1000x analytics cost efficiency gain
Analyzed 5B rows of data in 10 minutes
Built 222 AI agents in only 90 minutes
Ran 10 LoRA layers on a single GPU
25% multi-step agent performance boost
Aible rapidly deploys trusted generative AI agents aligned with core business KPIs. Leveraging NVIDIA's airgapped hardware and co-optimized GPU stacks with Google Cloud's serverless BigQuery and Cloud Run, Aible delivers secure, scalable AI impact within 30 days.
Aible rapidly deploys trusted generative AI agents aligned with core business KPIs. Leveraging NVIDIA's airgapped hardware and co-optimized GPU stacks with Google Cloud's serverless BigQuery and Cloud Run, Aible delivers secure, scalable AI impact within 30 days.
Putting business users first to solve Enterprise AI's toughest challenges
Aible operates with a disruptive mission: empower every business user to use artificial intelligence directly rather than leaving it solely in the hands of select teams. Founder and CEO Arijit Sengupta, author of AI is a Waste of Money, observed that the primary reason enterprise AI initiatives fail is that technical builders frequently do not understand the practical objectives of the business. Instead of focusing on technical metrics like token utilization, Aible focuses strictly on moving business KPIs, such as sales revenue, customer satisfaction, and logistics costs.
It’s well-known that enterprises looking to roll out AI face steep challenges, notably around data privacy and model hallucinations. A major blocker is a classic "chicken-and-egg" prototyping dilemma. Corporate stakeholders demand rigorous clarity regarding data security, output consistency, and token expenses before granting access to sensitive company files or dedicating budget. Yet, data teams cannot accurately guarantee these profiles without first testing models against the enterprise's unique data sets. Navigating the extensive corporate bureaucracy required to stand up isolated cloud projects for unvetted prototypes routinely stalls innovation for months.
"The most common reason why AI projects fail is that well-meaning experts have no idea what the business is trying to do," says Sengupta. "A business user doesn’t really care how powerful an AI is; they care about consistency and measurable impact on their business KPIs."
By using Google Cloud’s serverless architecture with NVIDIA Blackwell GPUs, we demonstrated a 1000x improvement in analytics efficiency. We can evaluate millions of questions across billions of rows of data in minutes, saving our enterprise customers immense costs while natively keeping their data secure and under their own control.
Arijit Sengupta
Founder and CEO, Aible
Additionally, expert-driven AI often falls prey to the "paperclip problem," where a model blindly optimizes a narrow technical goal at the expense of macro business health. For example, an inventory-management agent told to minimize product stockouts might continuously expedite freight shipments, running up catastrophic logistics costs that completely erase company profit margins. To safely and meaningfully scale AI across a global workforce, enterprises require a highly secure, consistent solution that bypasses bureaucratic logjams and focuses on delivering business impact.
Aible chose to work with Google Cloud and NVIDIA to address these issues and deliver a better path forward for enterprise AI.
Designing a serverless, multi-GPU architecture for secure and scalable prototyping of Enterprise AI agents
To break the prototyping gridlock, Aible collaborated with Google Cloud and NVIDIA to architect a frictionless, dual-phase deployment framework. The journey starts locally on an airgapped NVIDIA DGX Spark desktop agent computer powered by the NVIDIA GB10 Grace Blackwell Superchip. Because the hardware runs disconnected from the internet, enterprise teams can securely ingest sensitive data to evaluate safety, prune dead-end responses, and precisely profile token behavior without delay or data exposure risks.
Once verified locally, users only need to copy and paste a single JSON file to move the optimized, deterministic agent to production on NVIDIA Blackwell GPUs in Google Cloud. Once in production, Aible operates inside a secure-by-default architecture using Google Cloud Run templates.
Crucially, Aible respects data gravity; enterprise data remains safely stored inside Google BigQuery, and all analytical processing executes natively within the customer's cloud boundary without information ever leaving.
The alignment between NVIDIA and Google Cloud offers an incredible prototyping-to-production pipeline. Organizations can safely build and test autonomous agents in a secure, airgapped NVIDIA DGX Spark desktop environment, then push that proven agent directly to Google Cloud to immediately deliver at massive scale with total architectural consistency.
Arijit Sengupta
Founder and CEO, Aible
Aible’s infrastructure maximizes the co-optimized hardware and software stacks of Google Cloud and NVIDIA. Cloud Run dynamically handles serverless agent orchestration, memory, and tool calling, while BigQuery serves as a serverless data engine. To eliminate hallucinations, Aible applies a deterministic validation layer called "if it's blue, it's true," which automatically double-checks generative outputs back to source documents.
"Our agent orchestration runs serverless on Cloud Run with NVIDIA RTX PRO 6000 Blackwell GPUs, and data processing happens inside BigQuery," says Sengupta. "This serverless-first architecture means you can build thousands of operational agents without creating a management nightmare, because agents cost absolutely nothing when they aren't actively running."
