Samsung Backs $231M Non-GPU Architecture for AI Inference
Euclyd, backed by $231 million, is building a non-GPU processor and memory architecture for AI inference. The funding highlights a growing architectural push to lower energy and cost of foundational models, challenging GPU-centric data center design.
Beat this week
Last 7 days · Funding
Impact 6.6/10 (+0.4 vs prior). Counts are stories in our record, not a market forecast.
Open the change reportCoverage balance Positive coverage leads. Positive coverage exceeds negative coverage by 45 percentage points.
This story sits in Funding — the counts compare this beat's last 7 days with the previous 7 in our verified record, not a market forecast.
Figures are computed live from our source-verified story record (as of ) The volume change compares this window with the prior 7 days in the same record. — see our methodology for how impact and sentiment are derived.
AI briefing
Key takeaways
- Euclyd, backed by $231 million, is building a non-GPU processor and memory architecture for AI inference.
- The funding highlights a growing architectural push to lower energy and cost of foundational models, challenging GPU-centric data center design.
- CNBC
- Seeking Alpha
In this briefing
Mentioned
Key Intelligence
Key Facts
- 1Euclyd raised a €200 million ($231 million) Series A co-led by Samsung, Somerset Capital Partners, EQT's Scaleup Europe Fund, and Innovation Industries.
- 2Founded in 2024, Euclyd is designing an AI chip system for inference using a non-GPU processor and memory architecture.
- 3Nvidia became the world's most valuable company after its gaming GPUs were repurposed for AI training and inference, giving it a near monopoly on the highest-end AI chips.
- 4OpenAI announced its first AI chip, Jalapeño, in August 2026 with claimed industry-leading speed and efficiency, while Google, AWS, and Meta are also developing their own AI chips.
- 5Euclyd's systems are not yet proven at scale in commercial deployments, but the company claims they will reduce energy needs and costs for foundational model inference.
- 6Euclyd plans two revenue streams: selling hardware and rack systems for self-hosted enterprise inference and licensing chip IP to companies building custom silicon.
| Dimension | ||
|---|---|---|
| Primary workload | Inference | Training and inference |
| Architecture | Non-GPU processor + memory | GPU |
| Commercial maturity | Unproven at scale | Proven at hyperscale |
| Energy/cost profile | Claims to reduce energy and data center costs | Industry-leading performance but high demand |
Analysis
For AI engineers and infrastructure teams, the most important signal is architectural: Euclyd is betting inference needs a fundamentally different processor-memory system than the repurposed gaming GPUs that dominate training. With hyperscalers and OpenAI already designing custom silicon, Samsung's backing adds memory-manufacturing weight to the shift toward heterogeneous AI compute.
The pivotal development is Samsung's decision to co-lead a 200-million-euro, or $231 million, Series A investment in Euclyd, a Dutch startup founded in 2024 that is designing an AI chip system for inference with a fundamentally different architecture than Nvidia's graphics processing units. The round, disclosed by CNBC on September 14, 2026, also drew co-lead backing from Somerset Capital Partners, the EQT-managed Scaleup Europe Fund, and Innovation Industries. The size is extraordinary for a Series A, especially for a company that has not yet proven its silicon in commercial deployments, and it reflects a broader surge in capital flowing into alternatives to Nvidia's dominant AI compute stack.
Success will depend on execution in a market where OpenAI, Google, AWS, and Meta are simultaneously building their own chips and where Nvidia's ecosystem advantage remains deeply entrenched.
Nvidia's rise to become the world's most valuable company was built on repurposing GPUs originally designed for gaming into the engines of AI training and inference. That pivot created a near monopoly on the highest-end AI chips, leaving hyperscalers, startups, and now strategic investors searching for ways to diversify. OpenAI added momentum to that search in August 2026 when it announced its first AI chip, the Jalapeño, which it claimed offered industry-leading speed and efficiency. Google, AWS, and Meta are also developing their own AI processors. Euclyd is entering this crowded but capital-rich arena with a specific bet: inference workloads require a different processor and memory architecture than the GPU-centric systems that dominate training.
