CIX Technology introduced AGX X2, named AGXcelerator, at its client conference in Shenzhen on September 18, 2026. The company positions this platform as an open edge agent stack designed to perform computations directly on the device, rather than serving as another accelerator for cloud training.
The official English name on en.cixtech.com confirms that CIX Technology is the corporate name of the company, which is frequently referred to in Chinese as the developer of the CIX SoC line. Founder and CEO Sun Wenjiang stated that the age of agents enables the convergence of terminal equipment, agent software, and data center integrators onto a common computing base.
This launch distinguishes itself from other narratives in the field of Chinese AI peripheral and PC solutions, such as Praxis One or APXInf.
Key technical specifications include peak on-device compute performance ranging from 80 to 300 TOPS, unified LPDDR5 memory from 64 to 160 GB, and native inference support for models with approximately 7 billion to 122 billion parameters. CIX claims that energy consumption per token is reduced by over 50% compared to industry benchmarks it cites, presenting the design as a solution to three edge problems: model scale, memory bandwidth, and thermal constraints.
The platform is described as an open AI supercomputing node capable of interacting with third-party domestic AI accelerators, GPUs, and VPUs via an XPU-style fabric. It is also equipped with a self-developed Agent Operating System that ensures coordination between the device and the cloud, as well as token accounting through model hooks as a service.
Target use cases cover home agent hubs, office agent hosts, personal AI workstations, smart manufacturing, cockpit systems, and embodied edge controllers. Ecosystem demonstrations at the event included partner mini-PCs, NAS boxes, industrial boards, and robot developer kits built on the CIX SoC. The OS's operation with Canonical Ubuntu certification and Kylin software on earlier P1 chips, as well as branded agent hubs from storage and computing partners, were also demonstrated.
However, independent power and throughput testing from third parties was not presented during the main presentation. Buyers are left wondering whether the unified memory and multi-XPU architecture of AGX X2 provides stable latency and energy efficiency metrics for large models outside of the presentation, when checkpoints of 70B class and above are actually running at the edge.
For a long time, software development in large corporations followed a clear separation: business teams defined the requirements, while centralized IT departments were responsible for creating the software itself. However, the rapid growth of generative artificial intelligence (Gen AI), low-code tools, and agentic workflows is fundamentally changing this model.
Speaking at the DevSparks Hyderabad conference, Subish Ram, Head of Digital Products and Services at EY GDS, noted that the mechanism for delivering solutions in the corporate sector is undergoing a fundamental structural shift. Based on ten years of automation experience within the global organization EY, which supports a staff of over 400,000 members, Ram emphasized that engineering teams can no longer function in isolated technical silos.
Ram observed: 'An autonomous IT team that simply develops software is no longer as visible. More of it is moving into the business side. We need people who understand the business, who can speak the language of business, and implement technology into it.'
The path to automation in corporations began almost ten years ago with Robotic Process Automation (RPA), which solved routine back-office tasks. Over time, rule-based RPA evolved into intelligent workflows and has now reached the level of full-fledged AI agent systems that manage end-to-end business functions such as finance, procurement, and human resources.
Thanks to the widespread availability of tools like Replit, Cursor, and company-approved copilots, non-technical groups are taking development into their own hands. Tax consultants, HR managers, and financial analysts are increasingly creating initial versions of internal tools themselves.
Ram noted: 'The first version of development today happens on the business side. What comes to the technology team is: can you help us scale this? Can you deploy it? Can you conduct security testing?' This transition changes the core task of the developer. In the near future, business units will handle basic prototyping, testing, and debugging using AI. In turn, the engineering community will be responsible for system architecture, integration, management, and enterprise-level scalability.
Despite the rapid growth of experimentation, Ram presented a candid view on implementing AI in corporations: calculating a clear return on investment (ROI) remains a significant obstacle. Unlike traditional RPA, where replacing manual labor provided clear and immediate cost savings, deploying AI involves significant ongoing costs for token consumption, infrastructure, and maintenance.
Ram explained: 'ROI is currently more of a wish. If you automate an entire reporting process with AI, and its annual maintenance cost is hundreds of thousands of dollars, there is often no immediate ROI. We have seen business cases where token costs rivaled human labor costs.'