Aible leverages NVIDIA NeMo suite of libraries—particularly NeMo Customizer and NeMo Evaluator to optimize model performance with automated, live post-training cycles, appending lightweight LoRA layers to adapt open weight reasoning models like Nemotron 3 to an enterprise's unique terminology and processes. Remarkably, up to 10 of these fine-tuned LoRA variants run concurrently on a single NVIDIA GPU, using NVIDIA Multi-Instance GPU (MIG), without diminishing cost-performance efficiency.
For multi-step agentic workloads, Aible utilizes Google Cloud's A4X and A4X Max instances powered by NVIDIA GB200 NVL72 and NVIDIA GB300 NVL72 clusters to ensure high-throughout, low-latency execution of complex agent flows. By sharing memory directly across the CPU and GPU in NVIDIA Grace Blackwell-based systems, this architecture removes data-copying delays to significantly accelerate multi-step agent execution.
"By using Google Cloud’s serverless architecture, we demonstrated a 1000x improvement in analytics efficiency," says Sengupta. "We can evaluate millions of questions across billions of rows of data in minutes, saving our enterprise customers immense costs while natively keeping their data secure and under their own control."
Achieving 1,000x analytics efficiency and accelerating time-to-value in days
The combination of Google Cloud and NVIDIA has yielded tremendous operational velocity and cost efficiencies for Aible's enterprise clients. Production environments are live in just 15 minutes, and corporate users routinely extract measurable business outcomes within 5 to 15 minutes with the help of over 100 pre-built industry templates.
"The alignment between NVIDIA and Google Cloud offers an incredible prototyping-to-production pipeline," says Sengupta. "Organizations can safely build and test autonomous agents in a secure, airgapped NVIDIA DGX Spark environment, then push that proven framework directly to massive scale with NVIDIA Blackwell GPUs on Google Cloud with total architectural consistency."
The technical scaling delivered using BigQuery is immense. Aible successfully processed 5 billion rows of data and evaluated 10 million variable combinations in less than 10 minutes. Because BigQuery bills serverless compute based strictly on the exact window of execution, running millions of automated, AI-initiated queries costs a mere $10. Joint white papers co-authored with Google experts highlight that this serverless design delivers an unmatched 1000x improvement in analytics efficiency and cost reduction.
We co-innovate deeply with our partners to maximize hardware efficiency; we have successfully run up to 10 fine-tuned model variants on a single GPU simultaneously. In addition, using advanced configurations like NVIDIA Grace Blackwell on Google Cloud delivers a 25% performance boost for multi-step agent workflows.
Arijit Sengupta
Founder and CEO, Aible
Real-world deployments demonstrate the ease of adoption. The State of Nebraska deployed a team of 36 people—including interns—who successfully built 222 data-driven agents in just 90 minutes on Google Cloud. Another major enterprise client stood up 450 distinct operational agents within its first two months of deployment.
"We co-innovate deeply with our partners to maximize hardware efficiency; we have successfully run up to 10 fine-tuned model variants on a single GPU simultaneously," says Sengupta. "In addition, deploying on NVIDIA Grace Blackwell-based systems on Google Cloud delivers a 25% performance boost for multi-step agent workflows."
Aible’s migration to Cloud Run and G4 VMs with NVIDIA RTX PRO 6000 Blackwell took less than 2 hours and now positions it to better support long-running agents through easy model deployments. This also allows Aible to run models like the Nemotron 3 Super in Google Cloud for long-running agent planning tasks, so those agents can scale up and down on demand without complex infrastructure management or idle GPU costs
Looking ahead, Aible is in the midst of a major shift away from simple chat interfaces toward long-running, autonomous agentic workflows. These next-generation agents continuously ingest diverse inputs—ranging from warehouse inventory logs to physical loading-dock camera feeds—to automatically balance complex trade-offs like product expedite costs versus customer stockout losses. By swapping narrow technical token output for comprehensive business-impact tracking, Aible and its partners ensure enterprise AI remains securely aligned with corporate reality.
"Go prototype a bunch of agents quickly on real data to trace the economic impact on your KPIs, then click a button to scale on Google Cloud," says Sengupta. "If an agent doesn’t get used, it won’t cost you any compute. In a serverless world, there is an opportunity cost to not trying."
Aible is an enterprise AI platform dedicated to empowering business users to directly build and leverage artificial intelligence to optimize operational KPIs like revenue, costs, and customer satisfaction. Founded on the principle of business-driven AI consistency and transparency, the company eliminates hallucinations using deterministic data verification layers.
Industry: Technology
Location: United States
Products: BigQuery, Cloud Run, Google Kubernetes Engine (GKE), G4 GPUs
About Google Cloud partner – NVIDIA
NVIDIA is a global leader in accelerated computing, specializing in the co-optimization of advanced hardware and software stacks to power high-efficiency artificial intelligence workloads.
