Tag Archives: NVIDIA

Technology Shorts March 2026

Today’s blogs looks at some of the recent breakthroughs coming out of labs and research facilities that could have practical applications that could eventually benefit the broadband industry.

Rainbow Chip. Researchers at the Columbia University School of Engineering and Applied Science have created a chip that turns a single laser beam into a “frequency comb” that produces dozens of light channels at once. As often happens in science, the breakthrough was discovered by accident when the team was working on a project related to Lidar.

Normal laser beams used in telecom are not precise and transmit a closely bunched group of similar light frequencies that scientists refer to as a messy light signal. This new chip creates multiple laser beams in a range of colors, with each beam precisely at a single light frequency. The chip output is called a comb because there is a clear gap between each different beam, so there is no interference between separate light beams. This chip could revolutionize fiber optic technology by simultaneously sending dozens of even-spaced light channels at precise frequencies through a single fiber, with no interference between colors. Scientists have created precise laser beams in the lab for research, but this chip could bring the technology into practical use.

Energy Efficient Wireless Chips. Researchers at the University of Colorado Boulder have developed a new device that could revolutionize wireless technology. The breakthrough is the creation of a surface acoustic wave (SAW) phonon laser that can create ultra-high frequency vibrations on a single chip. The new device layers silicon, piezoelectric lithium niobate, and indium gallium arsenide to amplify radio vibrations much like a diode laser amplifies light. SAW technology is already embedded in smartphones, GPS, and radar systems and is used to filter signals and reduce noise. However, today’s SAW technology  requires multiple chips and external power. The new phonon technology simplifies this to a single chip that can be powered by a battery. The new chip can also reach far higher frequencies and currently operates at about 1 gigahertz, but has a clear development path to boost this ability to tens or even hundreds of gigahertz.

Efficient Power Module. Researchers at the National Renewable Energy Laboratory unveiled a breakthrough that could squeeze more power from existing electricity supplies. They’ve created a silicon-carbide-based power module they call ULIS (Ultra-Low Inductance Smart). The ULIS device dramatically improves the way electricity is converted and delivered inside devices. Most electronic devices contain a power module, which houses the power electronics that regulate the flow of electricity inside the device. The ULIS device is smaller and lighter while bringing up to a five times improvement in power efficiency. The device would make sense in data centers, electric grids, and any devices using next-generation electronics, like in ships and aircraft. The secret to the success of the new device is that it slows parasitic inductance by seven to nine times, which is the resistance to the process of changing or converting an electric current inside a device.

Some of the benefits come from its new design. Traditional power modules stack components inside a box-like package, while ULIS has found a way to arrange components in a two-dimensional octagon. This creates a smaller light-weight device that also minimizes magnetic interference. One of the most interesting features is that the device can be controlled wirelessly, without needing to be connected to communications cables.

ULIS is expected to impact multiple sectors. Probably the most beneficial is in the electric grid. Today, the devices in the grid require electricity to be converted into a usable form before entering every smart device in the grid. The ULIS device could make this conversion more efficiently and with less power loss in the grid.

Cooling Data Centers with Hot Water. One of the biggest challenges of large data centers is having a large supply of cool water for cooling. At CES this year, Nvidia CEO Jensen Huang announced the company is using water at 45 degrees Celsius (113 degrees Fahrenheit) to cool supercomputers. This is a big breakthrough because hot water doesn’t require water chillers and the accompanying power-hungry compressors. Those devices account for about 6% of the power used at a data center. This breakthrough could be a boon for two-phase liquid cooling systems. Most liquid cooling systems today circulate water, which then must be cooled before reuse. A two-phase system extracts more heat from computers by using the heat to convert the liquid to a gas and then converting back to a liquid. This is not a new technology and has been used on a limited basis for a few years, but the NVIDIA announcement will prompt data center owners to consider hot water as the primary way to cool data centers. The announcement instantly tanked the stock prices of companies that make cool-water chillers for data centers.

AI in Telecom

NVIDIA recently issued its third annual State of AI in Telecommunications report. The company manufacturers many of the cards used in AI data centers, so the company is clearly focused on AI adoption. NVIDIA issues similar reports for other industries.

The 2025 report is the result of a survey that NVIDIA administered to 450 telecom professionals across the globe. The respondents represented a global mix of telecommunications companies, including network operators, system integrators, internet service providers, network equipment providers, independent software vendors, and more. Sixty percent of respondents were executives and directors of companies. The report doesn’t tell us anything about the companies included in the survey but does note that NVIDIA has mostly been concentrating on cellular carriers.

The reason that is important is that I found this survey when somebody online mentioned this report when saying that most telcos have adopted or will soon be adopting AI. The NVIDIA report is fairly careful to say the results represent the views of the 450 respondents, but even NVIDIA occasionally slips up with global statements such as, “97% of telcos are adopting AI. Nearly half are already deploying it”.

Here is how NVIDIA claims AI is being used in telecom:

  • 37% are using AI in network operations and planning.
  • 44% are using AI to optimize the customer experience.
  • 58% are using AI to increase employee productivity.
  • A smaller percentage of respondents are using AI for improved security, customer retention, energy conservation, regulatory compliance, marketing optimization, retail operations, and logistics.
  • 21% say AI has helped them increase revenues by more than 10%, with another 24% saying revenues have increased 5-10%.
  • 11% claim that AI has helped them reduce costs by more than 10%, while another 26% say costs are down 5-10%.
  • 43% say that finding AI experts has been the key obstacle to further AI adoption.

All of this may be true for the largest carriers around the world, but I can’t imagine it is true for the large number of smaller ISPs I know.

I’m curious how many smaller ISPs have done anything serious with AI. It’s hard for anybody to claim they don’t use AI since it is now seemingly built into every online tool in some manner. I’ve talked to a number of folks at ISPs who use AI to help write emails, memos, and reports.

I’d love to hear from folks at small ISPs. Is your ISP using AI for anything like what is described by NVIDIA? That might include things like using AI to chat directly with customers online or using AI to monitor network performance. Everybody I’ve talked to is curious about the potential of AI, but almost universally they aren’t ready to trust it in real-time applications. I’m curious if there is a quiet AI revolution going on at smaller ISPs that I haven’t heard about.

In a few years small ISPs might not have a choice. I have to think that in a few years that many of the software packages that come with network gear will include AI application to monitor performance or troubleshoot problems, but I haven’t heard any claims about that yet from vendors.