For most of the last decade, the global race in artificial intelligence has felt like watching a single company extend its empire with little resistance. NVIDIA reigns supreme, its GPUs powering everything from research labs to hyperscale data centres, its software ecosystem dictating the pace of progress, and its pricing power turning the AI boom into one of the most lucrative monopolies in modern tech history. Across the Atlantic, however, a smaller and far hungrier challenger has appeared – a European startup called Euclyd. Out of stealth, Euclyd has unveiled Craftwerk, a radically different inference engine that carries with it more than technical ambition. It embodies a David vs Goliath clash, a European upstart daring to stare down not just NVIDIA but the American stranglehold on AI infrastructure itself.
Euclyd is not simply proposing another chip as an alternative to the GPU. It is reimagining the entire structure of inference hardware. Craftwerk is a system-in-package of near mythic proportions: 16,384 processors stitched together as chiplets that together promise 32 petaflops of performance. At its side sits a custom memory system, the Universal Bandwidth Memory (UBM), pumping one terabyte of DRAM at 8,000 terabytes per second of bandwidth. With this weapon, Euclyd insists it can deliver inference orders of magnitude faster than NVIDIA’s current flagship GPUs – 20,000 tokens per second on a single Craftwerk unit compared against NVIDIA’s 1,000, and more than 7.6 million tokens per second across a rack, enough to run the latest large language models at a scale previously out of reach except for America’s tech titans.
The weight of David’s slingshot
To understand why Craftwerk has been described with biblical undertones, one has to consider the imbalance between Euclyd and NVIDIA. On one side stands a European newcomer, fresh out of stealth, armed with a novel architecture but with none of the entrenched advantages of incumbency. On the other is NVIDIA, a trillion-dollar colossus, wielding CUDA as both shield and weapon, dictating supply constraints, holding the purchasing budgets of hyperscalers hostage while reaping margins unheard of in the semiconductor industry. In sheer scale, reputation, and resources, the mismatch could hardly be greater.
But what Euclyd grasps is that David’s victory was never about size; it was about precision. If NVIDIA’s weaknesses lie anywhere, they are in power consumption, cost per token, and accessibility. GPUs remain astonishingly effective general-purpose accelerators, but their reliance on memory hierarchies and bandwidth bottlenecks means inefficiencies multiply as models balloon to billions of parameters. Craftwerk, in focusing on solving those constraints outright, aims its slingshot directly at the forehead of NVIDIA’s strength. Deliver more tokens, at lower energy draw and cost-per-output, and you open markets the American Goliath has ignored. In the hands of resource-constrained players – universities, research institutions, small governments – the equation of access begins to shift.
This is where the symbolic European angle becomes unavoidable. For three decades, Europe has sat on the sidelines of advanced compute, its semiconductor champions focused on automotive or niche electronics, leaving America and Asia to dominate the bleeding edge of data centre performance. For Euclyd to emerge from this backdrop – not with a modest accelerator card but with an architecture bold enough to openly confront NVIDIA’s hegemony – is as much about sovereignty as it is about silicon. The David vs Goliath metaphor takes root not only in the underdog startup versus the silicon giant, but in Europe staking its fragile independence against America’s grip over AI’s hardware foundations.
Bandwidth as the new battleground
Traditionally, compute dominance has been measured in raw FLOPS, the raw arithmetic throughput a device can deliver. NVIDIA has mastered this game better than anyone, accelerating graphics processing into a new computing paradigm and monetising it with ruthless efficiency. Yet in the world of large-scale AI inference, FLOPS alone are no longer the decisive metric. As models like GPT-class systems devour ever larger volumes of data, memory access and bandwidth have become the limiting factor. Cores are useless if they sit idle waiting for parameters to arrive.
Euclyd’s UBM memory design is a direct response to that. By coupling one terabyte of DRAM with 8,000 terabytes per second of throughput, the company is attempting to obliterate the memory bottleneck altogether. Instead of relying on high-bandwidth memory augmentations bolted to GPU dies, Euclyd makes bandwidth the foundational resource of Craftwerk. This is not a minor tweak. It reframes the priorities of hardware design, elevating data delivery over sheer core density. If Craftwerk works as promised, tokens flow through continuously, cores stay active, and per-token efficiency soars. In other words, the startup’s slingshot aims at the exact place where NVIDIA’s GPU-based Goliath is most heavily armoured yet resolutely vulnerable.
