Why this matters now
Between 12 and 14 October 2026, four companies announced products at four different layers of the AI industry. Cisco unveiled networking hardware, Hesai a data-collection system for robots, TELUS Digital and Parloa a customer-service partnership, and Zeta an enterprise software platform. Each announcement is the company's own account, so we flag what is checked and what is not. The press-release source behind most of them, NewsEOD, has a measured reliability record: 56% of 4,960 checked claims held up.1 That is a coin-flip-plus, and it is the reason this piece separates what companies say from what filings confirm.
The foundation: networking and chips
On 12 October, at the OCP Global Summit, Cisco said the first systems powered by its G300 chip are shipping to customers. It also announced a new 1.6T G302 chip, and said systems built on its P200 chip are already deployed in production.1 Cisco and NVIDIA also said they plan to develop an open industry specification for a next-generation 400G-per-lane pluggable optics design. The aim is more bandwidth in the same space while keeping customers free to mix suppliers.1
Kevin Wollenweber of Cisco framed the problem 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."1 In plain terms, the bottleneck is moving from how fast one chip is to how well thousands of them talk to each other.
What Nvidia's checked numbers show
Nvidia is the company most associated with this buildout, and its SEC filings are the best-verified material we have. They show the cost of making what it sells growing very fast. Cost of revenue was $16.621 billion in fiscal 2024, $32.639 billion in fiscal 2025 and $62.475 billion in fiscal 2026.2 By our own arithmetic, that is roughly a doubling in each of the last two fiscal years. The same pattern shows in the first quarter: $17.394 billion in Q1 fiscal 2026 and $20.458 billion in Q1 fiscal 2027.2
Rising costs alone do not tell you whether the business is healthy. That depends on the revenue those costs produce, and revenue is not in our verified set, so we do not draw a conclusion about profitability. What the filings do confirm is the scale of the activity.
Nvidia's cash balance tells a less dramatic story. It stood at $7.280 billion at fiscal 2024, $8.589 billion at fiscal 2025 and $10.605 billion at fiscal 2026.3 In the first quarters, though, it fell from $15.234 billion in Q1 fiscal 2026 to $13.237 billion in Q1 fiscal 2027.3 The dossier does not say why. Cash can fall because a company spends it on growth, buys back shares or pays suppliers, so the drop is not by itself a warning sign.
One caution about earnings per share. The filings list $11.93 for fiscal 2024, $2.94 for fiscal 2025 and $4.90 for fiscal 2026.4 A fall from $11.93 to $2.94 looks like a collapse, but our dossier does not say whether the older figure has been adjusted for share splits. Please do not compare these three numbers as if they were on the same footing. We would rather flag the seam than invent an explanation.
For readers wondering whether they already own this: the dossier does not give Nvidia's weight in any index, so we will not guess. It does note that AMD competes with Nvidia, and that products such as NeMo Guardrails and the DGX Spark are developed by Nvidia.
The new frontier: data for robots
The most novel announcement came from Hesai, which on 14 October launched Kosmo, a "3D world data engine". It combines a camera called Kosmo One with a cloud platform called Kosmo World to turn real places into 3D digital spaces. Hesai says these can be used for robot training, building world models and creating immersive content.5
The pitch is that robots need real-world examples to learn from, and those are hard to collect. Shaoqing Xiang of Hesai said: "Physical AI needs real-world data that is accurate, affordable and available at scale. Meeting all three requirements has been a persistent problem."5 At the launch Hesai announced partnerships with robotics companies Galbot and Dexmal, and showed applications built with Pixar's RenderMan and with Amap, covering robotics, filmmaking and location services.5
Note what this is: a launch and a set of partnerships. The dossier contains no sales figures, customer counts or independent tests. The related MIT Technology Review newsletter on robotics (8 October) carries the headline "AI breakthroughs in robotics won't change your life any time soon". Our checks gave that publication a 0% record, but only 11 claims were checked, a sample too small to judge it by.6 We cite it only to show the cautious view exists.
Software that acts on its own
At the software layer, TELUS Digital said on 14 October it is partnering with Parloa, an AI customer-service platform founded in Germany. TELUS becomes an implementation partner, helping businesses set up and run Parloa's AI agents.7 Chris Silver said: "The research confirms what we've all experienced ourselves — consumer expectations are high, and brand loyalty is fragile."7
Zeta, at its Zeta Live 2026 event on 8 October, introduced AthenaOS. David A. Steinberg said it "starts with what you want to achieve" and described it as an expansion of Zeta's AI infrastructure beyond marketing.8
There is some demand-side evidence. A Snipp survey of 1,000 U.S. adults who regularly use AI for search and shopping found that 61.8% had purchased after an AI recommendation. It also found 41.6% start product searches with AI.9 The sample is people who already use AI regularly, so it says nothing about the general public. And the survey was published by Snipp itself.
The governance question
Commercial momentum is only half of the picture. The framing of this story includes OpenAI's safety record, and the evidence we can check is thin. We have two quotes from Mark Chen, OpenAI's chief research officer, in an MIT Technology Review item about the company's response to a hack. 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
That is one executive defending his employer. The dossier contains no source for a safety-lead resignation, for OpenAI's monitoring or open-source pledges, for Acer's warning about Micron memory prices, or for Samsung's record profit. We have therefore left those out rather than repeat them unchecked. They are worth following, but a reader should know we have not verified them here.
What to watch
- Nvidia's next filing: whether revenue keeps pace with costs that have roughly doubled twice, and whether cash keeps falling from its Q1 level.
- Cisco's shipments: the G300 is said to be shipping. Follow-up customer announcements would turn a launch into evidence.
- Hesai's partners: any named deployments from Galbot or Dexmal, which would show Kosmo moving past a launch event.
- Agent platforms: any customer results from TELUS Digital and Parloa, since a partnership announcement proves intent, not outcomes.
