Daily Briefing
25 August 2026 · 5 stories
🔥 Top story
1NVIDIA launches an ultra-fast AI chip — Nebius is the first customer
Imagine bolting a Formula 1 engine into a regular city car: that's roughly the idea behind NVIDIA's new Groq 3 LPX chip, unveiled this week at the Hot Chips conference. Its specialty: answering a user in real time, at a frankly unreasonable speed — 3,400 words per second on a mid-size model. To put that in perspective: that's faster than the speech rate of a TV news anchor. Nebius, which just raised 4.3 billion dollars, is the first to install it in its data centers. In practical terms, when you chat with an AI agent in a few months, there's a good chance the answer was computed by this chip. NVIDIA claims 35 times more energy efficiency than its previous generation. Source: https://nvidianews.nvidia.com/news/groq-3-lpx-full-production-hot-chips-2026-08-24
2Meta is preparing its own AI agent for 3 billion users — launch imminent
Three billion: that's the number of people who use Facebook, Instagram, WhatsApp or Messenger every month. And Meta is about to add a personal AI assistant to all of them, codenamed « Hatch », that can book a table, compare prices or fill out forms on your behalf. The information was revealed Monday evening by The Information. To train it, Meta first used Claude (Anthropic's AI), then will switch to its own model, Muse Spark. The test phase consists of simulating real websites — DoorDash, Etsy, Reddit, Yelp — to train the agent not to mess up in real-world conditions. If Meta pulls it off, it will be the first time such a powerful agent reaches that many people at once. Source: https://www.theinformation.com/articles/meta-hatch-ai-agent-coming-weeks
3Alibaba raises 10 billion dollars for AI — China opens the cheque book
Ten billion dollars raised in a single day: that's what Alibaba pulled off on Sunday in Hong Kong, the largest placement in the history of the Hong Kong stock exchange and the third biggest worldwide in 2026. To give a sense of scale: that's more money than it takes to buy 100 Airbus A320s. Alibaba promises to invest everything in AI — its own chips, its data centers, its models. Banks Morgan Stanley, HSBC and UBS led the operation, which was oversubscribed by sovereign funds. On the stock side, Alibaba shares still lost 10% at Monday's open — investors are digesting the fact that each share is now worth a little less, because there are more of them. For ordinary people: China is taking a head start, and the others will have no choice but to follow. Source: https://www.reuters.com/business/alibaba-10-2b-hk-share-placement-ai-2026-08-23
4The United States is about to tax Chinese goods more heavily — AI in the background
Washington is preparing a new 7.5% tariff on Chinese products, according to Bloomberg and Reuters. The official reason: Chinese « overcapacity », meaning the fact that China produces more than it consumes, flooding world markets at low prices. In practice, this tariff stacks on top of those already in place and could push the total US duties on China to the 20% ceiling. The measure is expected before the summit between Xi Jinping and Donald Trump scheduled for September. Worth noting: Reuters could not confirm the information independently, so take it with caution. For European companies that depend on Chinese components or sell to both markets, this is one more reason to plan for diversified supply chains. Source: https://www.bloomberg.com/news/articles/2026-08-24-us-tariffs-china-overcapacity
5DeepSeek beefs up its AI and cuts its prices — 14 times cheaper than US rivals
Chinese lab DeepSeek made official this week the « stable » version of its V4-Pro model, after 4 months of testing, and launched in parallel a multimodal variant capable of understanding images (V4-Flash-Vision-Exp). Above all, it introduced a new pricing scheme: cheaper at night than during the day, like the electricity at your home. Result: paying for a DeepSeek response costs about 3.50 dollars per million words versus 50 dollars at some US competitors — 14 times less. The model natively speaks the same « technical dialect » as OpenAI, which lets companies switch from one to the other without rewriting everything. For European software publishers who want a high-performing AI without depending on the United States, it's now a credible and affordable option. Source: https://api-docs.deepseek.com/news/deepseek-v4-pro-ga-vision-exp
📡 To watch
NVIDIA Groq 3 LPX chip: who will be the next to adopt it?
Nebius is the first customer, but the chip now has to convince AWS, Google Cloud or Oracle. The first benchmarks against Cerebras, Trainium 3 and Google's TPUs will set the tone in the coming weeks.
Meta Hatch: official confirmation expected this week
Meta hasn't communicated anything officially yet. If the rumor is confirmed, we'll discover at the same time the pricing grid of the Muse Spark model — including a « contributor » rate that could be up to 21 times cheaper than the standard rate for certain tasks.
Alibaba: where will the 10 billion really go?
The placement is to be finalized this week. The split between chips, data centers and models will be a strong signal for the whole sector — watch for announcements from the banks that led the operation (Morgan Stanley, HSBC, UBS, CICC).
GLM-5.3 open access: it's coming very soon
The Z.ai lab promised the release of the open weights (the internal engine of the AI) around August 28. If the license really is MIT — meaning truly free — it's a small revolution for European users who want a high-performing AI without depending on the GAFAM.
📊 Trend
This week is dominated by the big moves: 10 billion raised by Alibaba on Sunday, an AI chip 35 times more efficient at NVIDIA, and a Meta agent about to land on 3 billion users. Three forces are converging — money is flowing into AI, specialized machines are replacing general-purpose servers, and autonomous agents are leaving the labs to settle into everyday life. The big question for the next twelve months is now crystal clear: who — America, China or Europe — will set the rules of the game for these assistants that are about to become our digital colleagues.