News

We’re not at peak GenAI yet – but APIs move from hype to happening agenda

September was month of Generative AI deals, with Amazon prepared to spend up to $4bn to crash the big league writes Contributing Editor Annie Turner. Meanwhile, the network itself made a quieter but significant return to centre stage.

 

Anthropic, which is developing a generative AI (GenAI) platform, continues to attract substantial investments In August. SK Telecom invested $100 million in the platform to help fund the development of a large language model (LLM) for telcos to leverage GenAI.

Amazon dwarfed that sum in September, saying it will invest up to $4 billion, but its plan is to make Anthropic’s models more useful to AWS’ customers and acquire an unspecified minor shareholding, secured by the initial stake of $1.25 billion.

The deal was widely interpreted as an admission by AWS that its own AI development is lacklustre. Critics have been particularly scathing of Alexa’s lack of progress its understanding language over the last five years or so and Amazon failed to attract much third party interest in training its AI chips.

Now Anthropic will to use AWS’ machine learning accelerators, Trainium and Inferentia, to build, train and deploy new foundation models for AWS as well as technology for AWS’ machine-learning chips. This is a necessary because Trainium and Inferentia cannot run on Cuda, NVIDIA’s accelerator platform – and NVIDIA has about 85% market share in AI-enabling silicon.

 

Jensen Huang, CEO, NVIDIA

NVIDIA in trouble in the EU?

It seems that NVIDIA’s dominance of AI chips, which have made it into the world’s sole trillion-dollar chip maker, has attracted the European Union’s (EU) attention. The European Commission (EC) is investigating suspected anti-competitive behaviour, according to Bloomberg, in the graphics processing unit (GPU) market.

NVIDIA originally manufactured GPUs for gaming before they were used in AI applications. The EC is to assess whether regulatory intervention will be required.

In parallel, French authorities are in discussions with industry stakeholders about NVIDIA’s prominent position in the AI chip sector, which include its pricing strategies, chip shortages, and the potential impact on prices.

NVIDIA’s share price has soared, along with demand, since the AI chatbot, ChatGPT, hit the mainstream just under a year ago.

 

Europe’s efforts to keep up

Xavier Niel, the French telecoms billionaire, is to invest €200 million in a bid to ensure Europe remains competitive with China and the US in AI. He will invest the money through the telecoms group he controls, Iliad, across various projects. They are to include a cloud-based supercomputer powered by NVIDIA, naturally.

Niel said in a statement: “To influence the AI ​​market, you need computing power. To have computing power, you need supercomputers. And to have supercomputers, you have to invest massively.”

Iliad’s cloud subsidiary Scaleway has been striving to offer an alternative to the US-headquartered public cloudcos and their massive investment in GenAI. Niel continued in the statement: “By equipping [the cloud] with a supercomputer, we want — and we can — create a European AI champion. It’s a question of sovereignty: to protect our data, we need platforms established on our territory.”

Niel continued, “It is not enough to create a champion, but an entire French ecosystem. Because we want to contribute to it, we are going to create an AI research laboratory, providing the means and recruiting the best researchers.”

 

DT hopes to drive the API economy

On a different theme, Deutsche Telekom (DT) is to charge developers and business customers for the use of APIs on its mobile network in Germany. Ericsson’s Vonage platform will host the access portal and the service is branded MagentaBusiness API.

The intention is to allow developers and business customers to embed communication functions like video, voice, messaging and other capabilities into their products, applications and workflows. The platform approach is intended to should lower barriers to entry, making the API service available to small- and medium-sized enterprises (SMEs) as well as corporations.

MagentaBusiness API will be launched as a ‘friendly adopter program’. No information was available concerning how and how much third parties will be charged for use of the APIs.

Claudia Nemat, Board Member for Technology and Innovation at DT, commented, “With Vonage and Ericsson, we are the first to expose network APIs in a one-stop shop portal. APIs are a key strategic focus for Deutsche Telekom. This is underlined by our status as a founding member of the CAMARA alliance, which aims to make standardized APIs available internationally.”

 

BT looks to the network for XR

BT Group launched a testbed at its R&D facility in Adastral Park in eastern England to develop and trial immersive experiences in the areas of work, home, health and entertainment. The most interesting aspect of this announcement is that it is looking at how it can sweat its network assets to facilitate this, rather than focusing on user devices.

In particular, BT Group wants to understand how networks, platforms, services and apps can be optimised for cloud-GPU rendered extended reality immersive experiences delivered over EE’s public and private 5G networks. The testbed will look to support a range of extended reality use cases, with a focus on augmented reality which enables blended realities – mixing virtual content with the real-world.

Gabriela Styf Sjoman, MD Research and Network Strategy, BT Group, commented, “Network optimisation is a fundamental enabler for immersive experiences that will require high bandwidth, high capacity, and ultra-low latency networks which can be dynamically configured for the demands of different extended reality service applications.

“With this testbed we’re looking to understand what future extended reality immersive experiences might require from network service providers like BT Group, platform operators and application service developers. These requirements will obviously vary depending on the particular use case.”