Amazon’s goose is slowing down. Leader of the “cloud” market, its AWS division has just announced its weakest quarterly growth (at +20% despite everything) since the group began to detail its results in 2014. The sphere of cloud computing has so far lived under bright skies. The great migration to the cloud, ie the use of the memory and computing capacities of computers and servers distributed around the world and linked by a network, has revolutionized computing over the past fifteen years.
The official launch of Amazon Web Services (AWS) in 2006 enabled a generation of start-ups that lacked the resources to build and operate physical infrastructure to access quality service, based on their needs. The range of services, initially limited to hosting, quickly evolved into data processing, image or text recognition tools. The massive investments made by a few giants at an extremely low cost of capital have made it possible to lower the barrier to entry for new competitors capable of supporting a rate of hyper-growth.
Large organizations, including the most conservative ones such as banking or insurance, have followed these start-ups, and the announcements of “cloud-first” strategies, aimed at deporting physically owned servers to large infrastructure. hosts, have multiplied. Remember Total signing with great fanfare, in 2018, an agreement with Google Cloud to jointly develop artificial intelligence solutions applied to the analysis of subsurface data for the exploration and production of oil and gas .
Elon Musk wants to reduce his cloud bill
After years of moving everything to the cloud and relentlessly using artificial intelligence, machine learning and deep learning programs that require tens or even hundreds of GPUs, without worrying about the associated costs, however, customers are beginning to take a closer look at their bill. And to initiate optimization strategies. Last November, after buying Twitter, Elon Musk asked the bluebird social network teams to achieve up to $ 1 billion in annual savings on infrastructure costs, much of which was provided by GoogleCloud. Many software start-ups have also realized that by dint of using cloud services their gross margin, which was supposed to be high, was in fact largely eroded.
The issue of sovereignty has finally come to the fore for European players. Now, the most sophisticated companies are adopting hybrid strategies by repatriating storage, and sometimes even data processing operations, to local systems in their hands. Graphics process maker Nvidia estimates that relocating some large, specialized learning workloads to the user can yield savings of 30%. As a result, the leader AWS (39% of the market in 2022 according to Gartner) is not the only one to see the climate cool. In this area where critical size is everything, Google Cloud posted a loss of 3 billion dollars over the year. On the Microsoft side, the profitability of Azure is not known, but the giant has announced a deceleration in the growth of its cloud revenues.
The new wave of artificial intelligence, however, is expected to drive renewed demand for cloud solutions. Many companies will want to train their model on their private datasets, such as finance giant Bloomberg. Training its BloombergGPT model took about 53 days of computations running on 64 servers, each containing eight Nvidia A100 GPUs in Amazon’s AWS cloud. At the commercial price of $33 per hour, training the model alone would have cost nearly $3 million. Other companies like Airbnb, Expedia or Instacart have announced connectors with ChatGPT that allow the OpenAI chatbot to directly access their systems and make reservations, operations requiring computing power each time. The cloud providers who know this well continue to invest in their infrastructures (+ 2% in the USA, + 5% in China) while waiting for the return of good weather.
Robin Rivaton is Managing Director of Stonal and a member of the Scientific Council of the Foundation for Political Innovation (Fondapol).