In partnership with
Reading time: 6 minutes
🗞️ In this edition
Alibaba Stopped Waiting For Nvidia
Sponsored: HubSpot for Startups
A Province Sued OpenAI Over A Call It Never Made
Finland Minted An AI Cloud Unicorn
⚡ YOUR MOVE For people building the life they want, now.
In other AI news:
Ligent Popped 19%, Then Gave Most Of It Back
OpenAI Wants Washington To Write The Rulebook
Apple Wants A Look Inside OpenAI's Hardware Lab
4 must-try AI tools
Alibaba named the size of its next model, and it is bigger than anything it runs today. A Canadian province says OpenAI saw the warning and never called the police. And a Helsinki company most people have never heard of just crossed a billion dollars.
Keep reading.
Source: technode.com
WHAT HAPPENED
At its Apsara Conference, Alibaba's chip unit T-Head showed the Zhenwu V900, which it says delivers three times the performance of the Zhenwu M890 before it, with 216GB of memory, 1.2TB/s of chip-to-chip bandwidth and native FP8 and FP4 support. Mass production is planned for the first quarter of 2027. Chief executive Eddie Wu called it China's most powerful AI chip and said Alibaba plans to train a model with between 5 and 10 trillion parameters. Alibaba Cloud is targeting more than 20 gigawatts of data centre capacity by 2032 and clusters of up to 500,000 accelerators, while keeping its capex guidance at RMB 380 billion. Its Hong Kong shares rose about 2% on Tuesday and its US listing gained about 3% before the open.
WHY IT MATTERS
Alibaba now designs the chip, runs the cloud, trains the model and sells the answers. When one company is its own biggest customer, it does not need the market to believe a chip is good. It only needs a workload big enough to fill it, and a 10 trillion parameter model is exactly that. Export controls were built to stop China from buying the best hardware. They have much less to hold onto when the buyer and the seller sit in the same building.
OUR TAKE
The obvious read is that China finally has its own Nvidia. We think the model is the real announcement, and the chip is along for the ride. A 10 trillion parameter target is an order Alibaba is placing with itself, a guarantee of demand for a part nobody outside the company has tested.
Watch two things. If no independent benchmark of the V900 appears before mass production starts in early 2027, treat the "three times" as a sales number. And if the RMB 380 billion capex guidance goes up at the next results, the model was the budget request.
From Our Sponsor:
Watch the full system run live on Sep 30. Walk away ready to do it yourself.
Most founders have LinkedIn traction with nothing to show for it in the CRM. On Sep 30, Maria Gharib (Mindstream) and Valerie Chapman (Ruth AI) walk through the exact system live.
From AI-assisted content creation to sequenced outreach to booked meeting. You'll leave with a process you can run the same day.
Eligible startups also get the LinkedIn-to-Leads Toolkit: ad credits, Apollo, Captions, and HubSpot's Prospecting Agent.
Source: aljazeera.com
The province says OpenAI's own reviewers flagged the Tumbler Ridge shooter's ChatGPT account for threats of real world violence in 2025, banned it, and never told police, and that the shooter then opened a second account. The February 10 attack killed eight people. B.C. wants compensation for emergency response and recovery costs, money toward a new school, and court ordered changes to how OpenAI handles violent threats. Premier David Eby called Altman's April apology "necessary, and yet grossly insufficient." More than 30 other suits have been filed by families and survivors.
How did today's stories land?
Source: tech.eu
The oversubscribed Series B was led by Emergence Capital, with MUFG Innovation Partners, Supermicro, the Finnish pension insurer Varma, Lifeline Ventures and the state backed Tesi among those joining. Founded in 2020, Verda runs the whole stack itself, from physical data centres and hardware up to cloud services and research. It says it hit a $165 million annualised revenue run rate in July, has now raised more than $450 million in total, and just opened offices in London and San Francisco.
⚡️ YOUR MOVE
FOR PEOPLE BUILDING THE LIFE THEY WANT, NOW.
🧠 Xiaomi's MiMo-V2.6-Pro is a 1.02 trillion parameter open model under MIT, weights on Hugging Face, at $0.435 per million tokens in and $0.87 out. It ties Grok 4.7 on one major intelligence index. → Grab the weights
⚙️ Grok 4.7 keeps the same $2 in and $6 out as 4.6 but sits on a bigger base model, with a 500K window and Terminal-Bench up from 20.3% to 38%. It is already in Cursor and OpenRouter. → Run the numbers
💻 Google's first Googlebooks start at $899 from Acer, ASUS, Dell, HP and Lenovo, ship October 4 in the US, and come with a year of Google AI Pro and ten years of updates. One feature changes what a cursor is for. → See the lineup
🔎 In Norway, the share of 9 to 18 year olds using Google fell from 72% to 47% in two years. If your customers are young, they may never type your name into a search box. → Rethink your funnel
Ligent Popped 19%, Then Gave Most Of It Back – The Hisense backed optical transceiver maker raised HK$5.67 billion in Hong Kong, touched HK$39.30 on debut and closed at HK$34.48, up 4.6%.
OpenAI Wants Washington To Write The Rulebook – A new OpenAI post asks the US to lead global technical standards for frontier AI, recursive self improvement included, days before Trump and Xi meet.
Apple Wants A Look Inside OpenAI's Hardware Lab – In its trade secrets case, Apple is asking for OpenAI documents on metal finishes, power management and batteries, plus forensic images of former employees' devices.
Zapier: Connects the apps you already pay for so the busywork runs itself.
Grammarly: Catches the sentence that would have made the client email awkward.
ElevenLabs: Turns your script into a voiceover without booking a studio.
Runway Gen-4: Makes the b-roll you were about to buy from a stock site.
Three stories about what scale owes. Alibaba is promising a model far bigger than anything it runs, on a chip nobody outside has tested. British Columbia says OpenAI already knew enough to act, and did not. Verda raised on the bet that Europe will pay for compute it can control. Building big is the easy part now. The hard part is deciding who answers for it.
Ask who answers for it.
Interested in featuring your services with us? Email us at [email protected]
