Nearly 80% of companies are expecting to increase their overall AI spending in the next fiscal year, with Generative AI (GenAI) expanding its share of overall AI budgets, according to the latest Deloitte US State of Generative AI in the Enterprise survey of C-suite executives. That number is a testament to how clearly organizations see the value of AI and GenAI and understand its benefits.

From enhanced decision-making with advanced AI analytics to improved operational efficiency through AI-driven automation and optimization, these exciting technologies offer the potential to take businesses to new levels of productivity and efficiency. Leveraging AI can even bring faster innovation by improving or developing new products and services as well as drawing on data to drive business insights.

For nearly three-quarters of survey respondents, value from AI and GenAI is being achieved, with their most advanced initiatives meeting or exceeding ROI expectations. Furthermore, businesses have been improving their level of preparedness in the critical area of technology infrastructure, according to the survey, in anticipation of scaling their GenAI initiatives.

To keep momentum going, itโ€™s important for organizations to understand the practical steps they will need to take to accelerate the deployment of AI and GenAI across the enterprise. By ensuring that their infrastructure preparedness extends to computational capacity and data, organizations can lay the foundation for business innovation and the opportunities it can bring.

Laying the foundation

To scale GenAIโ€”and get the most valueโ€”the investment in infrastructure and necessary computing power will likely be critical. In fact, building, configuring, and managing scalable, robust infrastructure is one of the best ways to achieve a return on investment. Upgrading architectures can support the tremendous speed, scale, and agility needed for modern AI applicationsโ€”with workflows that often need to juggle billions, even trillions of parameters to train models.

To get AI projects off the ground, organizations should thoroughly assess what graphics processing unit infrastructureโ€”the specialized processors that handle visual and mathematical calculationsโ€”may be needed to support AI models and how to tailor it for AI workloads. One option is working with a third-party cloud platform provider that can supply the infrastructure resources needed without the upfront investment.

Optimal utilization of computational resources can also be critical to helping ensure AI workloads are processed swiftly and cost-effectively. With some AI models using 100 times more computer power than just a few years ago, tech companies are starting to optimize large language models (LLMs) for more efficient use of power and data. With scalable and reliable cloud resources, optimized resource management, and high-performance data storage, businesses can see an acceleration of AI workload performance and data retrieval times.

When it comes to data, organizations should also enhance and evolve their ability to identify and leverage AI-ready data as well as their data-handling practices. That is, how they use internal data to train accurate and effective AI models. Organizations may even want to investigate if third-party data will be needed to achieve their AI goals. These efforts should include data security, auditability, and governanceโ€”all essential for enterprise AI deployments. Some government entities and enterprises are also increasingly pushing for Sovereign AIโ€”to mitigate risks related to surveillance, cyber threats, and dependence on non-domestic technologiesโ€”via increased investment in developing local AI models, building national AI infrastructure, and funding ethical AI initiatives.

Key considerations to scale

To achieve the agility, performance, and speed needed for long-term AI success, organizations should also consider the following even wider-ranging factors:

  • Compute: As noted above, GPUs are designed to process multiple computations simultaneously and can be clustered for massive scale. Some organizations are looking at ways to optimize utilization given the fact GPUs can be underutilized during peak times.
  • Networking: AI workloads must be robust to transfer large datasets and to connect compute resources efficiently. This is important to diminish latency that may be magnified by the scale of AI computations and data.
  • AI data & storage platform: An AI data & storage platform combines AI specialized physical storage with a software platform that drives AI performance scaled to support the velocity, volume, and variety of data that GenAI processes are relying on.
  • Skill sets: Organizations should look to upskill employees and hire talent with key skills to gain competitive advantage.
  • Total cost of ownership: The cost of a fully scaled AI infrastructure is significant, varies greatly between archetypes (e.g., public cloud, on premises, private cloud providers), and is evolving rapidly. Another possibility, and perhaps the most effective, is a hybrid AI infrastructureโ€”that is, different archetypes combined for both technical and cost performance.

The AI revolution is here

AI and GenAI are here to stayโ€”and organizations should evaluate their infrastructure preparedness to best capitalize on these fast-moving technologies. Whether itโ€™s building computing capacity or upskilling staff, organizations that clearly understand what it takes to effectively support and scale AI will be in a much stronger position as they continue on their AI journey.

To learn more about AI infrastructure readiness and how Deloitte is providing comprehensive solutions, visit the Silicon 2 Service AI Factory pages on Deloitte.com