AI Infrastructure Attracted $4.7 Billion in August with the Bet Shifting Toward Inference

Five major funding rounds in August 2026 raising over $4.7 billion revealed a systematic shift in the AI investment thesis: after OpenAI and Anthropic alone accounted for nearly 14% of global venture capital investments in the first half of 2026, the actionable investment bet is shifting toward two loops: enterprise inference economics and inter-chip interconnect networks.

The deals clearly embody this shift; Fireworks AI raised $1.505 billion in Series D at a $17.5 billion valuation, and Together AI raised $800 million in Series C at an $8.3 billion valuation — both in enterprise inference. On the infrastructure front, Lumilens emerged from stealth with over $900 million led by Atreides Management to develop optical interconnects in data centers, while Point2 completed a $136 million extension of its Series B with a strategic investment from Arm.

Signals indicate that interconnect networks have become the real bottleneck: when Arm takes a strategic position in Point2 within days of Lumilens emerging with $900 million in the same segment, investors are pricing in interconnect constraints, not chip constraints — a bet where supply and demand dynamics appear more attractive than GPU chips dominated by big tech spending.

The core risk lies in "stack compression": Nvidia's emergence as a strategic investor in River AI, Volta, and several interconnect deals indicates that the sector's largest supplier is securing options at every layer of its own demand chain. If Nvidia succeeds in solidifying its positioning across both inference and interconnect networks, independent portfolio companies may find themselves facing vertical integration from the supplier they are supposed to be independent of.

What Do These Terms Mean?

Inference: Running a trained AI model to answer user queries, which is the primary revenue driver in the market today.

Interconnect: The hardware and protocols connecting chips and servers within data centers, setting the ceiling for actual computing speed.

Stack Compression: Control by a single provider over multiple layers of the sector's value chain, reducing ecosystem independence and concentrating pricing power.

Series D: A late-stage funding round targeting startups that have proven their business model and are preparing for accelerated expansion.