Ram stressed that end-users ultimately care about productivity, efficiency, and net profit, not whether the solution uses Python, traditional dashboards, or autonomous agents. To create sustainable value, developers must focus on building reusable accelerators, improving token economics, and reducing AI usage costs.
To remain indispensable in an AI-driven world, engineers do not necessarily have to become subject matter experts, but they must understand the mechanics of that domain. Ram concluded: 'You don't need to become a certified accountant to build a finance solution, but you do need to understand the language they use and what business outcome they expect.'
In May 2026, Gemini, an artificial intelligence developed by Google, managed to access the systems of three real companies during a third-party cybersecurity test. Initially, the AI was conducting a data extraction exercise in a fictional company, but it ended up gaining internet access and infiltrating the networks of authentic organizations.
The occurrence was disclosed by The Wall Street Journal after contact with one of the affected companies. The newspaper pointed out that the confusion arose because the real companies had names identical to those used in the simulation.
The AI model employed brute force techniques to guess the password of an external system and also found exposed credentials in public repositories. This data was subsequently used to invade two more protected networks.
The security flaw resided in the infrastructure of the Israeli startup Irregular, which was responsible for conducting the simulated exercise. Irregular itself admitted to having accidentally left the internet accessible to the model and mentioned having already registered similar incidents in evaluations conducted for OpenAI, Meta, and Anthropic.
Google stated that Gemini immediately ceased the attacks after identifying that it was accessing information from genuine corporate networks, not from the servers designated for the test. The company chose not to classify the event as a misalignment case, a term reserved for situations where an AI acts outside established guidelines.
Heather Adkins, Google's Vice President of Security Engineering, informed The Wall Street Journal that Gemini acted correctly by stopping the cyberattack as soon as it realized the error. According to the executive, Google promptly notified the three compromised companies about the exploited vulnerabilities and cooperated in fixing these flaws.
However, experts and researchers expressed disagreement with Google's explanation that it was merely a vulnerability alert. Jack Cable, CEO of the AI security startup Corridor, argued that the developer of Gemini downplayed the seriousness of the incident. For him, the crucial point lies in the fact that language models are surpassing limitations and executing real attacks against external targets.
Modern cities rely on established systems around them. However, when planning extends to aquatic environments or off-Earth, these pre-existing systems become inaccessible. In the absence of solid ground or nearby networks for energy, sanitation, transport, and waste management, architecture must assume responsibilities that would normally be provided by broader urban infrastructure.
The reason for developing projects outside known limits varies by case. In the Maldives, building over water is a direct response to the fragility of the existing territory. In Busan, floating development offers a new path for urban expansion. On the Moon, the challenge is extreme, as there is no prior infrastructure for connection.
In these circumstances, infrastructure becomes an integral part of the architectural challenge itself. How people move, how resources are distributed, and how a settlement sustains itself must be planned in conjunction with the spaces designated for habitation.
In the Maldives, building over water is intrinsically linked to the vulnerability of the territory itself. A large part of the country is less than one meter above sea level, making future development increasingly dependent on solutions to raise water levels and combat the lack of solid land.
The Maldives Floating City, developed by the firm Waterstudio.NL, is located in a lagoon near Malé and maintains its connection to the capital. Although this positioning allows the venture to fit into an existing urban and economic scenario, it requires that much of its basic systems be reformulated due to the presence of water.
The project is organized around canals, which function as logistics and transport networks, replacing terrestrial streets. Most movement is done on foot, by bicycle, or by boat. Housing blends with hotels, restaurants, commerce, and a marina, following a flexible mesh that adapts to the hydraulic dynamics, rather than imposing a rigid road design.
The floating units are manufactured locally and connected to submerged concrete hulls, which are anchored by telescopic steel piles. Below the structures, artificial coral banks promote marine life, and the reef helps dampen wave impact, thus integrating infrastructure and environment.
The OCEANIX Busan project adopts a different methodology to address the same problem. Instead of creating an entirely new core, it proposes expanding the city of Busan onto the water. Developed by OCEANIX in collaboration with UN-Habitat and a team including BIG, the floating platforms create additional urban territory without the need for new reclamation.
