Google Cloud Next 2025 – our biggest takeaways

Authored by Simon Poulton, Partner – Cloud at Mantel

The painfully obvious bit:

My key takeaway from Google Next is crystal clear: we’re witnessing an AI explosion driven by a surge in practical applications.

Unlike previous years, where AI was just a glimpse of what’s to come, this year it was front and centre in every product demo. Businesses aiming to maintain a competitive edge in 2025/2026 must embrace AI, or risk falling behind.

I see strong parallels to the advent of personal computing. Before the 1980s, leveraging compute power required “dialling into the mainframe”, and therefore, specialised training. The PC revolution put computers on every desk, at home and in the workplace – consequently, it empowered people with accessible technology.

AI is at a similar pivotal moment. Detailed models like linear regression are beyond most people’s mathematical skills, making traditional machine learning difficult to apply. Today, tools like Gemini and ChatGPT handle these complexities in the background. This is akin to consumers using computers without needing to understand advanced mathematics or programming, and is similarly democratising AI.

Just as when computers were liberated, the practical applications expanded, and they became integral to business and daily life. Forty years ago, you could enter the workforce without computer skills, but that’s unthinkable today.

I believe we’re only ten years away from the same being true for AI and agentic systems. While part of the tech industry will focus on building these tools, knowing how to use them effectively will become part of the requisite skills for all white collar jobs.

Google’s layered approach to enabling enterprise-grade AI:

• Google’s strategy for enabling enterprise-grade AI is built on a robust, layered approach. First, there’s the hardware. At Google Next, the team unveiled the Ironwood TPU, a specialised chip for AI that boasts an astounding 3,500 times the capability of its predecessor released just 8 years ago.
  Importantly, Google is also abstracting the underlying layers, making this powerful hardware easier to leverage at scale. Google will manage a significant amount of complexity in coordinating many of the moving parts across the infra layer (Google calls this the AI Hypercomputer Stack). This means more AI power with fewer technical hurdles.  For the high-level, see the keynote here for some further detail on the hardware.

• Next up is the models themselves. Gemini 2.5 represents a significant leap forward, now topping the charts on platforms like lmarena.ai. Its strengths lie in handling long context, coding, and complex reasoning tasks. While it can sometimes be slower and more costly per result, it firmly positions Google at the forefront of AI model technology.  For the cheaper option (thanks Deepseek for the price pressure), there’s Gemini 2.5 Flash.

• Finally, there’s the tooling. Google’s AI platform, Vertex, supports a diverse range of models, including those developed in-house, open-source options, and third-party offerings. Vertex provides crucial features like scalability, explainability, and repeatability, making it possible to deploy machine learning, large language models, and agentic solutions more cost-effectively and with reduced risk. This, of course, requires a solid strategy and framework for adoption.

• Google is also focusing on new ways to integrate agentic experiences. The ADK (Agent Development Kit) simplifies the integration of agents and digital experiences. Additionally, the A2A protocol enables agents to communicate openly with each other, which is crucial for building more complex, reliable systems. To protect these models, Google introduced Model Armour, a tool designed to safeguard against adversarial AI actors, regardless of the specific model or system in use.

Google are leveraging their strengths to enhance the customer experience:

Below are some examples of strong vertical solutions Google have created:

• Financial services: Insurance, Mainframe, Data Migrations with Doc Automation, Anti-Money Laundering (AML) and Risk being the key panel topics.
• Retail: Vertex AI Search for commerce,  Personalised agentic experiences was definitely the hot topic, with a pretty amazing demo in the day 1 keynote.
• Healthcare: Agentic workflows & Data were the hot topic here.

• Google have also been busy on the horizontal solutions front as well, a great example of bringing the power of their 25+ years of experience in Search to Vertex AI Search. Here’s an example from Mercado Libre:
• Finally, an exciting step forward was the many announcements around Google Data Centre (GDC). This product allows you to bring the power of Google’s data centres and managed services inside your data centre.  Critical for air-gapped networks, certain types of sovereign data and of course edge use cases.  NVIDIA’s CEO Jensen Huang spoke about it here.

Finally, Google continues to innovate at the layers below the business line:

We saw a stream of announcements which continue to help Google’s customers focus more on the parts of their technology ecosystem which deliver value to the business.  Google Cloud achieves this by continuing to innovate with Managed services across the stack, a tighter developer experience and super clever ways of deploying AI to simplify moving to the cloud.  Here are some examples:

• Improvements in the infra layer to help abstract cloud migration and reduce the layers teams need to manage.
  1. Cloud WAN brings Google-style availability to the customer and cloud network while reducing the cost of the service.
  2. There were loads of sessions on GCVE, including customer experiences on how to move legacy VMware safely and quickly to the cloud, where they cost less to run and manage, to allow teams to focus more on AI and transformation.  Google announced more flexibility in the types of nodes that can be run.
• Improving the speed of Data migration, and the quality and usefulness of data was a topic explored from a technology perspective, as well as from an organisational change perspective
  1. Leveraging AI to help execute data engineering tasks, and therefore valuable for driving AI, was a recurrent theme.
  2. Data quality & lineage are improved with the new universal catalogue for BigQuery and the metastore.

This is just a tiny sample of what went on over an amazing 4 days – much of the value was in panel conversations and smaller presentations which didn’t make the recording reel. If you’re interested in discovering the full details of these, or more about the above topics, our team has been running individual sessions. Reach out to us using the form below and we can continue the conversation there.

Finally my favourite takeaway is that the Killers still crank out as tight a set as they did last time I saw them in 2009 at O2 in Ireland. The unexpected takeaway???  At 55 if Wyclef Jean jumps into the crowd during a show, you better be ready to bring the energy (see the 26 minute mark, warning, inappropriate language!!)

That’s a wrap on everything I took away from Google Cloud Next 2025 – keep an eye on this space, as the rest of the team will be giving their own highlight recaps over the next few days and weeks.