The GPU tap is still very much on
Nvidia has raised its forecast for AI chip sales next year, expecting revenue to grow around 70% as demand for AI infrastructure stays strong.
The company made $96.2bn in revenue last quarter and expects about $108bn this quarter, both above Wall Street forecasts.
Amazon Web Services has also agreed to deploy another 2 million Nvidia GPUs.
But investors are looking more closely at how Nvidia is helping fund some of its own customers.
The company has invested in major AI firms including OpenAI and Anthropic, while also backing financing deals that help customers build data centres and buy more chips.
Critics say this could create a circular system where Nvidia helps fund customers that then spend money on Nvidia products.
Nvidia disagrees, saying demand remains strong and the risks are limited.
Its data centre business brought in $89bn last quarter, up 117% from a year earlier.
There are still some concerns. Nvidia expects profit margins to fall as memory shortages push up costs, while free cash flow dropped as some large customers took longer to pay.
In brief:
- Nvidia expects sales to grow around 70% next year.
- Data centre revenue reached $89bn, up 117% year on year.
- Supplier commitments jumped to $279bn, mainly to secure more memory chips.
The numbers are deeply unserious
Nvidia’s results show AI infrastructure spending is still growing fast, but investors are watching how much of that demand Nvidia is helping to finance itself.
Google’s AI naming problem
Google says Gemini should make AI easier by letting users ask for what they need without choosing between different tools.
But Gemini still has separate features like Chat, Spark and Daily Brief.
Daily Brief pulls updates from Gmail, Calendar and past activity.
The issue is that it can also bring up old searches or unfinished research that users may not want resurfaced.
Spark can take actions for users, but it is still treated as a separate mode. Ideally, people would just make a request and Gemini would decide how to handle it.
Here’s what you should know:
- Gemini still has several separate AI modes.
- Other AI companies are doing the same.
- The simpler the interface, the easier AI may be to use.
Please select your preferred flavour of intelligence
Google is not alone. Claude and ChatGPT also split their AI tools into different modes, which can make them harder to use.
Apple is taking a simpler approach by adding AI to tools people already know, like Siri, Photos and Spotlight.
Text-based AI assistants follow the same idea: send a message and let the AI work out the rest.
Anthropic releases a standard that lets AI agents safely control lab hardware
Connecting AI to physical machines has always been painful. Every device needs custom code, and integration takes weeks. Anthropic just shipped a fix.
Meet Model Hardware Standard (MHS): a shared interface so AI agents can talk to physical equipment without custom glue code for every device. Think of it like USB-C, but for AI talking to machines.
Real results from early testing:
- Drug-discovery experiment at Genentech, with live error handling
- Imaging experiment compressed from weeks to one day at HHMI Janelia
- Laser stabilization on quantum computers jumped from 58% to 99.3% at QuEra
It works with any device that has a programmable interface, and it is model-agnostic, so you are not locked into Claude. It runs on top of MCP, the same open protocol Anthropic released for connecting AI to data sources.
You can apply for the research preview now. Open source is coming later.
Z.ai drops open weights for GLM-5.3 tomorrow
Z.ai’s GLM-5.3 weights are dropping on Hugging Face today, and this is a big deal if you want to run a powerful coding model on your own hardware.
Here’s the quick backstory. GLM-5.3 launched on the API two weeks ago. Z.ai held the weights back for safety review, specifically because GLM-5.3 is their most capable model for cybersecurity tasks, with big jumps in vulnerability discovery and complex multistep security work. So they staged the release carefully.
What you’re actually getting:
- 1M token context window with up to 128K output tokens — huge for long coding sessions
- Same 743B-parameter base as GLM-5.2, with all gains coming from extra post-training
- Needs at least an 8x H200 node to self-host the flagship
- Always runs with reasoning on, with three effort levels: low, high, and max
The license has not been confirmed yet, so don’t assume GLM-5.2’s terms carry over. Check before you ship anything.
Google releases a lightweight model that predicts blood sugar patterns and diabetes risk
Blood sugar data is messy. Glucose sensors track your levels every few minutes, all day, but the readings mix together two very different signals: your slow background baseline and sudden spikes after a meal. Previous AI models just threw all of that into one stream and hoped for the best.
GlucoFM fixes that with a smarter approach. It uses two separate processing lanes, one for slow trends, one for short bursts, so the model actually understands what is happening instead of guessing.
