Elon announced that xAI Grok 3.5 will be released soonish. This likely means May.
We know that at the February Grok 3 announcement xAI had already installed another 100,000 H100/H200s to go with the 100,000 H100s that trained Grok 3.
I think xaI started training in February for Grok 3.5. They will likely use about 60 day or so of training. This will have 2.5X-3X the compute. Grok 3 likely had more days to get training. Grok 3 had 5X the number of 20K chips used for Grok 2. They claimed nearly 15X the compute which would mean about 2.5X the days of training.
Grok 3.5 will probably average 250K chips with the first batches of installed B200s. It took 92 days to install the last 100K GPU chips. It will take about 70 days to install the first 100K B200s. This means completing installation of the first 100K B200s at the end of March or April if there was some overlap with the H200s.
This rapid installation of AI GPUs is a major competitive advantage. Nvidia CEO Jensen Huang has said that xAI and Tesla are leaders and superhuman in their ability to rapidly install large numbers of chips in data centers.
The other major Tesla and xAI advantage is recreating the third and fourth layer of the ethernet protocol into a hardware solution for microsecond instead of millisecond communication. This lets the data center have unified (coherent) memory which is needed for large scale AI training for more than 30,000 GPUs. xAI has the rapid communication so they can directly scale to millions of GPUs.
The Grok v3.5 release leverages their installation-build and fast communication/scaling advantage to get into the lead again. This will be focused on AI for voice, video and coding.
ELON: GROK 3.5 SOON, GROK 4 LATER THIS YEAR
While livestreaming Path of Exile using Starlink, Elon revealed the upcoming Grok roadmap:
“Grok 4, later this year.
We’ve got Grok 3.5 coming out soonish.
That’ll be a lot of significant upgrades.”
Source: @elonmusk https://t.co/uWK9o7KJLk pic.twitter.com/fVWbxqhupP
— Mario Nawfal (@MarioNawfal) April 10, 2025
I have updated my table of expected AI releases with this new information.
This v3.5 release strategy also suggests there will be major interim releases like version 5.5 and other half point releases. I am expecting less of a leap in training compute with version 5. Just scaling to three times the number of chips needs to combined with either major algorithmic improvement or a new generation of chips to justify a half point release.
The half point release strategy and continual reinforcement learning also indicates that xAI is getting closer to continual improvement of models.
xAI has gotten the permits to install fifteen 16 Megawatt mobile natural gas turbines. They would need to have seven or eight installed to power another 100,000 Nvidia B200 chips.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
