Thomas Kurian 说，在起步较晚的情况下，Google Cloud 的 full-stack AI 策略正逐步缩小与 AWS 和 Microsoft Azure 的差距；自他上任以来，云端市占已从 7% 提升到 14%。虽然 Google Cloud 在约 4,180 亿美元（418bn）规模的全球云端运算市场中仍列第三，但动能在加快：2025 年第四季收入成长 48%，且 2026 年预期收入将超过 700 亿美元（70bn），高于 2024 年的 430 亿美元（43bn）。Kurian 强调，Google 不再必须将每一美元营收中的 80% 汇给外部模型或晶片供应商，因为它掌握了自己的 IP、模型与晶片。

Google 于拉斯维加斯发布第八代 TPU，其中一款专注训练模型，另一款以更大记忆体加速推理，并认为自建硬体与软体垂直堆叠可降低成本、提高毛利，而非依赖 Nvidia GPU。Google 表示其约掌握全球四分之一 AI 计算能力，大约为 380 万 TPUs 与 130 万 GPUs；而 Microsoft 则约有 320 万 Nvidia GPUs。面对 Jensen Huang 对 Google 晶片性能与效率的质疑，Kurian 指出前十名 AI 研究室中有九家使用 TPU，包括 Mira Murati 的 Thinking Machines，并表示若 Google 在效能、价格或品质上不具竞争力，客户本可选择不采用。

Google 亦以资本支出预估上升到 1850 亿美元（185bn）支持自有 AI 版图，并与 Anthropic 成立交易：投资上限 400 亿美元（$40bn）并在五年间提供 5GW 的计算能量，价值超过 2,000 亿美元（200bn）。Kurian 认为 OpenAI 与 Anthropic 为争夺计算力而每年亏损数十亿美元，且两者都已在本年融资超过 1500 亿美元（$150bn）筹划 IPO；他预测未来一至两年将出现清洗，若业者无法平衡经济模型，便可能被淘汰，这一风险同样可能牵动依赖这些 AI 提供者的 Big Tech 生态。

Thomas Kurian said Google Cloud’s full-stack AI strategy is helping it close the gap with AWS and Microsoft Azure despite a late AI start, with market share rising from 7% to 14% under his leadership. Google is still third in the roughly $418bn global cloud-computing market, but momentum is accelerating: Cloud revenue rose 48% in Q4 2025, and it expects more than $70bn in 2026, up from $43bn in 2024. Kurian stressed that Google no longer has to pass 80% of revenue to external model or chip providers because it owns its own IP, models, and chips.

Google launched eighth-generation TPUs in Las Vegas, with one chip focused on training and one with larger memory for faster inference, arguing that its vertically integrated stack lowers cost and improves margins versus reliance on Nvidia GPUs. Google said it controls about a quarter of global AI computing power—roughly 3.8mn TPUs and 1.3mn GPUs—while Microsoft has about 3.2mn Nvidia GPUs. Responding to Jensen Huang’s critique of Google chip performance, Kurian said nine of the top 10 AI labs, including Mira Murati’s Thinking Machines, use TPUs, adding that customers would switch away if Google were weaker on performance, price, or quality. (Key numbers: 380, 130, 320)

Google is also scaling spending with capital expenditures projected at $185bn to support in-house AI, and it has a deal with Anthropic for up to $40bn investment and 5GW of compute over five years, worth more than $200bn. Kurian said OpenAI and Anthropic are burning tens of billions a year as they compete for compute power and both have raised over $150bn in private rounds this year while preparing for IPOs. He predicts a one- to two-year shakeout, warning that providers that cannot achieve sustainable economics are likely to fail, and that this risk could also spread to Big Tech ecosystems dependent on them.