion AI In-a-Box to Jensen Huang, CEO and founder of NVIDIA, during GTC Berlin 2026. The pitch is simple on paper: run your AI models, agents and data on your own hardware, with no trip to a foreign cloud. According to La Vie éco , which cited a company statement dated October 22, NVIDIA selected ABA Technology to take part in the event. The outlet describes a product built from a software platform and an integrated computing infrastructure, sold in two configurations that can run apart or together. The story fits a pattern we have tracked all year. NVIDIA keeps recruiting partners well beyond the chip business, as our report on the NVIDIA and Nokia AI-RAN push, which signed up nearly 20 telecom operators , showed. ABA Technology now brings a sovereign, packaged AI stack to the same ecosystem. This analysis covers what was announced, what remains unconfirmed, and what buyers should watch next. · Google · Preferred Sources Don't miss new tech stories on Google Add Tech Insider once in the Google app and our stories appear in your news suggestions. Add Now What ABA Technology Presented at GTC Berlin 2026 GTC Berlin is an NVIDIA-organized event focused on accelerated computing and AI infrastructure. Industries.ma reports that ABA Technology’s team presented Fusion AI In-a-Box to Huang on October 22, 2026, and calls the selection a notable recognition for a Moroccan firm. The outlet frames the appearance as a step onto the international stage for a company that markets itself around sovereign AI. The details matter, so here is what the reporting supports. ABA Technology says NVIDIA chose it to participate in GTC Berlin 2026. The company then presented its product to Huang during the event. Nothing in the available coverage says Huang endorsed the product, invested in ABA Technology, or signed a formal partnership with it. A presentation is a presentation, and readers should treat any stronger framing as unconfirmed. ABA Technology’s own news page also lists an earlier announcement, dated September 15, 2026, about its Fusion AI offering being selected by NVIDIA for GTC Berlin. That timeline tells us the Berlin slot was known for about five weeks before the Huang presentation, which is typical for conference exhibitors preparing a booth and a demo. Fusion AI In-a-Box Architecture: Software Plus Integrated Compute La Vie éco attributes the core description to ABA Technology: the solution rests on pairing a software platform with an integrated computing infrastructure. In plain terms, the vendor ships the applications layer and the hardware layer as one package. Buyers do not assemble GPUs, orchestration software and agent tooling from separate suppliers. The company says the package covers four parts of the AI value chain. Those are data management, the use of multiple models, the deployment of intelligent agents, and the compute resources needed to run all of it. Industries.ma repeats the same four-part framing, which suggests it comes straight from the company’s release rather than independent testing. Why the In-a-Box Format Appeals to Enterprises Most enterprise AI projects stall between a working prototype and a production system. Teams need data pipelines, model evaluation, security review and monitoring before anyone trusts an agent with real work. A bundled appliance promises to shrink that gap by delivering a pre-integrated stack. Whether Fusion AI In-a-Box delivers on that promise depends on details the company has not yet published. Fusion AI Factory In-a-Box: Five Components for Building AI Apps The first configuration, Fusion AI Factory In-a-Box, targets teams that build AI applications and wire them into business operations. La Vie éco says it offers an environment to prepare data, evaluate and adapt models, design intelligent agents and automate workflows. The goal, per Industries.ma, is to speed the move from experiments to operational use cases connected to an organization’s data and processes. Five software components make up the Factory configuration. Industries.ma adds that the applications built on it should connect both to IT systems and business software and to industrial equipment and operational technology. That OT angle separates it from many chatbot-first enterprise offerings. Component Configuration Reported role Fusion Data Factory Data preparation and management Fusion Forge Factory Model evaluation and adaptation Fusion Agentic Factory Design of intelligent agents Fusion Automation Factory Workflow automation Fusion Performance Factory and Inference Observability and performance tracking Fusion Forge Inference Inference Inference management and orchestration Fusion Data Platform Inference (optional) Data management as an add-on The role descriptions in the table are our plain-language reading of the company’s wording as relayed by La Vie éco and Industries.ma. ABA Technology has not published product documentation we could check, so treat each role as a marketing-level label, not a technical specification. Fusion AI Inference In-a-Box: Running Models in Production The second configuration handles the part of AI that costs money every day: inference. Fusion