US AI startups raise $150b in 2025, led by OpenAI, Anthropic

AI startups in the US have raised a record US$150 billion in funding in 2025, according to PitchBook data.

This surpasses the previous record of US$92 billion set in 2021, as investors back companies like OpenAI, Anthropic, and Scale AI.

Large deals in 2025 included OpenAI’s US$41 billion round led by SoftBank, and Anthropic’s US$13 billion raise.

Other AI companies such as Perplexity, Anysphere, and Thinking Machines Lab have also completed multiple funding rounds this year.

VCs said the new capital could help startups withstand a potential downturn in AI investment as concerns about high spending on infrastructure grow.

Some investors advised founders to build larger cash reserves while enthusiasm for AI remains high.

Major VC firms like Thrive Capital, Andreessen Horowitz, and Tiger Global have started raising new funds to keep pace with the investment surge.

🧠 Food for thought

Implications, context, and why it matters.

$150 billion for US AI startups hides who gets money.
OpenAI at $41 billion and Anthropic at $13 billion make up over one-third of 2025 sum, which tilts funding to foundation model startups that build large, general-purpose AI models adaptable to many tasks.
We do not know the stage mix for seed and early rounds (initial checks) or late-stage growth for mature startups.
If money clusters in mega-rounds for incumbents, the trend could fade if investor sentiment turns.
GPU underutilization can hit 70-85% 1, so startups with large infrastructure budgets may burn cash despite record raises, which creates a sustainability gap not captured in totals.

Infrastructure founders see a cloud financial operations (FinOps) opening because generative AI (GenAI) inference, the cost to run models for outputs, makes up 80-90% of total AI spend 1 and 72% of IT plus financial leaders call that spending unmanageable 2.
Engineering and platform teams at AI firms can use dynamic scaling, which right-sizes compute to demand while cutting GPU costs by 40-70% 1.
GPU pooling shares GPUs across workloads and lifts utilization.
Prompt and token optimization reduces the units of text a model processes, which lowers token use by 20-40% 1.
The FinOps market sits at $5.5 billion in 2025 with a 34.8% Compound Annual Growth Rate (CAGR) 3.
This helps counter worries about high infrastructure spend that could trigger a downturn.