The startup's pitch is that its silicon systems for foundational models will reduce the energy needs and costs of AI data center infrastructure, a claim that remains unproven at scale. Euclyd is targeting two revenue streams: selling hardware and physical rack systems to enterprise customers that want secure self-hosted AI inference, and licensing its intellectual property to companies that want to develop their own chips based on pre-existing technology. CEO Bernardo Kastrup framed the broader stakes, saying that AI is becoming a foundation of economic growth, scientific discovery, and national competitiveness, but its potential will remain constrained unless infrastructure is fundamentally changed.
Samsung's involvement is more than financial. The company is one of the world's largest memory manufacturers and a major engineering force, and Kastrup emphasized that Samsung can help in more ways than money. For Samsung, backing Euclyd offers strategic access to emerging non-GPU chip architectures and possibly a closer relationship with inference-focused silicon that could drive demand for advanced memory. The co-lead structure with EQT's fund, Somerset, and Innovation Industries signals that deep tech investors see this as infrastructure-grade capital rather than a typical early-stage software bet.
What to Watch
The implications for the AI chip market are significant. A $231 million Series A for a two-year-old startup underscores both the capital intensity of custom silicon and the urgency investors feel about Nvidia's chokehold on AI compute. It also highlights a growing divide between training, where Nvidia remains strongest, and inference, where workloads are projected to scale rapidly as AI models move into production. If Euclyd can deliver on its energy and cost promises, it could pressure Nvidia's inference economics and give enterprises a viable self-hosting option. However, the road ahead is steep. Euclyd must move from architecture to proven commercial deployment, compete with hyperscaler custom silicon programs, and manage the complexity of licensing IP while also building hardware.
Looking forward, the key milestones will be tape-outs, pilot deployments, and whether the dual revenue model can produce actual customer traction. The round validates the investment thesis around GPU alternatives, but it does not yet validate the technology. Success will depend on execution in a market where OpenAI, Google, AWS, and Meta are simultaneously building their own chips and where Nvidia's ecosystem advantage remains deeply entrenched. Euclyd's ability to leverage Samsung's memory and engineering scale could differentiate it, but the next 18 to 24 months will determine whether this is a genuine architectural shift or merely another well-funded attempt to loosen Nvidia's grip.
Source cluster
Primary reporting
Cite This Page
"Samsung Backs $231M Non-GPU Architecture for AI Inference." AI Intelligence Brief, September 15, 2026. https://getaibrief.com/story/samsung-euclyd-231m-non-gpu-inference-architecture
How we covered this story
Every story in our AI coverage is assembled from multiple primary sources, cross-referenced for factual consistency, and scored along three independent dimensions: sentiment, operational impact, and source-cluster confidence. Single-source rumors and unverifiable claims do not pass our editorial gate. When a story shows "Verified by N sources" with N≥2, the development is independently corroborated; when N=1, we mark it explicitly so readers can weigh the signal accordingly.
Impact scoring uses a 1-10 scale weighted toward regulatory, financial, and operational consequence rather than coverage volume. A topic that runs in every outlet but moves no real decisions ranks lower than a niche regulatory filing that reshapes how operators in the AI space have to behave. Read our full methodology for the scoring rubric, our glossary for term definitions, and our trends index for the longitudinal view across the beat.
Sources are only linked to a story once they clear our classification pipeline at a minimum 35 percent relevance threshold. According to that methodology, reviewed July 2026, this follows multi-source corroboration standards recommended by journalism research bodies such as the Reuters Institute for the Study of Journalism.
See something wrong in this story — a wrong fact, a broken source link, a misattributed entity? Report a data issue.
| Signal on this page | What it tells you |
|---|---|
| Verified by N sources | Independent corroboration count. N≥2 is our confidence floor; N=1 is marked explicitly. |
| Impact score (1-10) | Regulatory + financial + operational weight. 8+ signals an experienced-operator action item. |
| Sentiment | Five-tier classification trained on labeled AI-specific corpora. |
| Timeline | Where applicable, the related-events sequence that contextualizes today's development. |