Learning from Hyperscalers: Our Takeaways from Google’s Accelerator

Last week, Spark presented at Google for Startups’ inaugural AI for Energy Accelerator at Google’s HQ in San Francisco. We were joined by 14 stellar teams building the future of energy. There has never been a more critical time to use AI to scale the infrastructure that will power this century — and this cohort is poised to meet the moment.

For Spark, it was a defining week: We announced our expansion into data center development, driven directly by customer demand (more on this in an upcoming post).

About the Accelerator

Led by Matt Ridenour (Head of Accelerator & Startup Ecosystem, IDEO alum), Erchit Sood (Program Manager, Startup Ecosystem), and the incredible Google for Startups team, this was a ten-week sprint bookended by in-person sessions and packed with workshops from some of the highest-leverage groups inside Google. We had deep dives with the DeepMind team, Google’s Advanced Energy group, the Data Center team, plus Marketing and Sales leadership. In classic Google fashion, we started with OKR setting and ended with sharing our OKR wins.

We rehearsed our pitch at least 10 times with coach Alex Gould, and we were lucky to have David White, a field CTO who integrates startups into the Google ecosystem, as our primary mentor. We also learned from Yingxia Yang (Global Energy Infrastructure) and Allegra Reister (Advanced Energy), who gave us a front-row understanding of Google’s enormous ambitions in sustainable, reliable energy.

Lessons from a Hyperscaler about Energy

Many hyperscalers strive to be good grid citizens: that means supporting a two-way grid, where data centers act as grid assets rather than pure liabilities. Speed to power is now the defining bottleneck in the entire ecosystem. Demand is outpacing everything, and that demand is what’s forcing innovation.

Google’s target is 24/7 clean energy. They’ve built an unprecedented power-procurement pipeline to do it. With 10 million Google queries per minute, 1 billion hours of YouTube videos streamed per day, and 70% of the world online, the scale is evident.

A few themes kept surfacing:

• Hyperscalers are preparing for AI load growth that no one fully understands yet. “Nobody knows what next quarter’s training demand will be.” So instead of chasing short-term spikes, build durable assets that stay valuable across cycles.

• The old world, where building a facility was the hard part, is gone. Now the constraint is the grid: interconnection timelines, local ordinances, and social license to operate. Community acceptance is now as tough as oil and gas was 15 years ago.

And one big, macro truth: data-center development is driving an outsized share of U.S. infrastructure build-out right now.

AI Insights

On the AI side, Google showed where model-driven development is headed. A few standout tools:

• Google AI Studio: a streamlined starting point for building, testing, and deploying with Google’s latest models.

• Vertex Model Garden: a centralized place to discover, customize, and productionize models from Google and partners.

This is an era where engineering managers are becoming individual contributors again because vibe coding removes barriers and requires architectural taste rather than pure coding execution.

We also had the chance to pitch Dr. Peter Norvig, one of the most influential figures in the history of AI. We asked him about the future of search — specifically, how LLMs, retrieval, and the traditional web stack will collide.

Our takeaways:

• Put LLMs at the center of data pipelines and business logic. The models will keep improving. Build around that trajectory.

• Architect for speed and determinism using background jobs, caching, and predictable retrieval.

What Spark AI Achieved

We entered the cohort with clear OKRs across Product, Engineering, and Growth, and this November is on track to be Spark’s strongest month since inception. A few highlights:

Product

• Shipped our battery-storage product, including BESS-specific regulations and fire codes across 16,000 new records

• Released sentiment heatmaps with color-coded county sentiment, shipped within days of an on-site customer visit.

• Debuting data-center community sentiment at Google Demo Day, the first of several data-center features is now in motion!

Engineering

• Integrated live search directly into Spark reports, unlocking nationwide coverage down to tens of thousands of townships.

• Replaced a sunsetted vector search provider with our own stack for semantic search, embeddings, and ranking, giving us more control and better accuracy.

Growth

• Launched a cohesive content + field marketing engine (thought leadership, podcasts, conferences), driving a 67% increase in qualified leads.

• Expanded the pipeline with more accounts progressing into the deal stage. Grew average deal size by 40%, strengthening unit economics and expansion potential.

• Signed the largest developer of distributed solar assets in the country as a customer.

The accelerator pushed us to operate with more clarity and more ambition, and the results are showing up everywhere in the business. We are truly grateful to Matt, Erchit, Allegra, Yingxia, David, Alex, and the team for this opportunity. Please check out the other rockstars in this cohort.

If you want to learn more about our accelerator experience or see what Spark is building next, let’s chat!