Alphabet, the parent company of Google, is developing a new server chip aimed at enhancing the efficiency of its proprietary Gemini models.

The upcoming chip, referred to internally as “Frozen v2,” is expected to launch sometime in 2028, according to a report by The Information, which cites unnamed sources. The report indicates that this chip could achieve efficiency levels between six and ten times higher than Google’s current AI chips, based on the tokens produced per energy unit.

In a statement to TechCrunch, the company did not explicitly confirm the report. It also did not refute it.

“Our teams are continually researching and testing new innovations to provide optimal performance and efficiency for our users and clients,” Google informed TechCrunch. “Though not every initiative advances to production, this thorough investigation is fundamental to our comprehensive approach. By co-developing our hardware and software from inception, we guarantee our systems are integrated and highly optimized for practical workloads.”

AI firms are increasingly attempting to manufacture their own chips to enhance the operation of their internal models and to tackle global shortages in AI computing power. This efficiency has become a crucial selling point for technology companies as worries about AI expenditures have tempered the previously exuberant market atmosphere. Concurrently, companies are striving to reduce their reliance on chipmaker Nvidia, which has historically held a dominant position in the AI chip sector, leaving major AI developers dependent on its technology.

In June, OpenAI unveiled its inaugural custom chip, an inference processor named Jalapeño. Recently, reports indicated that Anthropic is in talks for a new chipmaking collaboration with Samsung.

Investors have previously expressed concerns regarding Alphabet’s substantial planned investments aimed at advancing its AI strategy. Earlier this year, Google announced plans to invest between $180 billion and $190 billion. With significant funds involved, the company needs to demonstrate that these investments will yield positive returns.

News regarding the more efficient Frozen v2 chip seems to have calmed investors, providing Google with momentum ahead of its earning report scheduled for later this week. Following The Information’s report, the company’s stock rose approximately 3% on Monday morning.

The Model Context Protocol (MCP) serves as a fundamental component of AI interoperability, offering AI models a safe means to access external data resources and services. It acts as the infrastructure that allows a chatbot to tap into your calendar, your database, or your internal applications, eliminating the need for engineers to create custom solutions for each connection. An important update to this protocol is set to roll out next week; although it may not be evident to end users, it could significantly impact the development of the ecosystem.

The official specifications for the latest version have been accessible since May, but we received an unusually clear overview of the modifications Monday morning from the team at Arcade—a startup founded two years ago that has centered its entire enterprise around enabling AI agents to operate within actual companies, allowing them to securely connect to and utilize tools such as Gmail, Slack, and Salesforce.

Arcade secured $60 million in June, stemming from the belief that most AI agents do not falter due to weak foundational models but rather because the surrounding infrastructure is not yet prepared, which is what this update aims to rectify. Fundamentally, MCP is altering its approach to managing session IDs—the small tokens that servers use to retain the context of a conversation—enabling servers to function more effortlessly on a larger scale.

As Arcade’s founder Nate Barbettini explains:

[In the existing system] When an MCP client like Claude first connects to a server, it initiates a “hello”: I’m Claude, here’s my version, and here are my capabilities. The server then responds with its own capabilities and provides a session ID… Following that, the client sends that session ID with every request so the server recognizes it as the same conversation. Occasionally, the ID expires, necessitating the client to notice, request a new one, and continue….

Imagine a real-world deployment. You’re operating a server for millions of users, utilizing a load balancer whose sole function is to direct each request to whichever server in the array is available, sometimes across different regions. Now, every one of those machines must be aware of a session ID previously assigned by another machine. While not impossible, it poses significant challenges and complicates the load balancer’s job instead of assisting it.

In essence, the current configuration presumes a single server remembers the user, but real enterprises distribute traffic across multiple servers that do not communicate by default, necessitating today’s MCP servers to exert extra effort simply to identify users. This has created a notable challenge for anyone managing an MCP server at scale, partly explaining why we haven’t witnessed more companies launching large-scale, first-party MCP integrations despite the excitement surrounding agentic AI this year.

With the new system, the protocol will adopt a more flexible, “stateless” methodology for session IDs on the server side, akin to how most standard websites currently operate, which should simplify maintenance for the entire system and, in theory, reduce costs when scaling.

Although this is quite technical, it serves as a crucial reminder that not every aspect of AI development is progressing at lightning speed. While model training accelerates, much of the technical framework those models require is still subjected to the gradual consensus processes of standards bodies. Progress is indeed being made; it’s just occurring at a slightly slower pace!

Almost a year back, X, owned by Elon Musk, revealed plans to overhaul the Android version of its app, which had not performed as well as its iOS variant. On Monday, the company launched the revamped app, now ready for download.

The newly developed Android version of X was constructed entirely from the ground up and promises enhancements in loading speed, scrolling, notifications, and more, according to the announcement from X.

The update has been under construction for nearly a year. Last August, Nikita Bier, X’s product head, mentioned that the social media company was assembling an Android “dream team” to transform the experience. Later that fall, he also indicated that X experienced one of its best weeks ever for Android downloads in October — a key reason why the new app became a priority.

With today’s launch, Bier referred to the initiative as “one of the largest engineering projects” in the company’s history, emphasizing that the new Android app was developed from scratch instead of just receiving an update.

“It’s quicker, smoother, and more dependable. But most importantly: it will allow us to develop new features at incredible speed,” Bier shared on X. The social network owned by Elon Musk has recently been introducing a variety of new features, including X Money and X Chat, which now have their own separate applications.

The Android launch may also attract more users in global markets, where Android leads as the primary smartphone platform, encouraging them to either download or return to X, after years of platform disregard. (At one point last year, issues on Android were so severe that the X app wouldn’t load X posts when users clicked links.)

Nonetheless, Bier cautioned that there are still some rough aspects to address, including boosting performance on older Android devices and introducing support for Spaces, X’s live audio feature. Those upgrades are still in progress. Bier noted that additional features, such as the new video editor, the react-with-video feature, cashtags, and custom timelines, are also on the way for Android.

Current Android users can access the new X app by updating their existing application via the Google Play Store.