AI Inference In-a-Box is built to run models after they reach production. It combines Fusion Forge Inference, which manages and orchestrates inference jobs, with Fusion Performance, which monitors performance and observability. Fusion Data Platform is optional. ABA Technology says the design aims to use compute efficiently while keeping a lid on processing cost, energy use and latency, and while tracking service quality. Both La Vie éco and Industries.ma add that capacity can scale with usage, volume and performance needs. No throughput numbers, GPU counts or latency targets accompany those claims. Separating the two configurations makes commercial sense. A hospital or ministry may buy a finished model and only need a safe place to serve it. A bank building custom agents needs the full Factory. Letting customers deploy either one alone, or both together, widens the addressable market without forcing a single price point. Local Deployment and the Data Sovereignty Pitch The headline claim is local operation. Per La Vie éco, Fusion AI In-a-Box is designed to run on the infrastructure of the companies and public bodies that use it. Unlike solutions that ship data to external cloud services, it lets organizations keep control of their information, pick their own models and steer their own AI agents. That argument lands because data control now shapes AI buying decisions in regulated sectors. Our coverage of five big US banks blocking third-party AI agents shows how quickly finance teams lock down outside tools once data leaves their perimeter. A sovereign appliance targets exactly that anxiety. The risk of unsanctioned tools adds pressure. Workers already bring AI into the office on their own, which our analysis of shadow AI use by 45% of workers documented. An approved on-premises option gives IT departments something to offer in place of a ban. The NVIDIA Connection: What Is Confirmed and What Is Not Because the headline pairs a small vendor with a famous name, it helps to separate the verified from the implied. The table below sorts the claims by what the named outlets support. Claim Status Basis ABA Technology presented Fusion AI In-a-Box to Jensen Huang on October 22, 2026 Reported La Vie éco, Industries.ma NVIDIA selected ABA Technology for GTC Berlin 2026 Company statement La Vie éco, relaying ABA Technology Two configurations, deployable separately or together Reported La Vie éco, Industries.ma Product integrates NVIDIA Blackwell chips Reported in separate coverage 7news Morocco Jensen Huang endorsed or partnered with ABA Technology Unconfirmed No source Price, benchmarks, capacity, power draw Not published No source Commercial availability date Not published No source The pattern is familiar. Conferences let exhibitors present to executives, and companies understandably publicize the moment. That does not make it a commercial agreement. Readers who see this story shared as an NVIDIA deal should check the wording closely, because the sourcing supports a presentation and a selection to attend, nothing beyond that. Reported Blackwell and NVIDIA AI Enterprise Integration Earlier coverage from 7news Morocco says the compact Fusion AI-in-a-Box system uses NVIDIA Blackwell chips and was developed with global partners. That report dates to the platform’s April launch, well before the Berlin appearance. A separate report describes integration with NVIDIA AI Enterprise software, with a stated aim of energy-efficient, cost-controlled inference that preserves data sovereignty. We could not confirm which Blackwell part sits inside the box, how many accelerators ship per unit, or how much memory each configuration carries. Those facts decide everything about model size, concurrency and cost per token. Without them, no one can say whether the appliance suits a small clinic running one assistant or a ministry serving thousands of staff. ABA Technology’s own launch page, titled around a sovereign AI factory made in Morocco, and Industries.ma both say Fusion AI was developed with NVIDIA and Dell Technologies. That points to a conventional enterprise hardware chain, with NVIDIA silicon and software, a major server maker, and a local integrator adding the platform layer and the customer relationship. From April Launch to Berlin: ABA Technology’s Timeline The Berlin presentation is the latest step in a campaign that started months ago. Based on the verified coverage, the sequence looks like this. April 2, 2026: ABA Technology launched Fusion AI as a unified platform for sovereign AI, according to 7news Morocco, with Fusion AI-in-a-Box as the centerpiece. April 2026, GITEX Africa: TechAfrica News reported that the company signed a strategic partnership with Atos to scale Fusion AI across Africa, targeting sectors such as healthcare, finance, industry, public services and research. September 15, 2026: ABA Technology’s news page lists the announcement that NVIDIA selected Fusion AI for GTC Berlin 2026. October 22, 2026: The team presents Fusion AI In-a-Box to Jensen Huang in Berlin, per La Vie éco and Industries.ma. Read together, those dates show a company moving from a national launch to a