The Expanding Influence of AI

Generative AI, with its massive training workloads, is having a huge impact on a broad range of technologies and related hardware supply chains.

It is nearly impossible to avoid hearing about how artificial intelligence has begun impacting nearly every aspect of our lives. Already, AI is being applied to everything from medical diagnostics to judging beauty contests. The long-term repercussions of artificial intelligence technologies have been compared with change brought by the Industrial Revolution. One can only wonder if the promise of AI will become tempered over time and settle down to more reasonable expectations or if it will surpass what we can only imagine today.

The historical course of new technology introductions can be charted by the Gartner Hype Cycle. It illustrates the typical timeline of a breakthrough technology as it matures into a productive opportunity. Some innovations, such as LED illumination replacing incandescent light bulbs, take many years before gaining acceptance, while others, such as cell phones, quickly became a universal necessity.

Although early development began in the mid-1950s, AI burst into the market with the introduction of Chat GPT in late 2022 and has continued to advance at unprecedented rates. Some pundits have predicted we are still very early in the inflated expectations portion of the curve, while others claim we are already on the slope of enlightenment. Electronic and optical component manufacturers have experienced exceptional sales growth due to the race to build ever larger AI computers, and this is expected to continue through much of the current decade.

There is little doubt that generative AI with its massive training workloads is having a huge impact on a broad range of technologies and related hardware supply chains. By one estimate, tech industry leaders spent $200 billion on AI-related capital expenditures in 2024. Advances are being accelerated by the demands of bigger and faster computer clusters driven by large language models, which are expected to increase in size tenfold per year. The result is that the basic architecture of hyperscale data centers is being altered with new requirements. Remote data centers are being connected using high-speed, low latency optical links to create a virtual machine larger than practical in a single location. This also spreads the enormous consumption of energy and water among multiple sites.

Packaging high-performance computational power in high density envelopes has resulted in a booming market for new high-speed and power distribution components of all types. This has led to seven significant developments:

1. Copper backplane connectors are now being rated to 224 Gb/s PAM4.
2. Demand is growing for high-density and high-current power connectors that are designed for up to 600V DC to improve the efficiency of power distribution in data centers. Bus bar power distribution connectors are rated to hundreds of Amps.
   

   Saf-D-Grid from Anderson Power Products

3. Momentum to replace copper PCB traces with high-speed twinaxial cables that offer reduced signal loss at high speeds is continuing to grow.
   

   Amphenol’s OverPass™ cable system

4. High-speed links to top-of-rack and between computer racks require a mix of passive and active copper cables as well as active optical cable assemblies.
5. GPUs have evolved to become a key component that enables the acceleration of incredibly complex computing tasks. Leading edge AI machines utilize up to 100,000 accelerators and have made suppliers such as Nvidia the darling of the stock market.
6. The electrical energy demands of these massive machines, which can reach 120,000 watts per rack, result in major reconsideration of system design and construction. One estimate puts global data center power consumption at 800TWh by 2026, roughly equivalent to the total electrical energy consumed by Japan.
   1. System designers are under a mandate to reduce power consumption, especially at the chip level. Picowatts per bit has become a critical design efficiency criterion. Adoption of liquid cooling is being accelerated as traditional air cooling is struggling to support the thermal density of AI systems. Solutions range from cold plates on individual devices to full immersion systems. Historically shunned due to cost and complexity, liquid cooling is now being considered in the construction of conventional data centers. Interest in grid-connected battery backup using advanced sodium ion chemistry is seen as an alternative to local lead-acid batteries or diesel generators.
   2.  The site selection process for a new data center now includes the availability of adequate electrical energy, which often exceeds that of a medium-sized city. This limits the availability of preferable locations and increases the cost of real estate. Access to large quantities of water and a naturally cool environment have also become factors.
   3. The nuclear energy industry has been reinvigorated as designers are considering the use of local “Mini or Micro” nuclear generators to support new AI data center energy demands, creating opportunities for contractors with nuclear experience.
7. Requirements for extreme performance have pushed manufacturers to speed up the introduction of leading-edge components. This is particularly evident in optical interconnects.
   • • 1. 1. Switch ASICS have doubled performance from 25.6 Tb in 2020 to 51.2Tb now, and 102.4Tb already in the pipeline.
          2. Pluggable QSFP-DD and OSFP transceivers have seen explosive growth with coherent optics quickly moving into volume applications. Sales of optical transceivers are expected to exceed $4 billion in 2024, doubling 2023 sales.
          3. Digital signal processor chips that enable 800G+ optical transceivers are evolving from 7nm to 5nm technology to reduce power, increase performance, and increase packaging density.
          4. Emerging technology, including hollow core and multi-core optical fiber, is gaining attention for its ability to reduce latency and increase data transmission density.
          5. AI computing systems are driving optics inside the box and closer to the switch as co-packaged optics offer greater density and lower power consumption.
          6. Linear optics are being investigated to reduce both cost and power.
          7. The challenge of connectivity to next generation switch chips is being addressed with both twinaxial and fiber optic cable solutions.
             

