Google CEO Pichai says Gemini's next leap depends on building "much larger base models"

Alphabet 已将其 2026 年的投资预测上调至 1950 亿至 2050 亿美元，高于之前的 1800 亿至 1900 亿美元区间。公司表示，需求仍然超过投资。

谷歌报告称，2026 年第二季度收入为 1198 亿美元：https://s206.q4cdn.com/479360582/files/doc_financials/2026/q2/2026q2-alphabet-earnings-release.pdf，同比增幅为 24%，高于分析师预期的 1169 亿美元。Google Cloud 增长 82%，达到 248 亿美元，服务部门增长 15%，达到 945 亿美元。Gemini 应用的月活跃用户现已达 9.5 亿，高于 2 月的 7.5 亿。

自去年 10 月全球发布以来，Google Search 中的 AI 模式已突破十亿月活跃用户。谷歌表示，AI 模式像之前的 AI 概览一样，正在推动整体搜索查询的增加。这不应与相比传统搜索提升网站流量混淆：https://the-decoder.com/pew-finds-that-only-1-percent-of-users-click-a-source-link-directly-from-googles-ai-overviews/。

谷歌表示，它每周通过搜索中的 AI 功能向网站发送数十亿点击，但没有具体数据或与传统搜索量的比较，这一说法的信息有限。在广告方面，AI Max 已经结束测试阶段，已有 50 万广告商正在使用。Alphabet 表示，AI 正在释放数十亿以前难以变现的搜索查询。广告 DEC_D_Incontent-1

即使谷歌使用更强大的模型，AI 模式下每次 AI 响应的成本已降至自发布以来的最低水平。这显示了效率提升，这一点最近的 Flash 3.6 版本：https://the-decoder.com/google-ships-three-new-gemini-flash-models-but-its-frontier-3-5-pro-remains-lost-in-training/ 也显示了出来。随着时间推移，这种效率可能为谷歌在 AI 服务的利润率上带来优势。

这也引发了一个问题，即最前沿的模型对于谷歌日常运营的重要性到底有多大。对于AGI竞赛，它可能仍然很重要：https://the-decoder.com/deepmind-ceo-hassabis-says-nobody-in-the-world-knows-what-happens-next-so-cautious-optimism-means-building-guardrails-now/。但在为整个生态系统提供日常用例服务方面，可能就没那么重要了。谷歌CEO桑达尔·皮查伊表示：“我们看到Gemini Flash系列受到了大量需求，因为它在性能和成本之间达到了最佳平衡。”

谷歌仍在开发新的前沿模型，并计划每月发布新模型。但显然Gemini 3.5 Pro并不包含在这个节奏中，尽管谷歌表示它目前正在测试中。

在财报电话会议上，CEO桑达尔·皮查伊承认，谷歌首先需要一个更强大的基础模型，以赶上领先者。他表示，下一代前沿AI模型需要“更大规模的基础模型”。谷歌正在训练Gemini 4以达到这一标准。

皮查伊表示：“我们已经开始了迄今为止最雄心勃勃的Gemini 4预训练运行，对于在前沿取得的进展感到非常兴奋。”

他还指出了谷歌正在努力改进的具体弱点。“有些领域我们已经承认需要改进。编程和智能编程就是一个例子，我认为团队在这方面非常非常专注。”

保持AI领域的最新动态。清晰、有用、无虚饰。

关注The Decoder，获取AI新闻、背景故事和专家分析。

The Decoder：https://the-decoder.com/

Alphabet has raised its 2026 investment forecast to between $195 billion and $205 billion, up from the previous range of $180 billion to $190 billion. The company says demand continues to outpace investment.

Google reported revenue of $119.8 billion：https://s206.q4cdn.com/479360582/files/doc_financials/2026/q2/2026q2-alphabet-earnings-release.pdf for the second quarter of 2026, a 24 percent increase year over year that beat analysts' expectations of $116.9 billion. Google Cloud grew 82 percent to $24.8 billion, and the Services segment rose 15 percent to $94.5 billion. 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 in October. Google claims AI Mode, like AI Overviews before it, is driving more overall search queries. That shouldn't be confused with more traffic to websites compared to traditional search：https://the-decoder.com/pew-finds-that-only-1-percent-of-users-click-a-source-link-directly-from-googles-ai-overviews/. Ad

Google says it sends billions of clicks to websites every week through AI features in Search, but without specific data or comparisons to traditional search volume, that claim doesn't tell us much. On the ad side, AI Max has left beta and already has 500,000 advertisers using it. Alphabet says AI is unlocking billions of new search queries that were previously hard to monetize. Ad DEC_D_Incontent-1

The cost per AI response in AI Mode has dropped to its lowest level since launch, even as Google uses more powerful models. That points to growing efficiency gains, something the recent Flash 3.6 release：https://the-decoder.com/google-ships-three-new-gemini-flash-models-but-its-frontier-3-5-pro-remains-lost-in-training/ also showed. Over time, that efficiency could give Google an edge on margins for serving AI.

It also raises the question of how much a frontier model still matters to Google day to day. For the AGI race, it likely does：https://the-decoder.com/deepmind-ceo-hassabis-says-nobody-in-the-world-knows-what-happens-next-so-cautious-optimism-means-building-guardrails-now/. For serving daily use cases across its ecosystem, probably less so. "We're seeing tons of demand for our workhorse Gemini Flash series because it hits the sweet spot of performance and cost", Google CEO Sundar Pichai said. Ad

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.

CEO Sundar Pichai acknowledged during the earnings call that Google needs a more capable base model first to catch up with the leaders. The next generation of frontier AI models requires "much larger base models," he said. Google is training Gemini 4 to meet that bar. Ad DEC_D_Incontent-2

"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. Ad

He also pointed to specific weaknesses Google is trying to fix. "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."

Stay in the loop on AI. Clear, useful, no fluff.

Follow The Decoder for AI news, background stories and expert analyses.

The Decoder：https://the-decoder.com/