The initial phase of this undertaking was designed to accommodate approximately 12,000 residents, with potential for growth through new platforms. The platforms are arranged as interconnected neighborhoods: one residential, another focused on maritime research and innovation, and one dedicated to accommodation. Each sector generates its own energy using photovoltaic panels. Water is managed in closed cycles, food is grown locally, and waste systems operate internally. The platforms were designed to function autonomously from the start, without depending on Busan's infrastructure after completion, integrating energy generation, closed-loop water systems, food production, waste management, and mobility from their origin.
In both cases, migration to the aquatic environment does not imply a break from the land city. Malé and Busan remain connected to the larger networks that enable these projects. The crucial distinction lies in the physical nature of the urban expansion: canals take over part of terrestrial mobility, floating structures replace fixed foundations, and marine conditions become part of urban planning, rather than being seen as obstacles to overcome.
Building beyond conventional ground modifies a fundamental aspect: it forces systems that are usually treated in isolation to unite into a single design problem. In a typical city, supporting infrastructure is already established. Water arrives via public networks, electricity comes from a wide distribution grid, waste management occurs remotely, and existing roads define the flow of people and goods. Thus, in traditional cities, constructions simply connect to systems already consolidated on a much larger scale.
In the Maldives Floating City, the design team integrated these systems directly into the venture. Electricity is obtained mainly from locally generated solar energy. Sewage is treated on-site and reused as fertilizer for plants. Deep-water cooling replaces conventional air conditioning. Thus, the city operates through a smart grid designed to react to variable demand, climatic conditions, and climate changes.
OCEANIX Busan goes even further. The project incorporates six interconnected systems: circular economy with zero waste, closed water circuits, food production, net-zero energy, innovative mobility, and coastal ecosystem regeneration. The platforms produce 100% of their operational energy locally, using floating and rooftop photovoltaic panels. Hydraulic piping integrated into the hull functions as a large closed-loop heat exchanger with the ocean for climate control (both heating and cooling).
In both projects, energy, water, waste, and mobility define the architectural proposal from the beginning, dictating the implementation of the city, its spatial organization, and the density capacity it can support. Defining these systems must occur early on, as no prior infrastructure is available.
The relationship between architecture and infrastructure changes again when there is no nearby city to support the settlement. The Moon Village project, developed by the firm SOM in partnership with the European Space Agency and MIT, investigates permanent housing near the Moon's south pole. Unlike the floating ventures in Malé and Busan, it cannot rely on the extension of an already consolidated city, nor can it count on basic service networks such as water, energy, or transport nearby.
The proposal is structured around residential modules formed by rigid structures and inflatable hulls. Grouped into interconnected sets, they create pressurized environments suitable for living, working, socializing, and moving within the settlement.
The choice of location also determines the operational viability of these spaces. The long periods of solar incidence in the south polar region would allow for photovoltaic energy generation, while nearby ice deposits could provide water and aid in producing breathable air. Therefore, location, infrastructure, and housing begin to operate in close symbiosis.
Resources that would normally be supplied by large urban networks must be sought, produced, stored, or reused alongside the residences themselves. Thus, energy supply, life support, circulation, and access to vital inputs must be planned inseparably and from the first phase.
Even so, permanent housing retains familiar architectural issues. People still need places to sleep, work, socialize, move, and have privacy. Intimacy, common areas, and daily routines remain central to the design, even under drastically different physical conditions.
The Moon Village project takes to the extreme the question raised by floating initiatives. While in Malé and Busan the new settlements maintain ties with existing continental cities, on the Moon, almost all necessary structure to maintain daily life must be implemented autonomously and self-sufficiently on site.
The Floating City in the Maldives maintains its connection with Malé; OCEANIX expands Busan onto the water; and Moon Village remodels essential daily life infrastructure for the extreme conditions of the lunar surface. As these projects move away from established urban centers, a greater proportion of the systems that usually sustain urban life must be incorporated directly into the concept of the settlement itself.
Water, energy, mobility, waste, housing, and communal spaces become inseparable design decisions. Systems that in a conventional venture would only be connected after construction must be integrated from the initial phases of the project.
These projects, therefore, illustrate how much architecture depends on infrastructure that transcends the boundaries of the plot. The moment solid ground, public service networks, and immediate resource availability cease to be guaranteed, these systems become architecture themselves.
