Google CEO: Gemini's Next Leap Depends on Building Much Larger Base Models

Pichai's Statement: Why Google Needs Bigger Models

During Alphabet's Q2 2026 earnings call, CEO Sundar Pichai made a striking declaration: Gemini's next generation will require "much larger base models." The statement came as Google kicked off its most ambitious pre-training run yet for Gemini 4, signaling that the company remains firmly committed to the scaling paradigm despite growing debate about its limits.

"We have started our most ambitious pre-training run yet for Gemini 4, and are excited by the progress we are seeing at the frontier," Pichai said. This directly reinforces the importance of scale in frontier AI development, even as competitors and researchers question whether scaling laws are reaching diminishing returns.

Alphabet's Record Infrastructure Investment

Alphabet raised its 2026 capital expenditure forecast to between $195 billion and $205 billion, up from the previous $180–$190 billion range. The company reported Q2 revenue of $119.8 billion — a 24% year-over-year increase that beat analysts' expectations of $116.9 billion. Google Cloud grew 82% to $24.8 billion, while Services rose 15% to $94.5 billion.

These figures paint a clear picture: Google is not just talking about scale — it is spending unprecedented amounts on AI infrastructure. A $205 billion annual capex is unmatched in technology history and demonstrates Google's conviction that building larger models is the path to better AI.

The Gemini app now has 950 million monthly active users, up from 750 million in February. AI Mode in Google Search has crossed one billion monthly active users since its global launch, and Google claims it is driving more overall search queries.

Scaling Laws vs. Efficiency: The Debate Continues

Pichai's remarks are significant in the context of ongoing debates in the AI industry. Recently, some researchers have argued that Scaling Laws — the principle that larger models with more data and compute yield proportionally better results — may be slowing down or plateauing. DeepMind CEO Demis Hassabis himself has acknowledged uncertainty about what comes next.

Yet Pichai's statement makes clear that Google, at least for now, is doubling down on scale. The company is simultaneously investing heavily in efficiency gains. The cost per AI response in AI Mode has dropped to its lowest level since launch, even as Google uses more powerful models. This efficiency was also evident in the recent Flash 3.6 release.

As Pichai noted, "We're seeing tons of demand for our workhorse Gemini Flash series because it hits the sweet spot of performance and cost." This suggests Google understands the importance of balancing powerful frontier models with efficient, cost-effective workhorses for everyday use cases across its ecosystem.

The key question is whether Google can maintain this balance — investing in massive frontier models while simultaneously driving down inference costs — without spreading its resources too thin.

Competition with OpenAI and Anthropic

Google's primary challenge over the past two years has been the rapid progress of OpenAI (GPT-4o, o1, o3) and Anthropic (Claude 3, 3.5, Opus 4.6). Despite releasing multiple versions of Gemini, Pichai himself acknowledged that Google needs a more capable base model to catch up with the leaders.

He specifically pointed to weaknesses in coding: "There are areas where we have acknowledged we need to improve. Coding and agentic coding is an example of that, and I think the teams are very, very focused on it." This is a direct reference to competing with OpenAI's Codex and Anthropic's Claude coding capabilities, both of which have set high benchmarks in software development tasks.

Google is still working on a new frontier model and plans to ship new models monthly. But Gemini 3.5 Pro apparently isn't part of that cadence, even though Google says it's currently being tested. This suggests the company may have decided to leapfrog directly to Gemini 4 as a more significant jump.

What to Expect from Gemini 4

Based on Pichai's statements and Google's acknowledged weaknesses, we can anticipate several key improvements in Gemini 4:

• Substantially improved coding capabilities — addressing the gap with OpenAI and Anthropic in software development
• Stronger reasoning (Chain-of-Thought) — competing with models like OpenAI's o1 and o3
• Better agentic functionality — enabling autonomous task completion and tool use
• Enhanced multimodal capabilities — building on Gemini's native multimodality
• Greater efficiency in Flash variants — continuing the trend of cost reduction per query

The "much larger base model" approach suggests Gemini 4 will be significantly more expensive to train than its predecessors, but Google is betting that the resulting capabilities will justify the investment.

Implications for the AI Industry

Pichai's statement carries several important implications for the broader AI landscape:

First, the scale race continues. Despite efforts to create more efficient smaller models (from Microsoft's Phi series to Mistral's offerings), size still matters for frontier AI. Google, OpenAI, Anthropic, and others will continue building increasingly large models, driving up both capabilities and costs across the industry.

Second, capital intensity is accelerating. With annual investments approaching $205 billion, only the largest technology companies (Google, Microsoft, Amazon, Meta) can realistically compete at the frontier. This creates a significant barrier to entry for startups and open-source initiatives, potentially consolidating AI leadership among a handful of megacaps.

Third, Gemini 4's success or failure will define Google's AI trajectory. If Google can deliver a model that matches or exceeds the best from OpenAI and Anthropic, its massive ecosystem advantage (Search, Cloud, Android, YouTube, Workspace) could tip the competitive balance. If Gemini 4 falls short, Google risks being seen as a perpetual follower in the AI race.

Fourth, the efficiency paradox. The same earnings call revealed that AI response costs are dropping even as models grow larger. This suggests a bifurcating market — ultra-powerful frontier models for complex tasks, and efficient distilled or specialized models (like the Flash series) for everyday use. Google's strategy appears to be competing on both fronts simultaneously.

Conclusion

Sundar Pichai's declaration that Gemini's next leap requires "much larger base models" confirms that Google is ready to compete aggressively for AI leadership. The $205 billion investment and ambitious Gemini 4 training run represent a massive bet on scale as the winning strategy for frontier AI development.

However, success is far from guaranteed. OpenAI and Anthropic have significant momentum and continue to innovate rapidly. The next 12–18 months will be decisive in determining whether Google's "next leap" actually leapfrogs the competition or merely catches up to where the leaders already are. For the AI industry as a whole, the message is clear: the era of scaling is far from over — it is entering its most capital-intensive phase yet.

📖 Source