continental distribution deal to a global conference slot in about six months. The pace suggests a deliberate effort to build credibility abroad before commercial sales scale up. How It Compares With Other Ways to Run AI On-Premises Fusion AI In-a-Box does not exist in a vacuum. Organizations that want local AI have at least three routes, and each trades control, speed and effort differently. The comparison below is qualitative because ABA Technology has published no numbers to benchmark. Approach Data stays local Integration effort Vendor lock-in risk Best fit Packaged appliance (Fusion AI In-a-Box) Yes, by design Low, if pre-integrated as claimed Medium to high Public bodies and regulated firms without deep AI staff Self-built GPU stack Yes High Low to medium Teams with strong MLOps skills Hyperscaler cloud AI No, unless special regions apply Low High Fast prototyping and variable workloads Desktop-class local AI box Yes Medium Medium Small teams and individual developers At the small end, NVIDIA’s own desktop hardware has already pushed local AI into reach for individuals. Our look at the DGX Spark 64GB sale at $4,999 against the 128GB model at $6,950 shows where that tier stands on price. A fleet-scale appliance for ministries sits in a different category, with different support and compliance needs. Developers who want to test local models first can try the route in our guide to running an LLM locally with llama.cpp and Ollama . That exercise makes the appliance pitch easier to judge, because it exposes the setup work a bundled product claims to remove. Market Impact: Why Sovereign AI Appliances Matter Now Governments and large companies increasingly ask where their AI runs and who can see the data. La Vie éco ties the product to the growing weight of digital sovereignty in AI adoption plans, especially for firms and institutions that want control over their infrastructure and information. That demand is real, and it favors vendors who can offer an on-site option with a credible hardware partner. For NVIDIA, regional integrators extend its reach into markets where the company does not sell directly to every ministry or mid-sized enterprise. The more partners package Blackwell hardware and NVIDIA AI Enterprise software into finished products, the more AI spending flows through its stack. ABA Technology gains borrowed credibility from the association. Both sides benefit even without a formal deal. For competitors, the message is that sovereignty is becoming a product category, not a footnote. Server makers, cloud providers offering sovereign regions, and software firms building agent platforms all chase the same buyers. Our report on SUSE AI Factory adding NVIDIA OpenShell for agent security shows enterprise Linux vendors entering the same space from a different direction. Security and Governance Questions Buyers Should Ask Keeping data on site removes one risk and leaves others. Any buyer evaluating Fusion AI In-a-Box, or a rival appliance, should press the vendor on the points below before signing. Model choice: Which models ship by default, and can the customer load others? The company says organizations choose their models, but the supported list is unpublished. Agent permissions: What controls limit what an agent can read, write or trigger inside IT and OT systems? Update path: How do security patches and model updates reach an air-gapped or restricted site? Observability: Fusion Performance is described as tracking performance. Does it also log agent actions for audit? Exit terms: Can the customer export data, fine-tuned models and workflows if they leave? Support model: Who answers the phone when a GPU fails, ABA Technology, a hardware partner or a reseller? Local deployment is not automatically secure. An unpatched appliance on a flat network can be worse than a well-run cloud tenant. The sovereignty argument holds only if the operator also maintains the box. What Is Missing: Price, Performance and Availability The coverage we reviewed contains no price, no benchmark results, no capacity figures, no dimensions and no power specifications. It also names no customer deployments tied to the Berlin product and gives no commercial availability date. For a hardware-plus-software appliance, that is a long list of blanks. Conference announcements often work this way. The first goal is awareness, and spec sheets follow once sales teams are ready to quote. Still, procurement officers cannot compare an appliance on sovereignty alone. They need cost per user, tokens per second under load, rack space and kilowatts. Until ABA Technology publishes them, the offer stays a concept for most evaluators. One more caveat applies to the sourcing. Nearly all of the English and French coverage repeats the same company statement. We found no independent hands-on review, no third-party test and no comment from NVIDIA on the product itself. That is normal for a launch week, but it caps how firm any judgment can be today. Historical Context: From Cloud-First to Keep-It-Local For most of the last decade, enterprise strategy leaned toward the public cloud. Data protection rules such as the EU’s