Over the past 12 months, we have seen a huge number of new AI organizations being developed leveraging the latest advances in underlying models, technology and demand. Although AI is often viewed as acting as a ‘co-pilot’ rather than an ‘autopilot’, there are still many amazing feats AI can achieve compared to classical computing. Recently, we saw a startup that can provide accurate text-to-sign language conversion, multilingual transcription, automated audio-video generation with realistic avatars, and more.
But like all startups and scale-ups, these new organizations face many challenges. Some are unique to the AI industry, others are common to all growing brands. But with the right level of support, founders can thrive and move the industry and humanity forward.
High computing power for training AI models
One of the main challenges AI organizations face is training. Training AI models requires a large amount of computational power, which can be challenging for deep tech companies that tend to operate on an operational cost basis rather than a capex basis. Deep learning algorithms such as neural networks require many iterations and adjustments to achieve optimal results. Without access to high performance computing resources, this can be time consuming and expensive. Additionally, this data has to be stored somewhere, which is prohibitively expensive to purchase outright and can be expensive to maintain.
Flexibility in resource allocation and cost control
The resource requirements for training and deploying AI models vary greatly depending on model complexity and dataset size. As with most start-ups, a company’s direction can change almost overnight, and it can be challenging for both people and technology infrastructure. As a result, most AI startups are cloud-native by default so they can pivot to new hardware when things start moving in a different direction.
Backward compatibility issue
AI frameworks such as TensorFlow and PyTorch are continuously updated and improved, but many iterations of these frameworks are not backward compatible with previous versions. This puts significant pressure on organizations to stay up-to-date with the latest frameworks, at the risk of functionality issues and downtime. Users often expect startups to have initial problems, but extended downtime can dramatically erode trust.
With these issues in mind, how are existing successful AI startups overcoming challenges?
AI in action: OVHcloud powers must-haves for Customs Bridge
Customs Bridge is a “deep tech” startup that uses artificial intelligence algorithms to create an automated product classification engine for European importers. The company’s mission is to create the most reliable product classification engine possible for assigning the correct customs code to products whose descriptions are not fully formalized.
However, Customs Bridge faced major challenges in training AI models. The on-premises infrastructure had limitations, had large data processing requirements, and needed a state-of-the-art AI framework. The existing infrastructure was not sufficient to effectively train and deploy AI models, and it was difficult to access and process the large amounts of data required to train the models.
To overcome these challenges, Customs Bridge turned to OVHcloud’s AI and machine learning solutions. The team implemented OVHcloud’s model training solution, AI Training, leveraging OVHcloud instances to deploy models into production and support data power pipelines. This allowed Customs Bridge to process massive amounts of data, enhance AI models, and improve overall productivity and efficiency.
Customs Bridge was able to leverage OVHcloud resources for data enrichment and advanced AI model training. They relied on about 2.5 TB of data to train their first Transformers model, but thanks to his NVIDIA V100 GPU provided by OVHcloud, the compute time it takes to train Transformers on 250,000 lines is It was only about 30 minutes. It’s fast and cheap, and Customs Bridge is now able to scale data volumes without limiting its infrastructure. The cloud-based approach gave the company the freedom to experiment until it found the amount needed to achieve the desired accuracy.
In addition to increasing the flexibility and scalability of AI model training, Customs Bridge enables cost-effective and efficient resource allocation, simplified AI framework implementation and deployment, innovation and experimentation for optimal results. It also benefited from the ability to By leveraging OVHcloud’s AI and machine learning solutions, Customs Bridge was able to overcome challenges and build an innovative and effective product classification engine.
Advancing deep technology with specialized cloud services
One of the first steps for any growing AI startup is to understand their ecosystem, not just in terms of understanding their competitors. There are many organizations that offer incubators, accelerators, and support schemes that can directly help with mentoring, administrative assistance, or in the case of the above example, support of the technology infrastructure.
Cloud services offer flexible resource allocation and cost control, allowing deep tech companies to change resources as their needs change. This adaptability ensures that businesses only pay for the resources they need, allocates resources more efficiently, and operates on operational costs rather than capital expenditures.
Scalable storage solutions are also an important part of the cloud service model. These solutions enable deep tech companies to process and store massive amounts of data to train AI models. These solutions are built to scale easily and allow AI companies to scale data volumes without service interruption, unlike physical storage where installing and managing new drives can pose many challenges. can be increased.
move the industry forward
Deep tech AI companies experience many of the same problems as startups in other industries, but they also present some unique challenges. For example, the huge datasets required to train AI models require correspondingly high-powered compute and storage capabilities that are out of reach for young organizations operating with seed funding. is often
This is why many AI companies are cloud-native by default. With the cloud, such organizations will be able to scale more easily without paying for infrastructure upfront, and benefit from managed her solutions that eliminate the need for day-to-day management by the founders and their teams. Needless to say. However, startups should be careful when setting up their cloud service agreements to avoid both spiral and hidden costs. Wrong setup or wrong provider (for example, overcharging inbound/outbound costs) can cause technical strain. But with the right partners, the right solutions, and a truly collaborative approach, startups can forget the administrative details and instead focus on their core mission of creating a new world of AI. increase.
