Lenovo's hybrid AI pitch: Optimize your token economics

Lenovo's fourth quarter results included the usual complement of better-than-expected results, traction in servers, PCs, storage and even smartphones. But tokens were mentioned repeatedly--13 times in fact.

Welcome to the new era of hardware vendors talking hybrid AI deployments and improving your token economics, which frankly suck right now.

Lenovo's fourth quarter results are just part of a broader argument for on-premises AI deployments. Dell Technologies made its case at Dell Technologies World this week. Dell will also talk about AI infrastructure on its earnings call. HPE will likely do the same.

• Dell Technologies courts on-premises AI workloads from workstation to data center
• AI inference costs are going to be a big concern: What's the fix?
• Where’s tokenomics for the rest of us?
• The art and science of blending human and AI labor

The argument is relatively simple. Work out your token optimization on premises so you're not burning through your budget just trying to figure out models. Of course, there's a broader problem that token burn isn't exactly predictable because LLMs do something different every time, but that's an issue for another day.

On Lenovo's earnings call, executives mentioned hybrid 17 times and token 13 times.

Ashley Gorakhpurwalla, Executive VP & President of the Infrastructure Solutions Group at Lenovo, said on the company's earnings call that on-prem AI is at an inflection point.

Gorakhpurwalla said:

"Enterprise use of AI agents, coupled with the economics of more efficient tokens, but literally hundreds more token usage is really poised to grow the addressable market in hybrid AI for our industry to $1 trillion by 2029, and we have an awesome opportunity in front of us.

We have had the biggest refresh of our product line in our entire history of the Think Series portfolio around AI inferencing and how to optimize token economics."

Sure, Lenovo is talking a lot about AI factories, HPC and scale to get enterprise customers and hyperscalers to "speed to first token," but the company is also expecting a lot of on-premises and inference away from the cloud.

Lenovo's Shao-Min Cheng, Group Chief Financial Officer, said:

"Combining private and public environments to accelerate time to first token, improve token efficiency and maximize value per token, Lenovo hybrid AI helps enterprise customers shorten time to ROI to less than 6 months while delivering production-ready AI environment in as little as 90 days. The Rubin-based platforms deliver up to 10x lower cost per token versus previous generations helping customers bring AI workloads on premises with greater efficiency and stronger control over data and a clear repeatable business outcomes."

Lenovo is also spinning up its hybrid AI library of more than 60 enterprise use cases across multiple industries.

Lenovo posted strong fourth quarter results with revenue of $21.6 billion, up 27% from a year ago, and net income of $559 million doubled. AI-related revenue was up 84% from a year ago. The company is targeting $100 billion in revenue in two years, up from the $83.1 billion delivered in fiscal 2026.