Amazon Web Services has started talks to sell its own AI chips, the Trainium and Inferentia processors, to data centers outside AWS. It's a significant strategic shift. Until now, these chips were available only within Amazon's cloud ecosystem.
CEO Andy Jassy called the move a $50 billion opportunity. So AWS is no longer just competing with Nvidia on performance: it is now entering the hardware distribution market too. But the challenge is a complex one. Nvidia holds a dominant share of global AI infrastructure, with an established partner network and a software ecosystem, led by CUDA, that is hard to replicate.
For Italian cloud-native SMEs, this scenario opens up real prospects. More competition among AI chip suppliers tends to lower computing costs in the medium term. At SHM Studio, we track these dynamics to steer our clients' infrastructure choices toward more efficient, sustainable AI solutions. In short: the AI chip market is opening up, and SMEs would do well to follow how it evolves.
Amazon's change of course in the AI chip market
Until a few months ago, the AI chips developed by Amazon, the Trainium processors for training and Inferentia for inference, were accessible only through AWS infrastructure. No external data center could buy them or integrate them into its own systems. Today, according to TechCrunch, that logic is changing.
AWS has started formal talks to sell its chips to third-party data centers. CEO Andy Jassy has publicly put the value of this opportunity at $50 billion. That figure signals a clear ambition: no longer just a cloud provider, but a global AI hardware supplier.
So Amazon no longer competes with Nvidia only on cloud services. It is now entering the physical distribution of processors, a field where Nvidia has built a structural advantage in recent years.
Why Nvidia remains a tough rival to dent
Nvidia now controls an estimated 70% to 80% of the AI chip market, according to Gartner analysis. Its advantage isn't just technological. It's ecosystem-wide.
The CUDA platform, Nvidia's parallel programming framework, is built into nearly all major machine learning frameworks: PyTorch, TensorFlow, JAX. On top of that, the network of hardware and software partners Nvidia has built over time is a considerable barrier to entry.
Amazon, by contrast, is betting on an alternative ecosystem based on the Neuron SDK. But that stack is still less mature than CUDA. As a result, convincing data centers to replace or complement Nvidia GPUs with Trainium chips will take no small amount of technical evangelism.
Even so, the strategic signal is clear. Amazon has no intention of staying confined to its own cloud. In some ways, the move recalls the strategy Amazon itself followed with AWS in the 2000s: turning internal infrastructure into a commercial product.
The AI chip market in 2026: an industry being redefined
The backdrop to this move is already in motion. In 2025, several hyperscalers stepped up development of proprietary chips: Google with its TPUs, Microsoft with the Maia project, Meta with its own custom accelerators. So the push to reduce dependence on Nvidia isn't unique to Amazon.
What sets AWS's strategy apart is its willingness to sell externally, not just use chips in-house. That creates a new market segment. Independent data centers, the ones that don't belong to the big hyperscalers, could find in Amazon an alternative supplier at potentially more competitive prices.
On top of that, diversifying hardware sourcing has become a geopolitical priority. Tensions in semiconductor supply chains, which surfaced strongly between 2022 and 2024, have pushed many operators to look for alternative suppliers. Amazon is positioning itself in this space with strategic timing.
The real impact on Italian cloud-native SMEs
For Italian small and medium-sized businesses that operate in cloud environments, or are considering bringing AI solutions into their processes, this news has practical implications, even if not immediate ones.
First, more competition among AI chip suppliers tends to lower computing costs in the medium term. So machine learning workloads, from document classification to predictive sales analysis, could become affordable for smaller companies too.
Second, the availability of alternatives to Nvidia chips could translate into more flexibility in cloud providers' offerings. For example, AWS could offer Trainium instances at lower prices than Nvidia GPU-based instances, encouraging migration to more efficient AI architectures.
Finally, for SMEs already using AWS as their main infrastructure, the news points to a stronger Amazon ecosystem. So building skills in the Neuron SDK and Trainium instances could pay off over the next two years.
What the official press releases don't tell you
Andy Jassy's statements are optimistic, as you'd expect from a CEO introducing a new line of business. Still, some questions remain open.
The first is software compatibility. Selling chips to third-party data centers means they must be able to run them on their own. Without a mature, well-documented software ecosystem, the risk of slow adoption is real. At SHM Studio, we often see this dynamic in digital transformation projects: the technology is available, but adoption takes time and skills.
The second is competitive positioning. Selling chips to rival data centers means, to some extent, strengthening infrastructure that could compete with AWS itself. It's a strategic tension Amazon will have to manage carefully.
The third is timing. Talks are still ongoing. There is no product on the market at scale yet. As a result, SMEs shouldn't expect operational changes right away.
What to watch over the next 12-18 months
The most relevant window for assessing the impact of this strategy runs through the end of 2027. A few indicators deserve particular attention.
- Partnership announcements between AWS and independent European or American data centers.
- Neuron SDK updates that improve compatibility with standard frameworks.
- Price changes for cloud instances based on Amazon chips compared with Nvidia-based ones.
- Nvidia's reaction: possible exclusive deals or price cuts to keep partner data centers loyal.
If you manage SEO strategies, Google Ads campaigns or LinkedIn campaigns with AI components, the cost of inference is already a relevant variable today. Likewise, if you build applications on web platforms with integrated language models, this evolution is worth monitoring.
To dig deeper into how these dynamics shape SMEs' technology choices, the SHM Studio team is available for a consultation. You can also find regular analysis of AI, cloud and digital infrastructure for the Italian market on our blog.
In short: Amazon's move is ambitious and consistent with its history. But the road to a direct challenge to Nvidia is still long. Italian SMEs would do well to follow how it evolves, without expecting overnight revolutions. The AI chip market is opening up again, and in the medium term that's good news for anyone investing in applied artificial intelligence for business.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
