AI-based applications are now quite easy to develop with the tools available on the market. MaaS (Model-as-a-Service) is a way that provides pre-trained AI Foundation Models as a service to power AI applications for a client.
Most developers want to avoid training, hosting, and managing AI models to power their applications, as difficulties in deployment may lead to the abandonment of an application idea.
Also read: IBM Open Sources Granite Code Models, Revolutionizing Software Development
MaaS makes it easy to leverage the already-trained machine learning models for application requirements without the costs involved with model development. It allows startups to focus on their core applications rather than developing AI models independently.
Lower Overhead Costs Are Beneficial for Users
Training and deploying AI models is a different ball game than selecting between the foundation models like Microsoft Phi-3, Google’s Gemini, OpenAI’s ChatGPT, Meta’s Llama 3, or other models available. It’s different from IaaS (infrastructure-as-a-service), which Amazon and HuggingFace offer.
You can compare MaaS to cloud services, with the difference that it provides your products with complex machine learning abilities rather than simple cloud services.
It saves overhead, human resources, and capital associated with training AI models, which are not affordable for all businesses.
The principal program manager of the Microsoft AI platform, Seth Juarez, says,
“If you’ve ever tried to deploy a model, there’s a series of combinations of incantations, Pytorch versions, and CPU and GPU stuff.”
Juarez, while talking to VentureBeat, explained that MaaS removes all those factors. If you want to use an open-source AI model or an API-based model like OpenAI, it can now be accessed through an endpoint.
MaaS Provides Business Opportunity
MaaS offers big tech companies like Microsoft, Alibaba Cloud, and Amazon an opportunity to tap corporate business, which makes it attractive for them.
Microsoft introduced its MaaS program in 2023. Initially, it offered only Meta’s Llama 2 and Mistralv7B, but recently it added more to its pile. Now it has also added Cor42 JAIS and Nixtila’s TimeGen 1, while it says that stability AI and Cohere will also be available soon.
Also read: AI Foundation Models Market Raises Concerns in the UK
Alibaba Cloud, the technology backbone of Alibaba Group Holdings, also offers MaaS services through its ModelScope platform for several open-source models. It was launched a year ago and has 300 ready-to-use artificial intelligence models for developers and researchers.
ModelScope’s offerings cover an entire array of AI models, from NLP (natural language processing) to computer vision.
Tencent, the world’s largest video game publisher, also launched a MaaS platform last year. Its CEO and Senior Executive Vice President, Dowson Tong, who, in his speech last year at the World Artificial Intelligence Conference in Shanghai, said,
“By leveraging these capability models, partners can easily create their own exclusive models by adding their own unique scene data. Through privatized deployment, authority control, and data encryption, we prioritize data protection for enterprise users, ensuring secure and reliable model usage.”
Tencent Cloud MaaS offers pre-trained AI models for a wide range of different business sectors, including media, finance, healthcare, and education.
Rent It and Let the Company Do the Maintenance
The models included in Maas catalogs are either a result of a partnership with the companies or an API structure has been built for the models to make them function uniformly. Currently, the models available adhere to these limitations, and more complex models are unavailable as they require a more complex way to be deployed. Juarez said,
“That’s why you see some enabled as Models-as-a-Service and others you see you can push into your own container and run in what we call managed inference.”
Tencent said it created a robot customer service model for an online travel company by refining the model, leveraging industry-standard large models, and incorporating enterprise data. The service provides customers with comprehensive vacation planning services to enhance their experience. We cannot confirm Tencent’s accuracy and customer satisfaction claims because the customer’s name was not disclosed. However, such services can help derive sales and improve revenues.
Juarez says that in the future, developers will select AI models on a rent-or-ownership basis, as people do for homes and other properties.
“Basically, you own the container, the model, and Azure ML, and you’re paying the rent and doing the upkeep, so to speak.”
He explained that the company does the maintenance on the Models as a Service setup, and users can rent the model for their requirements, but, he said, some users may prefer their private virtual network to use these models.
Nvidia, the GPU maker, also launched some large-language AI models as a cloud service as part of the Nvidia AI Foundation. Its CEO, Jensen Huang, said that AI language modeling as a Service is “potentially one of the largest software opportunities ever.”
Cryptopolitan reporting by Aamir Sheikh
