The short version, and a caution about the evidence
Between 20 October and 6 November 2026, a run of announcements moved the AI conversation away from what models can do and toward what it takes to run them: securing them, feeding them data and putting them on hardware. Two caveats frame everything below. First, this is thin on Europe. The dossier we drew on names no EU regulator, rule or company, and the only European-relevant hook is a phrase in Fortinet's announcement about "sovereignty requirements".1 Second, almost every source is a company press release. Our own measurement of the wire service carrying most of them found that 54% of 6,613 checked claims held up.1 The one non-press-release source, Crunchbase News, held up on 49% of 41 checked claims, a smaller sample.4 We flag which statements are checkable arithmetic and which are management assertion.
Securing AI: the European hook is sovereignty
On 26 October Fortinet launched the FortiGuard Inline AI Security Service, which adds AI-specific inspection and guardrails to its FortiGate firewalls through a cloud service.1 Michael Xie said: "Organizations need AI security that can scale with how quickly AI is being adopted across the enterprise... By extending advanced AI inspection and guardrails through FortiGate, we are making AI security available through the infrastructure customers already use."1 The part most relevant to a European buyer is the claim that FortiAIGate lets customers choose how to deploy and consume AI security based on cost, performance, privacy and sovereignty requirements.1 That is a vendor's description of a product feature. The dossier does not say how, or whether, it meets any specific EU rule, so we cannot say it does.
T-Mobile offers the one claimed result from the security side. In a report published on 2 November, Jeff Simon said: "Today, phishing-related credential compromises among our people are down more than 99%."2 The report also covers what held up when tested, including Bug Bounty program results, and says the teams skipped pilots and multi-year rollouts.2 This is T-Mobile's own account of its own programme. The dossier has no independent audit of the 99% figure. T-Mobile did publish it, which makes it checkable in principle, and the company says it is "showing its work" because trust "has to be earned and verifiable".2
Cisco's angle is distribution. It said on 27 October that Cisco IQ is generally available to partners and that expanded Cisco Cloud Control offerings give them new ways to build, secure and manage AI-ready infrastructure.3 Tim Coogan said: "Our partners are where Cisco meets the customer, and the AI era is asking more of all of us."3 In plain terms, the AI market is being sold through the resellers and integrators who already hold the customer relationship.
Physical AI: the data problem, in the industry's own words
Crunchbase data shows global venture funding to robotics startups of $9.9 billion in Q3 2026, more than double the $4.3 billion of Q3 2025, with most of it going to companies based in China and the US.4 The funding is large. The more revealing comment comes from Ulrik Stig Hansen of Encord, who described the obstacle as data: "Physical AI is very different because there is no internet of physical AI data, and so the data first has to be collected, and you have to convert it from sensor tokens to action tokens."4 He added that "storing all that data, curating all that data, enriching all that data, and validating that all the data works for these systems is a very expensive operation", and said the industry is still in a pre-training phase.4 Encord sells in this market, so he has an interest in the problem being seen as large.
The chip side makes the same point from the hardware end. On 22 October at the 9th China Robot Summit in Hangzhou, Harvest Praxis launched its "Hyper AI Brain" platform, built on MediaTek's 3nm Genio Pro 5100 chip. It claims the first fusion of a robot's perception and decision-making "brain" and motion-control "cerebellum" on a single chip.5 Founder Philip Cai Zhenhui said: "The industrialization of embodied intelligence relies on three critical pillars: large models, real-world data, and on-device computing."5 The "world's first" claim comes from the company and we have not verified it.
Edge hardware: consumer devices and chips
Apple unveiled a new iPad mini on 26 October with an 8.5-inch OLED display and an A20 Pro chip.6 Pre-orders opened on 28 October and general availability followed on 6 November.6 The dossier carries no detail on NVIDIA's RTX Spark laptops, so we do not describe them. In the data we do have, the clearest sign of real demand for edge silicon comes from Impinj, which makes RFID chips for tracking items. It reported record third-quarter revenue of $112.1 million, with record chip revenue and unit volumes.7 CEO Chris Diorio credited the M800 chip, resilient retail demand and growth at its "lighthouse accounts" (its key customers).7 Impinj guided Q4 revenue to $104.0 million to $107.0 million.7 That is below the Q3 record, so even the strongest hardware result here comes with a guided step down.
Consolidation and a cautionary small-cap
On 20 October Salesforce agreed to acquire Braze for about $5.66 billion in enterprise value, paying $47.50 in cash per Braze Class A share.8 Braze helps marketers coordinate messages across apps, email, SMS and WhatsApp.8 Braze CEO Bill Magnuson said the company "shipped at the fastest pace in our history" last year and that Salesforce shares its commitment to marketers.8 For scale, the price is roughly 8,800 times the $640,000 quarterly revenue of the small company discussed next. That is our arithmetic, not a figure from either release. The deal is a bet that AI-driven customer engagement will be bought as a platform, but the dossier gives no revenue or growth figures for Braze, so we cannot judge the price.
TEN Holdings (Nasdaq: XHLD) shows how different the other end of the market looks. Its third-quarter revenue rose 17.9% to $640,000, according to CEO Virgilio Torres.9 The quarter also carried a $1.8 million write-off of capitalised service contracts and about $0.5 million of legal expenses.9 The company says it cut annual operating costs by about $2.5 million through cost reductions, process improvements and AI-driven tools. It also says short-term related-party borrowings fell from $4.6 million to about $0.1 million, and total liabilities fell about 74%, from $7.7 million to $2.0 million.9 A 17.9% rise on a base this small is not evidence of scale, and the one-off charges exceed the quarter's revenue. The balance sheet has been cleaned up, but the dossier does not show a business that has yet proven its software model.
What to watch
- Whether Fortinet or others spell out how their "sovereignty" options map onto EU requirements. The dossier is silent on this.
- Whether T-Mobile's 99% phishing figure is independently tested.
- Impinj's actual Q4 result against its $104.0–107.0 million guidance.
- Whether the Braze deal's terms are followed by a regulatory process. The dossier contains none.
- Whether TEN's next quarter shows revenue growing without the one-off charges.