GDPR, in force since 2018, began pulling some workloads back, and AI sharpened the debate. Training and serving models means moving large volumes of sensitive text, images and records, often to infrastructure owned by foreign firms. Vendors responded with sovereign cloud regions, confidential computing and, increasingly, boxed systems that sit in the customer’s own data center. Fusion AI In-a-Box belongs to that third wave. What makes it unusual is its origin. A Morocco-based group, backed publicly by national ministries according to 7news Morocco, is selling a sovereignty product rather than buying one from a US or European supplier. Predictions: What to Expect Over the Next 12 Months These are our analytical judgments, not reported facts. Each could change if ABA Technology or NVIDIA releases new information. Specs will follow the stage time. Expect ABA Technology to publish configuration details, likely tied to named hardware, within the next few quarters, since buyers will demand them. More regional integrators will copy the model. Packaged, sovereign AI stacks built on NVIDIA hardware will appear from other countries’ system builders as governments push for domestic control. Africa and the Middle East will be the first test markets. The Atos agreement and the company’s regional positioning point there before Europe or North America. Inference, not training, will drive sales. Most buyers need to serve existing models safely, so the Inference configuration may attract earlier orders than the Factory. Security scrutiny will rise. As agents gain access to OT systems, auditors will demand logging and permission controls, and vendors without them will lose deals. For broader context on models, hardware and enterprise AI, browse our AI and machine learning coverage . Bottom Line on Fusion AI In-a-Box The takeaway: ABA Technology earned a spot at GTC Berlin 2026, presented Fusion AI In-a-Box to Jensen Huang on October 22, and offers a two-part product aimed at local, sovereign AI. The architecture is coherent, the market need is real, and the NVIDIA association adds visibility. The proof is still missing. Without price, performance data, availability or independent testing, the product remains a promise. Watch for published specs, named customers and any statement from NVIDIA that goes beyond hosting the event. Those three signals will show whether this was a conference moment or the start of a real product line. Frequently Asked Questions What is Fusion AI In-a-Box? It is ABA Technology’s sovereign AI product, which pairs a software platform with an integrated computing infrastructure. It covers data management, multiple models, AI agents and compute, and is designed to run locally. Did Jensen Huang endorse ABA Technology? No source we reviewed says so. The coverage reports that ABA Technology presented the product to Huang at GTC Berlin 2026 on October 22. An endorsement, investment or formal partnership with Huang is unconfirmed. What are the two configurations? Fusion AI Factory In-a-Box is for building AI applications and integrating them into business operations. Fusion AI Inference In-a-Box is for running models in production. They can be deployed separately or together. Does the product use NVIDIA Blackwell hardware? 7news Morocco and a separate report say it integrates NVIDIA Blackwell chips, and that report adds NVIDIA AI Enterprise software. The specific Blackwell model and quantity per system have not been published. How much does Fusion AI In-a-Box cost? No confirmed price exists. The announcement coverage also omits performance figures, capacity, dimensions, power specifications and a commercial availability date. Does the system send data to the cloud? ABA Technology says it does not need to. The product is built for local operation so organizations keep control of their data, choose their models and manage their agents without transferring data to external cloud services. Who are the likely customers? The company targets public and private organizations, and coverage of its April launch and its Atos agreement points to sectors such as healthcare, finance, industry, public services and research. ABA Technology’s launch page also lists Industry 4.0, public services, digital twins, and research and education as target uses. Where can I read ABA Technology’s own description? The company’s launch page for Fusion AI describes the sovereign AI factory and the in-a-box concept, and industry coverage from Industries.ma and TechAfrica News adds context on the Berlin appearance and the Atos agreement. Elias Virtanen Cybersecurity Analyst Elias Virtanen is the Cybersecurity Analyst at Tech Insider, bringing hands-on expertise from his background in penetration testing and security consulting. He previously worked as a security researcher at F-Secure in Helsinki, where he focused on threat intelligence and vulnerability disclosure. Elias covers ransomware trends, zero-trust architecture, and the evolving regulatory landscape including NIS2 and the EU Cyber Resilience Act. He holds a CISSP certification and an MSc in Information Security from Aalto University. View all articles