Greylock Bets $1.5B on the Next AI Wave

Venture capital firm Greylock Partners has announced the launch of its 18th fund, a substantial $1.5 billion dedicated to backing early-stage AI entrepreneurs. The move signals continued strong investor confidence in the artificial intelligence sector, with Greylock aiming to identify and support the next generation of AI leaders.

Greylock's AI Investment Strategy

Saam Motamedi, General Partner at Greylock, discussed the firm's strategy, drawing parallels between the current AI boom and the early days of the mobile revolution. He stated, "We're in the equivalent 2008 or 2009. We're one to two years after the iPhone moment, and I think we're gonna look forward and many of the defining AI companies of this generation have yet to get started." This new fund is designed to capture that potential.

The full discussion can be found on Bloomberg Technology's YouTube channel.

Greylock's investment thesis for AI spans three core layers:

• Model Layer: Building on their existing investments in foundational model companies like Anthropic and OpenAI, Greylock will continue to back new ventures developing novel approaches or focusing on specific data domains.
• Infrastructure Layer: The firm anticipates a significant transformation of the technology stack around AI agents. They see a parallel to the rise of cloud infrastructure and hyperscalers, predicting the creation of new, purpose-built services for AI workloads. Companies like BaseTen (AI inference) and Snorkel (data) are cited as examples in this space.
• Applications Layer: With agents now capable of long-running tasks, Greylock expects a proliferation of AI-powered applications. They point to companies like Crest AI (customer service) and Abnormal (cybersecurity) as early examples, and highlight Resolve AI, which builds agentic on-call engineers, as a testament to the practical impact of this technology.

Assessing New AI Models: Kimi K3 and Benchmarks

The conversation also touched upon the recent release of Moonshot AI's Kimi K3 model, which has garnered attention for its performance on certain benchmarks and its large context window. Motamedi expressed a measured perspective on the hype, noting, "Benchmarks are imperfect." He cautioned that real-world performance and trade-offs are crucial for evaluating a model's true capabilities.

Regarding cost, Motamedi challenged the narrative that Kimi K3's pricing would necessarily lead to price erosion for frontier models. He pointed out that Kimi K3 is reportedly more expensive per token than its predecessor and is less token-efficient, meaning it may consume more tokens for a given task. "This conclusion that it's gonna lead to price erosion for the frontier is perhaps a bit premature," he stated.

Motamedi also emphasized the multi-faceted moats of leading AI companies like OpenAI and Anthropic, citing not only their model advancements but also their significant revenues derived from API businesses and first-party products like Cloud Code and Cloud Tag. He suggested that the assumption of a single "best" model winning might be too simplistic, as the overall AI economy is projected to grow exponentially.

The growth in token usage, with OpenAI's API alone increasing from 30 million tokens per day in 2023 to a projected 15 billion by March 2026, highlights the massive expansion of the AI token economy. Motamedi concluded by stressing the importance of recognizing this overall growth, stating, "The only mistake one can make is underestimating the size of this overall wave."