Cerebras $2.5B: physical AI changes the rules for SMBs

The timeline of an exit the market didn't see coming

In May 2026, TechCrunch has reported the details of Cerebras Systems' $2.5 billion exit. The Eclipse fund, founded by Lior Susan, is among the main beneficiaries of the deal. However, the story begins much earlier.

About ten years ago, Susan started investing in companies that built technology for the physical world. It was a lonely position. Most venture capital looked at software, SaaS, digital marketplaces. In contrast, Eclipse focused on chips, sensors, robotics, and tangible computational infrastructure.

Cerebras has become the most visible demonstration of that thesis. The company has developed extraordinarily large chips — the Wafer Scale Engine — capable of processing large AI models with reduced latency. Furthermore, it has built complete systems, not just raw silicon.

Cerebras architecture: why silicon matters more than software

To understand the value of the operation, it’s helpful to look at the technical architecture. Cerebras chips don’t follow the traditional GPU cluster paradigm. Instead, they integrate thousands of cores on a single silicon wafer.

Consequently, communication between compute units happens without the typical bottlenecks of inter-GPU networks. This translates into concrete advantages on specific workloads: training language models, high-frequency inference, complex physics simulations.

According to analysis published by Gartner , by 2027 over 60% of companies adopting AI in production will have to make structured decisions about inference hardware. Therefore, the choice of chip is no longer a matter to delegate to the IT department: it becomes a strategic decision.

Specifically, for SMEs building internal AI pipelines — predictive analytics, document automation, product recommendations — the availability of specialized infrastructure changes the make-or-buy calculations.

Winners and losers in the AI hardware ecosystem

Cerebras' exit reshapes some hierarchies. Among the most obvious winners are Eclipse and its LPs, but also the entire segment of physical AI investors. In fact, the operation legitimizes an investment thesis that for years remained marginal compared to software-first.

Among the players who need to rethink their positions are generalist cloud computing providers. Companies like AWS, Google Cloud, and Azure have built their advantage on horizontal flexibility. However, the vertical specialization of AI hardware creates niches where traditional cloud is not competitive, neither in cost nor in performance.

Similarly, traditional GPU makers — led by NVIDIA — must monitor the rise of alternative architectures. Despite this, NVIDIA maintains a huge ecosystem advantage thanks to CUDA and its installed base. The duopoly is not in question in the short term.

To dive deeper into the competitive dynamics in AI hardware, the report McKinsey State of AI offers a structured take on the ongoing tensions between industry giants.

SHM Studio's take: what changes for non-hyperscalers

We at SHM Studio we work daily with Italian SMEs that are integrating AI into their processes. Therefore, we observe this situation from an operational, not speculative, perspective.

The main point is this: physical AI infrastructure is becoming accessible even outside of large data centers. This is happening through two main channels.

• Specialized cloud: providers like CoreWeave or Lambda Labs offer access to Cerebras chips and NVIDIA alternative hardware on an hourly basis. Therefore, an SME can test workloads on different architectures without purchasing hardware.
• Hybrid on-premise models: some manufacturing and logistics companies are considering local deployments for latency and data sovereignty reasons. In this scenario, the choice of chip becomes part of the overall architectural project.

So, the question for an SME is not 'do I need to buy Cerebras chips?'. The question is: 'does my AI roadmap in the next 24 months require a reflection on computational infrastructure?'. In most cases, the answer is yes.

For companies building a structured digital presence, our AI services include a technological assessment phase that also considers infrastructural choices. Likewise, our approach to Digital marketing always integrates the smart automation component.

Operational implications for Italian SMEs

Translating financial news into concrete actions is the job that SMEs often struggle to do internally. Therefore, let's try to indicate some practical directions.

First of all , it's useful to map current and future AI workloads. Not all use cases require specialized hardware. For example, a text classification model on medium volumes runs perfectly on standard cloud instances. Conversely, real-time inference on industrial sensor data can benefit from dedicated architectures.

Afterwards , it's worth monitoring the evolution of prices in the specialized cloud market. Competition among providers is compressing costs. Therefore, solutions that seemed inaccessible 18 months ago are now within the budget of companies with 50-200 employees.

Finally , it's worth structuring governance for AI technology choices. This means not delegating infrastructure decisions solely to the IT department, but also involving strategic management and, where present, the CFO. The implications for TCO (Total Cost of Ownership) are significant.

For those managing digital campaigns with automation components, choices regarding Google Ads and LinkedIn Ads are evolving towards more intensive use of predictive models. Therefore, the quality of the underlying infrastructure directly influences campaign performance.

Still a work in progress: where Eclipse is heading after Cerebras

According to TechCrunch, Lior Susan considers Cerebras's exit only the beginning of the physical AI thesis. Eclipse has other companies in its portfolio operating at the intersection of hardware, robotics, and artificial intelligence.

This signal is relevant for understanding where capital will be concentrated in the coming years. Among other things, the "physical world AI" thesis overlaps with trends already underway: the digitalization of manufacturing, logistics automation, and computer vision systems for retail.

For Italian SMEs, many of which operate precisely in these sectors, the message is clear. In fact, AI is no longer just a topic for software houses or digital startups. It's a topic that concerns those who produce, distribute, and sell in the physical world.

Our activity of SEO and Copywriting for B2B SMEs takes this change into account. As does our approach to web development , which increasingly integrates data-driven personalization components.

Next moves: three questions to get your bearings

We close with a simple framework. Three questions that every SME should ask themselves after reading this news.

• Which business processes could benefit from real-time AI? If the answer includes physical operations — production, warehousing, field service — then computational infrastructure becomes relevant.
• Does our current cloud provider offer access to specialized AI hardware? If the answer is no, it's worth exploring alternatives before the AI roadmap clashes with technical limitations.
• Do we have governance for AI tech decisions? If choices are made on a case-by-case basis without an overall vision, the risk of architectural inconsistency grows over time.

To explore these topics with a consultative approach, you can contact the SHM Studio team . Also, our Blog regularly publishes analyses on AI, digital infrastructure, and strategies for Italian SMEs.