Despite the fast pace in artificial intelligence development, organizations are falling behind in their ability to handle AI demands, according to Cisco’s 2024 AI Readiness report released last week.
Only 13% of organizations are “fully ready” to realize the possibilities of AI, a decrease from 14% in 2023, the report revealed. In addition, only 21% of organizations have the required GPUs to satisfy current and future AI demands.
Despite the small number of companies reporting that they are prepared for AI, 98% said the urgency to use AI-powered technologies rose in their organizations during the previous 12 months.
Cisco worked with an independent third party that interviewed 7,985 senior business leaders tasked with AI integration. Organizations had 500 or more employees and covered 30 markets around the world.
The State of Enterprise AI Infrastructure
The AI Readiness report grouped organizations according to four criteria:
1) Pacesetters, who are leading AI adoption and integration. Only 1 in 7 organizations considered themselves pacesetters, according to Cisco.
2) Chasers are those workers in organizations who have above-average AI readiness and are progressing well in AI adoption.
3) Followers, which are organizations that have momentum in adopting AI but are below average in preparing to use it.
4) Laggards, the companies that are least prepared to adopt AI.
High Urgency to Satisfy AI Workload Demands
Companies are experiencing a real urgency from their leadership to make the most of AI within the next 18 months. Close to 85% of companies surveyed believed they only have 18 months to begin showing how AI is impacting their business. Meanwhile, close to half, or 59%, give themselves only 12 months to show AI’s impact on their organizations.
Companies expect to deliver returns from new revenue streams coming from AI, explains Mark Patterson, chief strategy officer for Cisco.
“This urgent time crunch on AI is because companies expect AI projects to open up new revenue streams and increase profitability,” Patterson tells Network Computing via email.
He adds, “Given the rapid market evolution and the significant impact AI is anticipated to have on businesses, this gap between urgency and ability is especially startling.”
Inadequate compute, data center network performance, and cybersecurity infrastructure contributed to the biggest decline in infrastructure readiness. Organizations also lack the power consumption required for AI workloads, according to Cisco.
In fact, 30% of organizations lacked the ability to protect data in AI models. To do so, Cisco reported that they will need end-to-end encryption, security audits, continuous monitoring, and instant threat response.
Companies are also behind in addressing the quality of data for AI. Of the companies surveyed, 80% reported inconsistencies or shortcomings in the preprocessing of data as well as data cleaning for AI projects, which is in line with the 81% rate from 2023.
AI Infrastructure Readiness Requires Training Talent
To build more AI infrastructure readiness, skilled talent will be key to overcoming a deficit in workers needed to maintain IT infrastructure, Patterson suggests.
In fact, only 31% of companies believed their talent was in a “high state of readiness” to fully make use of AI. In addition, 24% of those surveyed did not believe their companies held enough talent to address the “growing demand for AI,” the Cisco report revealed.
Expanding the AI talent pool will require forming a learning culture for innovation, he says. That includes talent development and forming clear career paths.
Leadership feels the pressure to achieve AI readiness, but workers are hesitant to use AI, according to the Cisco AI readiness report.
“While organizations face pressure from leadership to bring in AI, the disconnect is likely due to hesitancy among workers within the organization who must take steps to gain new skills for AI or fear AI taking over their jobs,” Patterson says. “To remedy this, companies should encourage AI adoption across departments, incentivize innovation, and recognize and reward successful AI initiatives.”
Other Studies Find Similar AI Infrastructure Issues
A Nov. 21 report released by Capital One Financial Corp. revealed a similar disconnect. Only 36% of tech practitioners and 47% of business leaders believed their companies had the skills and expertise required to carry out complex AI projects. Although 87% of business leaders were confident in their organization’s ability to carry out and deploy AI, 70% of technical practitioners were struggling with data problems by spending up to four hours per day fixing data problems, conducting quality checks, and rectifying errors. These issues with data management hold back AI success.
Generative AI brings resource-intensive demands on organizations, notes a recent report by the Al-Enabled ICT Workforce Consortium, led by Cisco along with Accenture, Eightfold, Google, IBM, Indeed, Intel, Microsoft, and SAP.
Skills and Training Critical for AI Success
The consortium report looked at the role of AI on information and communication technology jobs. It revealed that 92 percent of jobs will undergo high or moderate transformation as a result of AI advancements.
“Across the Consortium member companies, we have made it our collective responsibility to train and upskill 95 million people over the next 10 years,” Francine Katsoudas, chief people, policy & purpose officer for Cisco and founding member of the AI-Enabled ICT Workforce Consortium, said in a statement. “By investing in a long-term road map for an inclusive workforce, we can help everyone participate and thrive in the era of AI.”
Key skills that will become more essential include AI ethics, responsible AI, prompt engineering, AI literacy, and large language models (LLM) infrastructure, the consortium report said.
Patterson said industry reports like that of the consortium help provide the information organizations need to train workers and bridge the AI skills gap.
“While many organizations may not have a fully defined AI strategy or use case yet, scalability, simplicity, and security will be critical as they take their next steps,” Patterson says.
This training also includes education on cybersecurity risks around AI and how to adapt infrastructure to work with this new technology, Patterson says.
“If you can’t secure AI, you won’t be able to successfully deploy AI,” he says.
Meanwhile, tech professionals should develop a holistic view of the infrastructure required to adopt AI while incorporating observability and security, according to Patterson.
A holistic view of infrastructure will bring “easier operations, resiliency, and efficiency at scale,” Patterson says.
They should also learn how to customize or “right-size” data center infrastructure for specific use cases and outcomes, Patterson says. That includes dense compute nodes for high-density GPU AI workloads such as model training or less GPU-intensive inferencing workloads.
The Cisco report revealed that 79% of respondents believe they need more data center GPUs for future AI workloads, an increase from 76% in 2023.
Training in AI workloads can also include network planning exercises. A report by cloud-based network management software company Auvik Networks found that only 50% of respondents said their company participated in network planning.
Infrastructure knowledge also involves learning about the right interconnects required for AI applications. Interconnects include PCIe 5.0, CXL 2.0, or hybrid switches.
To become prepared for AI, organizations should consider which AI use cases will provide the best ROI, how to optimize the use of expensive GPU clusters, and whether AI apps are secure from threats and leaks, Patterson says.
“Addressing the talent shortage for AI infrastructure will be an industry-wide effort, but individual organizations can take action to reskill their workforces,” he says.
