45k user reviews & 1 million votes, over 29k web hosting brands & 85k plans since 2004!
Menu
WHTop → 📋 News → Google Splits TPU Roadmap to Challenge NVIDIA In AI Race
Apr, 2026 : Google Splits TPU Roadmap to Challenge NVIDIA In AI Race
📅 - Google is splitting its latest AI chip strategy in two, unveiling eighth-generation TPUs built separately for training and inference as pressure from NVIDIA and other rivals intensifies. At Cloud Next, the company framed TPU 8t and TPU 8i as infrastructure for the agent era - where models do more than answer prompts, instead reasoning, coordinating, and acting continuously.
For years, Google's TPU story was relatively straightforward. Build custom silicon, use it internally for systems like Gemini, and offer it to cloud customers as an alternative to NVIDIA-heavy infrastructure. Now the company is making a sharper distinction. Training large models and serving them in production are no [...][... Check source for end of article ...]
Want to add a website news or press release ? Just do it, it's free! Use add web hosting news!
Related news
📅 - Google Lands Nuvem Cable, Expanding US-Europe Cloud Capacity - Google has completed the Nuvem subsea cable landing in Sines, Portugal, adding a high capacity route between Europe and the United States as cloud and AI traffic accelerates. The 7,000 kilometer system links Portugal with South Carolina through the Azores and Bermuda, carrying up to 384 terabits per second across 16 fiber pairs at full design capacity.
For infrastructure buyers, the immediate change is not simply another line on a network map. Nuvem gives Google more control over a route carrying traffic between its American and European cloud regions, adding redundancy across the Atlantic. Customers may eventually see better resilience, latency or availability. None of that is automatic. [...]
📅 - Deloitte Opens London AI Studio With Google Cloud - Deloitte has launched an AI Studio in London with Google Cloud, aiming to help UK organizations move agentic AI projects from prototypes into production. The facility, opening in late July, will support clients in public services, finance, retail, healthcare, life sciences, and TMT, while Deloitte trains 1,000 UK AI and data specialists on Gemini Enterprise.
The announcement lands in a market that has started to tire of AI demos. Enterprises have spent the past two years testing copilots, chat interfaces, internal search tools, and productivity assistants. Useful in places. Underwhelming in others. Agentic AI is being sold as the next stage: systems that do not merely answer questions but [...]
📅 - Google’s $920M-A-Month SpaceX AI Deal Prices GPU Scarcity - Google has signed a contract to rent AI compute from SpaceX for $920 million a month, according to a regulatory filing tied to SpaceX's expected IPO. The agreement runs from October 2026 to June 2029, covers about 110,000 NVIDIA GPUs, and puts a near-$30 billion price on access to scarce infrastructure over time for model workloads globally.
The interesting part is not only that Google wants more AI capacity. It is that Google, of all companies, is prepared to buy it from SpaceX.
Google has its own cloud, its own data centers, its own networking, its own AI teams and its own custom AI chips. It is not a startup standing outside the GPU supply chain with a purchase order and a dream. Yet [...]
📅 - Telefónica Anchors Google Cloud Sovereignty Service in Spain - Google Cloud and Telefónica have introduced a sovereign cloud service for Spain, pairing Google's Madrid region with Telefónica-managed encryption keys and local operational controls for regulated buyers. The arrangement gives public agencies and enterprises another way to use hyperscale cloud while keeping data residency, personnel access and key management under Spanish jurisdictional oversight pressure now directly visible.
The deal is not Google building a separate Spanish cloud. Not quite. It is a sovereignty overlay on Google Cloud's local infrastructure, with Telefónica Tech acting as the trusted local operator for encryption and control functions.
That distinction will matter to [...]
📅 - Thales and Google Cloud Plan Sovereign Cloud in Germany - Thales and Google Cloud plan to launch a sovereign cloud service in Germany, with a locally controlled Thales entity operating infrastructure designed to keep sensitive workloads beyond non-European access. The preview is open now, general availability is expected by late 2026, and the offer targets regulated enterprises, public agencies, healthcare, finance, and critical infrastructure buyers in Europe.
The move is not really about another cloud region. Germany has those. It is about who operates the cloud, who can touch the data, what law applies, and whether a regulated organization can use hyperscale services without creating a sovereignty problem in the risk register.
📅 - Google is splitting its latest AI chip strategy in two, unveiling eighth-generation TPUs built separately for training and inference as pressure from NVIDIA and other rivals intensifies. At Cloud Next, the company framed TPU 8t and TPU 8i as infrastructure for the agent era - where models do more than answer prompts, instead reasoning, coordinating, and acting continuously.
Reads: 53 | Category: General | Source: Hosting Jurnalist : Hosting Jurnalist | Author: