The AI news cycle rarely slows down, but the last few days have been especially dizzying — swinging from genuine technical leaps to a bizarre branding fight in Washington.

Google doubles down, twice

Just two weeks after shipping Gemini 4 Argon, Google DeepMind pushed out Gemini 4 Carbon, and it's not a minor point release. Just two weeks after Gemini 4 Argon, Google DeepMind is releasing Gemini 4 Carbon, which tops the Artificial Analysis Intelligence Index v5 with 59 points. That's a meaningful jump from Argon's score, and Google is positioning the model specifically for extended agent workflows and the full cyber defense lifecycle. Rather than a wide public launch, Google is rolling out the model gradually and participating in the U.S. government's pre-release testing process, citing Carbon's advanced security capabilities, with selected cybersecurity experts getting access first. Pricing is aggressive for a frontier-tier model, reportedly starting around $2 per 1 million input tokens and $10 per 1 million output tokens before rising to standard rates.

On the research side, DeepMind also previewed Genie 4, described as a real-time interactive world model that generates photorealistic scenes with improved physics and native audio produced in the same forward pass as the visuals. It's only available to trusted testers for now, but it signals DeepMind's continued push toward models that simulate environments rather than just describe them — a capability with obvious implications for robotics, gaming, and synthetic training data.

Washington's identity crisis

Meanwhile, the federal government is having a strange week of its own. President Trump said the government will move away from the term 'AI,' with reporting indicating the U.S. will officially be banning the term 'AI/Artificial Intelligence' in relevant government documents, wanting 'Super Intelligence,' or SI, used instead. The irony is hard to miss: the federal government is simultaneously increasing its use of the technology Trump wants renamed, with Anthropic announcing a broader GSA agreement the same day giving government organizations discounted access to Claude. In other words, agencies may soon be buying more of the thing they're no longer allowed to call by its name.

Infrastructure consolidates around neutrality

At Oracle AI World in Las Vegas, the infrastructure story got more interesting too. Oracle co-CEO Clay Magouyrk and Sam Altman announced that ChatGPT and Codex will be embedded in Oracle's business applications and database products, while Oracle Cloud Infrastructure will add Google TPUs, OpenAI's custom Jalapeño inference chip, and Cerebras systems — marking the first time Google TPUs run on another hyperscaler. That kind of multi-vendor chip strategy suggests cloud providers are hedging hard against any single compute bottleneck, even as rivals' chips end up sitting side by side.

Taken together, the week captures where AI actually is right now: capability gains arriving almost biweekly, infrastructure players scrambling to stay neutral and well-stocked, and political institutions still arguing about what to even call the thing everyone is racing to build.

Sources

AI InfrastructureAI PolicyArtificial IntelligenceDaily AI UpdateGeminiGoogle DeepMind