Industry access to AI computing infrastructure and services Is access to AI computing infrastructure and services a key issue for organisations developing, adopting or researching AI? We spoke to 200 AI practitioners across Australia to find out. Most important factors when Difficulty obtaining AI services?* progressing AI projects Recruiting or partnering for the right skills 92%8% Only 31% of respondentsnoted any difficulty in Access to quantity, quality dataNo difficulty59% Yes, experiencingdifficulty31% Other10% obtaining AIservices, with the top cause being AI compute with appropriate 83%17% Access to adequate AI compute infrastructure 77%23% privacy and security controls. Other factors causing difficulty were sovereignty of Organisational and customer data and AI literacyAI, data and infrastructure, and regulatory compliance. 65%35% *‘AI services’ describes end-to-end services, cloud platforms, pre-configured or pre-trained AI modules, and Platform as a service (PaaS). Very important/important/ Minor/no importancemoderately important What concerns AIpractitioners?95%Data privacy94%Ethicsin AI92%Environmental impactof AI infrastructure Location of organisation’s AI computing Global cloud 72% service provider Data science workstations 54%and/or laptops Data centre capacity owned and 39%operated by own organisation Components of an AI computing stack 53%CPUs33%GPUs23%High density servers13%High performance computing interconnects Other accelerator types Data centre capacity Only 7% of respondents have processes in place to measure how much AI computing theorganisation hasand/or uses. 5% 34%leased from third-party16%Other Local cloud 34%service provider27%None AI computing or services future challenges Top issues faced when Future consumption seeking to grow AI adoptionExpected AI computing consumptionwill rise by an average of 57% in the next 12 months, driven by expected changes to AI R&D related activities (60%), followed by an expected change in the AI maturity within the respondent’s organisation (59%), and changes in demand from customers and partners (53%). 70% 60% Over half of the respondents expect Lack of skilled Insufficientgeneral computing costs to rise over persons within understanding of AI the next 12 months (61%), as with AI this businessacross the economyservice costs (57% of respondents). Overall, privacy and security are the primary anticipated challenges when obtaining AI computing or services over the next 12 months. AI compute with appropriate security AI compute with appropriate privacy AI computeaffordability AI compute that is located in Australia Access to AI services e.g. PaaS Latency of AI compute AI compute minimum commitments Very important/important/ moderately important Opportunities Overall, machine learning was reported to be the AI field with the greatest opportunities. Younger organisations (in operation for less than 5 years) tended to note natural language processing as a significant opportunity. 78%Machine learning 38%Robotics 35%Computer vision Knowledge34%representation Natural language 31%processing Automated26%planning Expert 25%systems Other 4% 76%24% However for AI developers, AI compute that is located in Australia is of more concern, with75% of AI developer respondents perceivingthis to be an important futurechallenge. AI and data literacy Insufficient understanding of AI across the economy is the second most important challengeto those seeking to grow AI in Australia. Low AI literacy among customers is a hurdle for 86% of organisations developing, adopting or researching AI. Only12% of respondents reported a strongor very strong understanding ofAIwithin their organisation. There is a tendency for organisations with200–500 employees or those in business for more than 10 years to have more pronouncedconcerns about AI literacy levels. 72%28% 72%28% 68%34% 61%39% 60%40% 51%41% Minor/no importance The National AI Centre and CSIRO’s IM&T commissioned Aalto to investigate the availability of AI computing and AI services in Australia. Findings are drawn from a review of relevant published materials and direct engagement with industry via 10 x 1:1 interviews, and an online survey of 200 AI practitioners in Australia across the AI market including AI adopters and AI service providers. B&M | 22-00724