The short version for a European reader
This week's AI news is mostly American and Chinese corporate announcements. The dossier contains almost nothing on EU regulation, and we will not invent any. The clearest European thread is Parloa, a customer-service AI company whose launch partner describes it as established in Germany with a "robust European enterprise footprint".1 The rest of the picture comes from companies whose products European businesses will buy. The strongest evidence is a set of audited numbers, so we start there.
The hard numbers: what Nvidia spends to make its products
Nvidia's cost of revenue, meaning what it spends to produce what it sells, is in the dossier as checked data from SEC filings. It was $16.621 billion in fiscal 2024, $32.639 billion in fiscal 2025 and $62.475 billion in fiscal 2026.2 That is almost four times the fiscal 2024 level in two years.
The quarterly figures show the same direction. Cost of revenue was $5.638 billion in Q1 2025, $17.394 billion in Q1 2026 and $20.458 billion in Q1 2027.2 One quarter's cost in Q1 2027 is now larger than the whole of fiscal 2024.2 To picture the scale, $62.475 billion is what it cost Nvidia to supply customers for a single fiscal year.2
Rising cost of revenue is not profit. It shows how much product is being made and shipped, but these figures do not tell us how much Nvidia earned on it. The dossier's EPS figures are 11.93 dollars per share for fiscal 2024, 2.94 for fiscal 2025 and 4.9 for fiscal 2026.3 They do not move in a straight line, and the dossier does not say why. We draw no trend from them and suggest you do not either without checking how the company adjusted its share count.
Cash held at the end of fiscal years was $7.28 billion in 2024, $8.589 billion in 2025 and $10.605 billion in 2026.4 Q1 2027 cash was $13.237 billion, below the $15.234 billion of Q1 2026.4 A company spending this much more on production is not simply stockpiling cash.
Do you already own this? The dossier does not say how much of any fund Nvidia makes up, and we will not guess. If you hold a broad share fund, check its top holdings. Nvidia's rivals and suppliers also appear in the dossier: AMD competes with it, and TD Synnex supplies to it.
The networking layer: Cisco and an open standard
On 12 October at the OCP Global Summit, Cisco said it had begun shipping its first G300-powered systems to customers, introduced new 1.6T G302 silicon, and deployed P200-powered systems in production.5 Cisco and NVIDIA also plan to develop an open industry specification for 400G-per-lane pluggable optics, which Cisco says would increase bandwidth density while preserving customer choice.6
Cisco's Kevin Wollenweber put the argument this way: "AI is redefining what it means to scale infrastructure. It's no longer just about individual data centers or clusters, but how we connect compute at massive scale without compromising performance or resilience."6 The word "open" matters to buyers, because an open standard means they are less tied to one supplier. It is also a company's own pitch. The dossier contains no independent test of these products.
Physical-world data: Hesai's Kosmo
On 14 October, Hesai launched Kosmo, a 3D world data engine. It combines the Kosmo One AI spatial camera and the Kosmo World cloud platform to turn physical spaces into 3D digital ones for robot training, world-model development and content creation.7 Hesai announced partnerships with the embodied-AI companies Galbot and Dexmal, and showed applications built with Pixar's RenderMan and Amap.8
Hesai's Shaoqing Xiang described the problem: "Physical AI needs real-world data that is accurate, affordable and available at scale. Meeting all three requirements has been a persistent problem."7 Hesai says Kosmo is "the world's first AI spatial camera", and the dossier has no independent check of that.7
Agents in software and customer service
TELUS Digital announced it will become an implementation partner for Parloa, offering enterprises implementation, integration, change management and managed services.1 Chris Silver said: "The research confirms what we've all experienced ourselves — consumer expectations are high, and brand loyalty is fragile."1 For European firms the practical point is that a German-founded product is being sold through a Canadian-based services company.
Separately, Zeta introduced AthenaOS. Its chief executive David A. Steinberg said it "marks an expansion of Zeta's intelligent AI infrastructure beyond marketing."9
The safety and governance strain
Alongside this momentum, our topic summary points to OpenAI's capability gains, a resignation of a safety lead, and pledges from its leadership on monitoring and open-source timelines. The dossier contains no source document for the resignation or the pledges, so we cannot detail them.
What the dossier does hold is OpenAI chief research officer Mark Chen defending the company after a hacking episode. He said: "I do kind of reject the premise that OpenAI is a company with visible impacts in the world and therefore OpenAI is not training safe and aligned models."10 In a second report he is quoted as saying: "If you disappeared OpenAI, that would be bad for the world."11 Both are one executive's defence of his own employer.
For a European reader this matters because the safety debate is exactly where regulation usually enters. We found nothing in the dossier on how EU rules apply to any of these products, so we make no claim about it. Nvidia lists safety tooling among its own offerings, including NeMo Guardrails, though the dossier says nothing about how well it works.
Memory costs: the pass-through risk
Our topic summary also notes Acer warning about Micron memory pricing and Samsung posting a record profit. The reading is that AI demand for memory chips is strong enough to push up costs for device makers, who may pass them on to buyers. The dossier holds no figures for either company, so we cannot size this.
How far to trust the sources
Most announcements here come from press-release distribution, which is the companies' own description of themselves. Via News measured that source feed as holding up on 56% of 4,960 checked claims.1 The MIT Technology Review items carrying the OpenAI quotes score worse, with 0% of 11 checked claims holding up.10 That small sample is a caution, not a verdict on the quotes. Quotes are accurately attributed, but they are statements by the speakers, not findings.
What to watch
- Nvidia's next quarterly cost of revenue against the $20.458 billion of Q1 2027, and whether cash keeps falling from its Q1 2026 level.24
- Whether the Cisco and NVIDIA 400G optics specification is actually published as an open standard.6
- Whether Kosmo's partnerships with Galbot and Dexmal produce shipped products.8
- Whether any European regulator addresses AI agents and safety monitoring. Nothing in our material says one has.
