As organizations explore new opportunities with AI, the industry is rising to meet the need for more data, better models and the infrastructure to support it. Companies often need external data as well as AI foundation models to help them build their AI solutions. But AI is challenging, and many projects don’t make it to the finish line. According to 451 Research, part of S&P Global Market Intelligence, on average organizations abandon over one-third of their AI/ML projects before they enter production, with IT infrastructure performance the main driver of project abandonment.[1] To change this, businesses are realizing that they need to collaborate with many partners for AI to succeed.
AI model marketplaces have become a powerful way for enterprises to find the external data and AI models they need to succeed with their AI initiatives. Organizations often lack the necessary datasets, data curation expertise and resources to train AI models themselves, and AI marketplaces are rising to fill that need. These marketplaces are a perfect place for all the players in an AI ecosystem to meet and exchange value. Also, as organizations look for ways to use their data for new models and services, the ability to maintain, store and access data will be more relevant than ever.
In our previous blog post, we introduced data and AI model marketplaces and discussed the types of marketplaces as well as how marketplace participants interact. Now, let’s examine the needs and priorities of the different marketplace participants.
What AI marketplace participants need
AI model and data providers
The organizations that deliver AI models typically use proprietary data and intellectual property to create new AI models and services. To achieve this, data and AI model providers need:
- To monetize their data and models by listing them in the marketplace
- The ability to directly share data with target customers privately, securely and in real time, without routing that data via the marketplace operator
- To maintain control and ensure their models and data aren’t used for unauthorized purposes
- To ensure compliance with data residency requirements
- To maintain flexibility to share data and models via multiple marketplaces
- To allow consumers flexibility to bring in their own data and customize their AI models—while ensuring they acknowledge the use of the provider’s data or model
AI model consumers
The consumers of AI models are typically enterprises and service providers that want to use external data and AI models to jumpstart their AI initiatives. In some instances, the consumers add value (via their proprietary data and models) and subsequently become data and model providers themselves. AI model consumers also have their list of requirements:
- A sandbox where they can bring their data and models and customize them using data and models from the marketplace
- To get updates when changes are made to data and models and newer versions are available
- To know the lineage of an AI model and ensure it doesn’t violate any regulations
- To ensure the quality of their data and models on a continuous basis
- To have privacy regarding why they’re using a given AI model
- The ability to test out a model and billing services
- The flexibility to pivot and change AI models as needed
- To access data from providers in real time
- Arbitration services to resolve conflicts when the quality of data deteriorates
AI marketplace operators
AI marketplace operators want AI infrastructure—both hardware and software—on which they can run their marketplace and give consumers the ability to train or finetune their AI models. They want:
- Compute, network, data, storage and security services—either from the infrastructure provider or created themselves
- AI marketplace software that provides catalog services, AI/ML orchestration and wallet/monetization services
- A legal governance framework, created independently or with help from industry vertical-specific consulting companies
- Support for two types of data transfer:
- Direct provider to consumer (B2B), where the data doesn’t transit via the marketplace: This is desirable when providers don’t want the data stored in third-party clouds because they want full auditability with respect to the number of data copies. The marketplace control plane provides security keys and a URL to the provider’s data location, and handles the payment process. The data transfer takes places directly between providers and consumers.
- Data transit via a marketplace staging area: This is desirable when data providers don’t want consumers to take the data out of the marketplace and potentially use it for unauthorized purposes.
AI marketplace deployment options and tradeoffs
Operators can host their AI marketplaces in the cloud, in private or local colocation data centers, or at an interconnection hub like Equinix. They can also get private AI infrastructure as a managed service from Equinix and use it as a key building block for their AI marketplaces. The following are the tradeoffs between hosting an AI marketplace in different locations:
Why AI marketplaces belong at Equinix
Equinix is helping organizations access the external data sources and models they need to see a strong ROI from their AI initiatives. We are the neutral place where everything in an AI ecosystem can come together—the hardware and software, networking, AI models, data sources and much more. Thousands of cloud, network and IT service providers are already exchanging value on Platform Equinix®.
If an AI marketplace is hosted at Equinix and data exchange happens here, there are no variable data egress charges, and consumers don’t have to pay for data requests (e.g., per read/write costs for object storage in the cloud). At Equinix, you have full transparency over the number of copies of your data, and you’ll never be locked into a single model provider. Instead, Equinix offers a cloud-neutral location where you can access AI models and services from different providers.
Hosting an AI marketplace at Equinix also provides more predictable infrastructure costs for marketplace operators, who can get AI infrastructure as a managed service. Operators can then focus on their governance model and operational aspects of the marketplace, rather than worrying about the infrastructure. Private AI offers AI marketplace operators, providers and consumers greater control over data and auditability.
Wherever organizations are with AI, they’ll likely need access to external datasets and AI models. Equinix is the best place to access AI-ready infrastructure or build your own private AI solution as you prepare for the future. Hosting AI marketplaces at Equinix allows operators to better address price, performance and privacy requirements.
Learn more about designing infrastructure for AI in The Equinix Indicator.
[1] 451 Research, Voice of the Enterprise: AI and Machine Learning, Infrastructure 2023.
Subscribe to the Equinix Blog