The recent announcement of Amazon’s second-generation Trainium AI chip is just another proof of the seismic transformation reshaping one of computing’s most foundational pillars: microprocessor chips.
This is an exciting time for the world of computing. For the first time in decades, we’re seeing a wave of new hardware chips being designed and brought to market by numerous players. Some of these, like Amazon’s Trainium and Google’s TPU, are purpose-built for AI workloads, while others are general-purpose CPUs that are more efficient than ever, delivering incredible performance per watt of energy consumed. This diversity of chip designs is opening up new possibilities and changing the game for what chips can do—and who is leading the charge.
We’re living through what feels like a “chip renaissance,” a period that brings to mind the technological creativity and boldness of the 70s and 80s. Back then, a wide variety of unique and experimental CPU solutions hit the market at a rapid pace. This was before the 90s, when the industry began to consolidate around mainstream CPUs using the x86 instruction set, driven by the success of IBM-compatible personal computers and the rise of Windows as the dominant operating system.
This powerful partnership between Microsoft and Intel created two big winners and relegated many once-prominent competitors to the pages of history books. Take Motorola, for instance, once a key player in providing chips for some of the most popular home computers of the 80s and 90s, now largely forgotten in the world of processors. For years, Intel held the crown as the dominant chipmaker, with only occasional challenges from AMD. It wasn’t until the relatively recent resurgence of AI that Intel’s dominance began to waver. The explosive demand for GPU architectures, perfectly suited to AI workloads, catapulted NVIDIA to the forefront. So far, NVIDIA reigns as the undisputed leader in GPU-based computing.
How New Players Are Reshaping the Chip Value Chain
Things change fast, and the landscape of chip design has shifted dramatically. Fresh designs are rolling out at a breakneck pace, with multiple players launching new hardware every few months. But what sets this moment apart from the 70s and 80s isn’t just the speed of development, it’s the players involved. Many of the biggest names driving this wave of change aren’t traditional hardware manufacturers, but tech giants and cloud providers stepping into the game. The rules of chip design are shifting, and with them, the entire computing landscape.
Unlike the 70s and 80s, when traditional hardware companies dominated the scene, many of today’s biggest contenders come from outside the usual hardware world.
Until recently, each part of the value chain in computing operated within a single layer. Hardware vendors focused solely on designing chips, which they sold to manufacturers who integrated them into computers. These computers were then sold to end users, often through yet another intermediary. But the process didn’t end there, users still had to purchase an operating system and install software to actually make the computer functional and finally get direct value from it.
The key point is that each link in the chain was historically handled by a different vendor. The business was neatly divided into layers, with each player taking a share of the pie: Intel focused on making chips, not selling PCs; Dell assembled and sold PCs but didn’t produce chips or operating systems; and Microsoft sold operating systems and application software, staying far away from hardware. But that clear separation is no longer the case. Today, Microsoft is not only in the software business but also building its own AI accelerators and CPUs. Amazon has joined the fray, designing both AI-specific accelerators and general-purpose CPUs. Google, too, is doing the same, developing its own AI chips and CPUs.
This shift fundamentally changes the rules of the game. And why is that? It all comes down to the fragile economy of chip design…
The Economy of Chip Design: Who Has the Upper Hand?
Designing and building a new chip is an extraordinarily expensive and complex undertaking. It’s not just about creating the chip itself, there’s an entire chain of dependencies that must align to ensure the chip has a viable market and generates enough sales to justify the investment.
Think about it: even if you successfully design a cutting-edge chip (a monumental and costly achievement on its own), you still need to sell it to a company that can integrate it into computers and bring those products to market. That’s a massive challenge in itself. And even then, the job isn’t done. To make the chip truly useful, you need an ecosystem of compatible software so that end users can get actual value from the computers built with your hardware.
It’s an incredibly tough business! No wonder the “old incumbent,” Intel, managed to dominate the industry for so long. And now, the “new incumbent,” NVIDIA, seems poised for a substantial reign as well, though it’s unlikely to last quite as long as Intel’s.
The fragile economic chain of chip design shifts dramatically when it comes to hyperscalers—massive cloud providers like Amazon, Google Cloud Platform (GCP), or Microsoft. Why? Because hyperscalers inherently solve the toughest part of the economic equation: volume. They can afford to design chips exclusively for their own use, powering the massive computing demands of their sprawling data centers.
What’s more, these companies have the financial resources and patience to develop chips over time, gradually refining and integrating them into their infrastructure. Even if they lack the knowledge and expertise in chip design initially, their financial resources and long-term strategies put them in a unique and powerful position to build their very own silicon ecosystems.
This combination of scale, capital, and patience gives hyperscalers like Amazon, Google Cloud Platform (GCP), or Microsoft a significant edge in the chip game.
Plus, hyperscalers are the gatekeepers to the world of large-scale computation. This gives them the unique ability to strike a careful balance between buying chips from external suppliers like Intel, AMD, and NVIDIA, and working on their own chip designs. What’s more, they can test their in-house chips in real-world scenarios, learning and improving as they go, all without disrupting their daily operations. It’s a win-win that lets them build expertise while keeping their options open.
This is what you might call an “unfair advantage” that hyperscalers have over traditional chip designers, even compared to industry giants like Intel, AMD, and NVIDIA.
Welcome to the New Chip Ecosystem
So, what does this mean for the chip ecosystem?
First, it’s highly likely that these three hyperscalers—Google, Amazon, and Microsoft—will succeed in rolling out advanced computing chips at scale. At the same time, they’ll be able to make their chip design and production efforts economically sustainable. This essentially means that the ecosystem now has three new, fully capable hardware designers who are here to stay. And with their massive scale and unique business models, they might even operate with more favorable economics than their traditional competitors: AMD, Intel, and NVIDIA
Second, there’s a good chance these hyperscalers won’t just keep their chips for themselves. If selling chips turns into a lucrative, high-margin business, we might see them offering their designs to other companies, even rival hyperscalers. And why stop there? They could eventually tap into the consumer market too, which remains a huge opportunity. Who knows? In the not-so-distant future, we might walk into a store and see laptops with a label like “Powered by AWS.”
This will likely shake up the software world too, in ways we can’t fully predict yet. Software developers might have to start creating multiple versions of their code to take advantage of the unique features each chip offers.
In short, we’re on the brink of witnessing a big bang in the diversity of computing solutions offered by hyperscalers. Some of these will be accessible through the cloud, while others might even be available for on-premise deployment. It’s an exciting time for computing!
We’re likely to see a surge in the variety of computing solutions provided by hyperscalers, from cloud-based to on-premise options.
From Chip Diversity to the Allocation Problem
What does this mean for engineers and scientists working on large-scale simulations and scientific computing?
With so many computing solutions available, offering different performance levels and price points, how does an engineer or scientist allocate resources effectively for large-scale computing jobs? How do you decide on the specific machine type, configuration, or architecture to match your workload? Should you run your workload on the cloud, and if so, which provider and instance type will give you the best value? Or is on-premise hardware the better option?
The allocation problem becomes even more complex when you consider cost-efficiency. How do you avoid overspending on compute resources while ensuring you meet performance requirements? Each computational job may have specific hardware needs, some might require GPUs for parallel processing, others might demand CPUs optimized for high memory bandwidth. How can you navigate this complexity and make smart, informed decisions every time?
Allocation isn’t just about picking the right machine, it’s about understanding the unique requirements of your workloads and matching them to the right hardware, at the right price, every time. This is precisely where Inductiva can make a real difference. In our next blog post, I’ll show you how.