Here is what makes it worth paying attention to:
- Trained on 109,066 hours of real glucose data from 477 people, with no labels needed
- Beats the best existing glucose AI models by 4.1 points on a standard accuracy measure
- Works across diabetes risk, insulin resistance, and post-meal response prediction
- Transfers well to new datasets it has never seen before
This opens the door to building health apps that actually understand metabolic patterns from wearable data, without needing expensive clinical labels to train on.
Nvidia Pauses Revenue-Sharing Deals
Nvidia has paused its revenue-sharing program with AI companies, according to the Wall Street Journal. The move signals a strategic shift as Nvidia re-evaluates how it partners with AI startups amid surging demand for its chips and growing market dominance.
Claude Now Controls Lab Equipment
Anthropic is testing a system that lets Claude operate physical scientific instruments and industrial robots. Early results include QuEra’s Claude fixing quantum computer lasers overnight — reducing human expert time from 5-10 minutes to 6 seconds with 99.3% success rate.
Tech Giants Warn of AI Cyberattacks
OpenAI, Anthropic, Google, Microsoft, and 100+ other companies signed an open letter warning governments that AI-enabled cyberattacks are about to become far more widespread and sophisticated. The coalition called for urgent global cooperation on AI security defenses.
Salesforce Surges 22% on Strong Earnings
Salesforce shares rocketed 22% in its second-best trading day ever, leading a broad software rally. The surge came after the company reported stronger-than-expected quarterly results, driven by AI-powered product adoption and robust enterprise spending.
Nvidia to Acquire Hugging Face for 12.9B
Nvidia is reportedly acquiring AI model hub Hugging Face for 12.9 billion. The deal would bring the industry’s most popular open-source model repository under Nvidia’s umbrella, strengthening its AI ecosystem play against cloud hyperscalers.
Musk Predicts SpaceX .5T Revenue by 2033
Elon Musk predicted SpaceX could reach .5 trillion in annual revenue by around 2033, driven by Starship launches, Starlink expansion, and deep-space missions. The projection would make SpaceX revenue roughly 11.5x Nvidia’s current trailing twelve months.
OpenAI Renews Bid to Toss Apple Suit
OpenAI renewed its push to dismiss Apple’s trade secret theft lawsuit, arguing the claims lack merit and that the tech giant is using litigation to gain competitive advantage. The legal battle stems from allegations that OpenAI poached Apple engineers and used proprietary technology.
Socure Raises 56M at .2B Valuation
Digital identity verification company Socure raised 56 million at a .2 billion valuation and acquired AI startup Fravity. The round underscores continued investor appetite for AI-powered fraud prevention as digital identity becomes critical infrastructure.
AI Assistant Startup Instinct Raising 50M
Consumer-focused AI assistant startup Instinct is reportedly raising 50 million, signaling strong venture interest in the AI assistant market. The company is building a general-purpose AI companion that competes with offerings from major tech platforms.
Chinese GLM-5.3 Matches Top Models Without Nvidia
China’s GLM-5.3-Flash matches top-tier AI models at a fraction of the cost and runs without Nvidia hardware. The model demonstrates China’s growing ability to compete in AI without access to advanced Western chips, raising questions about the effectiveness of export controls.
IREN Reports B Contracted AI Cloud ARR
IREN reported FY26 results with billion in contracted AI Cloud annual recurring revenue. The data center firm secured a multi-year contract with a leading frontier AI lab, achieved Nvidia Exemplar Cloud status on GB300 NVL72, and is executing global expansion across 5GW+ pipeline.
EPA Sued Over Fast-Tracked Data Center Chemicals
The EPA faces a lawsuit over claims it fast-tracked approval of toxic chemicals used in semiconductor manufacturing and data centers. The chemicals, used as photoacid generators, may cause sudden death from acute exposure and are long-lasting PFAS compounds.
Half-Price Starlink Near Starbase Louisiana
Elon Musk announced half-price Starlink for residents near SpaceX’s Starbase Louisiana facility. The discount aims to build goodwill in the community as SpaceX expands its Starship production and launch operations in the region.
Tesla FSD Keeps Mistaking Trains for Roads
Tesla’s Full Self-Driving system continues to have trouble with trains, attempting to drive onto railroad tracks. The recurring issue highlights persistent edge-case challenges for vision-only autonomous driving systems.