By now, everyone has seen the diagrams of Nvidia, OpenAI, Oracle, AMD and others participating in a “circular economy”. Today’s announcement that Nvidia will invest $1 billion in Nokia opens the possibility of another circular financing opportunity. Nvidia is “juicing the market” for AI-RAN by making a bold investment.
The typical diagram implies a free lunch: Invest $1 billion and create over $100 billion in shareholder value. The companies invest the money in a new architecture, which is expected to justify that value in the future. It sounds like a perpetual-motion machine. A money amplifier. But people are drawing the picture the wrong way.
Here’s my view of how the picture should be drawn:
The money doesn’t appear by magic. Today’s announcement by Nvidia and Nokia alerts people that manage mutual funds and retirement funds that there’s a big connection between AI and networking. The fund managers decide to invest, and $100 billion of capital flows into these two companies. (Edit: After writing these words an hour ago, Nvidia’s boost has grown to $160 billion).
The overall pool of capital invested in public tech companies includes a total of about $27 trillion. That’s a lot of moolah, and it appears infinite for most purposes. But
Nvidia now represents about 20% of the total money invested in tech. How high can that go before the fund managers start to expect an ROI?
In this case, the ROI needs to come from technology synergy. Will AI (in the form of GPUs) become a critical part of mobile networks? Will data center optical networking be significantly enhanced through AI/ML? I have serious doubts.
In my Virtual RAN forecast, I have thoroughly investigated the possibility of 5G or 6G networks running on centralized GPUs. Nvidia has demonstrated great performance with SoftBank, but when I peel back the layers, I find that most of the performance benefits came from distributed MIMO, not from GPU-based AI algorithms. So far, GPU-based AI-RAN seems to be too expensive for mobile networks, and most of the benefits can be achieved with AI-RAN on CPUs, whether that comes in the form of ASICs or SoCs.
There may be other areas of synergy that I’m not aware of, but I can say that betting $100 billion of investor goodwill on GPU-based RAN is a very speculative investment.
Nokia will certainly benefit
Nokia will certainly benefit from the partnership. It’s getting very expensive to develop ASICs for the mobile market, so a shift to vRAN can become a lifesaver on R&D costs. Samsung has already made this step, betting big on vRAN for multiple reasons and saving money on ASIC development along the way.
In the long term, AI models and AI-based automation are likely to drive demand for mobile connectivity. I’ll be releasing a report this week on edge computing that illustrates a new computing architecture where device-based AI models (small language models) rely on low latency to access data storage that’s too large to be stored on the phone, or which comes from many sources. The point is that the distributed AI models will need inference at one location, with training and data storage in a different location.
In this way, AI models and networks will help each other to grow.
I don’t think that cramming 5G or 6G into a centralized GPU is the answer. But AI will create demand for better networks. Network-based AI edge clouds could be the answer to interesting problems in industrial automation or consumer applications. I encourage Nokia and Nvidia to focus on these long-term architectures instead of pushing for short-term synergies.
Joe Madden is principal analyst at Mobile Experts, a network of market and technology experts that analyze wireless markets. Disclaimer: Nokia is a client of Mobile Experts.
Opinions from industry experts, analysts or our editorial staff do not represent the opinions of Fierce Network.