At SlatorCon Silicon Valley 2025, executives from Google, Microsoft, and Amazon Web Services (AWS) came together for a panel discussing how startups and larger enterprises can leverage major cloud platforms to build and scale language AI solutions.
Experts Raj Sachde, Cloud Migration and Modernization Leader at Amazon Web Services (AWS), Prerak Garg, Partner Director for Cloud + AI Strategy at Microsoft, and Sabri Hammad Senior Solutions Architect with Google Cloud Consulting, began the discussion moderated by Slator’s Managing Director, Florian Faes, by exploring how startups, scale ups, and even buyers could build on their foundational platforms.
All three panelists agreed that startups should take advantage of their existing technology, rather than trying to build new models on their own. “Don’t start from scratch,” advised Hammad, telling the audience that Google, his co-panelists’ companies, and others like OpenAI “all have really great foundational models you can use.”
Partner spotlight
Riding the AI shockwave
The future of AI adoption depends on trust, ethics and human expertise.
Furthering his point, Hammad took a problem-solving perspective, advising startups to “do an assessment, run your comparison. See which tool fits your business model best, and go after it.”
Sachde took the stance that “technology is a given,” emphasizing the breadth of available options by reminding attendees that “we [AWS] have a hundred plus models available. So [a] model is always a given, an afterthought”.
Sachde thought that startups should focus on differentiation and developing their pricing and distribution systems. He encapsulated the point, saying, “So in my mind, it is more about the problem you’re trying to solve. And, technology is something that we will ensure that you are successful.”
Garg framed his advice as a time-saving strategy, highlighting the need for startups to move quickly from idea to prototype to market. He discussed this pathway, saying, “I think the key is, ‘how do you move from initial idea to prototyping?’ And then, don’t waste too much time in the initial prototyping stage. Get it out in the market, test it with users, and if it works, then think of scaling.”
According to Garg, partnering with a hyperscaler platform should help a startup scale quickly by reducing the pressure on its technical development. He noted, “these platforms can help you scale really fast because you don’t have to worry about the underlying infrastructure… plumbing, all of that is taken care of. And all the tooling – whether it’s evals, monitoring, etc. – is all available under one platform.”
In addition to leveraging existing technology from large platforms, the panelists advised that startups take advantage of funding opportunities available from their companies to address the financial pressure felt by small, growing companies.
Stand Out From The Crowd
Another theme that emerged was the need for startups to differentiate themselves from hyperscalers and competitors in order to succeed in the market.
When asked about the challenges inherent to working with AI language startups, Hammad noted that while hyperscalers like Google have strong engineering teams, robust infrastructure, and the ability to work at scale, they lack domain-specific expertise.
Garg extended the idea of specialization from experts to data. He suggested asking the question, “am I building a product that can become better with scale and more usage?” He encouraged startups to prioritize bringing specific proprietary data to their product, then using the domain-specific user data they collect over time to fine-tune their model for better performance.
He argued this creates what Microsoft internally calls a “signals loop,” allowing a domain-specific app to improve over time based on “signals” from users, and compete more successfully with “very horizontal apps like ChatGPT or Copilot” that have less ability to specialize.
A Few Challenges
Hammad also pointed out a fundamental challenge from the power requirements needed to continue expanding the “Grand Canyon” of data centers he observes on the east coast of the US. “Where are we going to get the power for this? That’s why all three of the hyperscalers here have actually invested significantly in nuclear and things like that, to try to get around that scale,” he commented.
The panelists also addressed sensitive data concerns given confidentiality regulations like the Health Insurance Portability and Accountability Act (HIPAA), agreeing that anonymizing/sanitizing data to protect patient privacy was an essential step.
Hammad further pointed out that essential data-anonymization steps also slow down the development process, observing that regulations like HIPAA “have to catch up with where AI is.”
Featured
Partner spotlight
Boost Language Access
Improve health outcomes and ensure compliance for individuals with LEP
![[SCSV25] Google, Microsoft, AWS Panel 1](https://tech.slator.com/assets/2025/09/SCSV25-Google-Microsoft-AWS-Panel-1.png)