The historical precedent is tempting to draw. NVIDIA itself once disrupted giants, transforming from a gaming graphics upstart to the core engine of modern AI by exploiting parallelism long before Intel or others took it seriously. Euclyd is attempting the same manoeuvre in Europe – read the battlefield differently, exploit a chokepoint the empire has grown comfortable ignoring, and attack with an asymmetric design. It is a strategy true to the David role: don’t fight where the giant is strong, strike where he least expects.
Power, cost, and the monopoly problem
Behind all the raw throughput, the deeper battle is financial and ecological. Today’s data centres bristling with NVIDIA GPUs are staggering consumers of electricity. Entire towns could be powered on the wattage required to keep a hyperscaler’s AI fleet alive. Regulators are already growing restless, environmental groups are sharpening their knives, and the absolute cost of generation and cooling infrastructure is spiralling upward. If Craftwerk can indeed deliver tokens at a fraction of the power draw, it becomes more than an engineering breakthrough; it becomes politically and economically vital.
Cost-per-token is the other glaring weak spot. A rack of NVIDIA GPUs might require capital running into the tens of millions of USD$, plus ongoing costs that lock smaller enterprises out of the game. Craftwerk’s promise is to overturn that equation – not only providing better throughput at lower wattage but doing so in a footprint that reduces the barrier to entry across industries and regions. This is where the David vs Goliath story resonates most sharply. The startup is not only fighting NVIDIA; it is fighting the gravitational field of dollar economics that has concentrated AI power in a handful of American skyscraper campuses. Deliver cost and energy efficiency at scale, and you deliver independence.
That independence matters on a geopolitical level. America has turned NVIDIA into a quasi-strategic asset, just as Europe stumbled into dependence. By offering an alternative, even if initially limited in adoption, Euclyd signals to governments, universities, financial institutions and even defence agencies that the American monopoly is not invincible. A David with a weapon strong enough to at least bruise Goliath forces a rethink not only in technology but in sovereignty.
Europe’s chance at sovereignty
For policymakers in Brussels and beyond, Euclyd is not just a startup unveiling ambitious silicon. It is a potential beachhead in the wider war of technological sovereignty. Europe has long feared being left as a client continent in the 21st century’s battles for digital wealth and influence. With semiconductors, cloud infrastructure, and now AI platforms dominated by foreign incumbents, the arrival of a bold local contender offers more than excitement. It offers a potential rallying point.
It is therefore likely that Euclyd’s journey will intersect with state policy in unusual ways. Support from industrial sovereignty funds, subsidies under the European Chips Act, or even contracts from governments seeking to shore up their digital autonomy could provide the crucial confidence needed to overcome NVIDIA’s entrenched dominance. Much like David rallying his people against a giant that none dared to face, Craftwerk might emerge as both symbol and tool in a European narrative of resistance.
The risks remain enormous. Chiplet packaging at this scale is complex, demanding fabrication partners outside Europe, and the realities of yield, heat, and system integration could bring down even the most promising architectures. Moreover, CUDA’s hold over the software ecosystem is deep, and overcoming it even in inference will require sustained effort. Many promising startups have already tried and found themselves marginalised into niche markets. Yet should Euclyd falter, it will not erase the symbolic power of having tried. By forcing the conversation beyond NVIDIA’s monopoly, it may inspire further challengers to step forward.
A fight worth watching
In truth, the David vs Goliath metaphor may be too narrow. What unfolds here is not only a struggle between a brave startup and a trillion-dollar giant, but between Europe and America over who gets to define the hardware fabric of the next stage of AI evolution. If NVIDIA retains unchallenged control, the spoils of the AI age risk flowing to a narrow handful of US companies. If Euclyd or others like it succeed, the field becomes more balanced, giving rise to a more distributed and democratic landscape. Both scenarios are possible, but the entrance of Euclyd means the fight cannot be dismissed as a foregone conclusion.
In Silicon Valley, eyes will remain on CUDA’s roadmap, on Blackwell and its successors. In Europe, however, there is a sense of something different brewing – a small stone in the sling, a burst of daring where resignation had settled in. Whether Craftwerk lives up to its claims remains to be seen. But one truth is already established: NVIDIA and America’s grip on AI hardware will not go unchallenged. For the first time in years, the giant feels the sting of a challenger aiming high.
The world has seen this story before. Sometimes the giant fells the underdog. Sometimes, improbably, the underdog lands the strike that changes history. Euclyd is making that bet in silicon, and the world will be watching closely to see if Craftwerk becomes the slingshot that finally reshapes the battlefield.