             Samtec AcceleRate® HP cable system

          8. In order to increase the efficiency of optical transport networking, 800G embedded optical engines will evolve to 1.6 Tb+. Development will require advances in sub 5nm chip technology, increasing demand for next generation semiconductor fabrication equipment. Super C is an example of expanding transmission spectrum to increase optical fiber capacity.
          9. The ongoing architectural transformation to disaggregation is seen as supporting increased compute density and power reduction demanded by AI computers.
          10. The size and complexity of AI clusters is increasing demand for advanced Test & Measurement equipment required to maintain and eventually create automated management systems. Failure in these huge and complex systems becomes unacceptably expensive, making dynamic fault detection and compensation essential.
          11. Advances in silicon photonics are being proposed as a solution to I/O data bottlenecks and packaging density challenges.
          12. Advanced packaging technology includes chiplet architecture, which replaces system-on-chip design by partitioning chip functions into discrete elements. This allows creation of an open ecosystem of modular components, customization, and reusability, as well as optimized performance and power efficiency. Interconnect standards between chiplets is currently a hot topic among designers.
          13. AI is driving the growth of intelligent automation using AI-enabled robots. This new breed of robots uses machine vision and has demonstrated the ability to learn entirely new tasks simply by observing a process or even by a series of simple verbal commands. In addition to working in collaboration with humans, robots can now work cooperatively with other robots.
          14. One pundit estimated that advances in AI technology have accelerated the achievement of general intelligence by 20 years.
          15. Recognizing the importance of AI technology, an international leadership race has already begun. Developers in both the United States and China have chosen an open-source software approach which allows independent developers to advance AI technology faster. The issue of potential harm by rogue AI software is largely undetermined. The U.S. government has implemented a trade embargo on the most advanced U.S.-designed AI chips. This has spurred China to accelerate development of its own chip design and fabrication capabilities. China has also recently nationalized the mining of the country’s rare earth resources — essential to many high-tech devices.

We can only speculate where the application of AI will lead us or how long the current hyper-growth of AI hardware will last, but the AI revolution is stimulating unprecedented and unexpected advances in a broad array of related technologies despite many unresolved issues. It is unclear if the incoming U.S. government administration will continue to support AI research. Will concerns about harmful use of AI result in more government regulation, slowing progress? Will the global race to AI superiority become a key element of economic and political dominance in the future?

What is clear is that AI technology is likely to result in fundamental changes in ways we cannot predict while generating outsized business opportunities. It is going to be an interesting journey.

Look for Bob Hult’s upcoming review of DesignCon 2025 in February, and read his Tech Trends series  high-speed coverage and more show reports in his Connector Supplier archives. Datacom/Telecom

Like this article? Check out our other Artificial Intelligence, Data Centers articles, our Datacom Market Page, and our 2024  and 2025 Article Archive. 

Subscribe to our weekly e-newsletters, follow us on LinkedIn, Twitter, and Facebook, and check out our eBook archives for more applicable, expert-informed connectivity content.