The SuperAI Singapore conference, held June 18-19, 2025 at Marina Bay Sands Convention & Exhibition Centre, showcased the rapidly evolving landscape of enterprise AI adoption. The event brought together established tech giants, innovative startups, and specialized solution providers, painting a picture of an industry in transition.
Not since the “Intel Inside” campaign of the early 1990s, or Cisco’s “Empowering the Internet Generation” of the DotCom era, have I seen the industry so excited about hardware.
In my early career I worked with 8-bit microcontrollers, having to squeeze as much functionality as possible into only 1k (1,024 bytes) of program memory and only 128 bytes of RAM. But efficient coding and storage use became less important in a world of huge increases in capacity. It’s very rare now to see a memory card less than 8GB (8,000,000,000 bytes) in size, and even a low-end laptop typically has 250GB of solid state disk space.
The Generative AI boom, with LLMs and associated inference and fine tuning tools, requires several orders of magnitude more capacity, in number of processor cores, specific types of processor, memory and network speed. Even before Generative AI, enterprises require data storage capacities in Terabytes (1024 Gigabytes), Petabytes (1024 Terabytes), Exabytes (1024 Petabytes) and beyond.
The Need For Speed
Processing all this data has fuelled the demand for faster silicon. Processing speed in Teraflops (trillions of floating point operations per second), memory bandwidth in TB/s. Nvidia’s Graphics Processing Units (GPUs) went from being a niche product for gaming computers to a must-have processor for Generative AI. Now more semiconductor makers are coming out with GPUs, Language Processing Units (LPUs), Neural Processing Units (NPUs) and other appliances.
At SuperAI we saw new semiconductor players including Groq, FuriosaAI and AMQ Semiconductors, as well as renewed vigour from established hardware companies such as Intel, AMD, Cisco and Palo Alto Networks.
Counting the Cost
The drive to AI increases data centre demand, especially in Singapore, Malaysia and Thailand. Analysts at Maybank report that data centre demand is expected to grow at around 20 per cent a year for the next five to seven years.
All this computing capacity comes at a cost. The realisation of this cost of AI was illustrated by such talk topics as “How To Run AI At Scale Without Going Broke”. In this session by Jack Collier, CMO of io.net, he shared how they need to double their GPU power every six months just to keep up with the current level of demand.
Scott Albin, Groq’s General Manager Asia Pacific, emphasised the need to focus on the cost per token (measure of inputs and outputs in AI models), and improving the Joules per token (the power required to generate a token).
Three Types of Enterprise AI Players
In addition to the semiconductor firms, the conference revealed three distinct categories of companies shaping enterprise AI adoption. Each represents a different approach to the market opportunity.
- Established enterprise software providers like Oracle NetSuite and Chargebee are embedding AI assistants and co-pilots directly into their existing platforms. This pragmatic integration approach suggests the market is maturing beyond experimentation into practical enhancement of familiar workflows.
- Existing infrastructure vendors are extending their core platforms to enable and protect AI initiatives. Kong added AI Gateway capabilities, while Boomi expanded into AI agent orchestration. Security providers like Cisco and Palo Alto Networks are addressing both AI protection and AI-powered security, recognising the dual nature of AI as both opportunity and risk.
- New specialised service providers are emerging to fill specific AI infrastructure gaps. Companies like Airia position themselves as orchestration platforms, while BrightData provides unblockable web access for AI agents. WIZ.AI focuses on multilingual customer service agents, representing the trend toward vertical-specific AI solutions.
Looking Forward
SuperAI Singapore 2025 revealed an AI industry maturing beyond experimentation, with enterprises now grappling with the practical realities of cost, skills gaps, and integration challenges. The emphasis on specialised hardware and orchestration platforms signals that efficiency and manageability are becoming as critical as raw capability.
For enterprises, an incremental approach appears to be prudent, given the continued rapid change in the types and capacity of AI building blocks.
References:
SuperAI session recordings
https://www.youtube.com/@superai_conf
“South-east Asia emerges as global data centre hot spot as AI usage rises”
Straits Times, 14 October 2024
“A case study on the Intel Inside campaign”
https://thebrandhopper.com/2024/09/14/a-case-study-on-intel-intel-inside-campaign/
“Cisco: a bridge to tomorrow”
https://brandthechange.com/case-study/cisco-a-bridge-to-tomorrow/