Wafer, An Inference Provider That Uses Non-Nvidia Chips, Lands Acquisition Offers and $200 Million-Plus Valuation

Developers and chip designers are increasingly upbeat about the benefits of using AI to supercharge the process of designing and optimizing AI chips, as my colleague Phoebe wrote about last week.

Now, a year-old inference provider, Wafer, is trying a similar approach: using AI to improve the way models run on a variety of chips and then helping businesses to run those models.

The San Francisco-based startup recently raised $40 million in Series A funding co-led by Marathon Management Partners and Chemistry VC, cofounder and CEO Emilio Andere told The Information exclusively. The startup’s valuation now stands above $200 million, according to a person with knowledge of the round. Andere, who cofounded Wafer with his college best friend Steven Arellano in May 2025, declined to comment on the valuation.

Andere said Wafer has developed AI agents—which it calls AI inference engineers—that can take a business’ specific AI workload and optimize open-source models for that workload. For instance, Wafer’s AI inference engineers would make different optimizations to an open-source model for a voice-based AI agent, which needs to be able to quickly handle repeated prompts and bursts of traffic, than they would for a coding agent, which needs to be able to remember important context over longer periods of time, Andere said.

Wafer differs from other inference providers by running its AI on chips from both Nvidia and Advanced Micro Devices. It also hopes to eventually apply its AI agents to a host of other non-Nvidia chips, such as those from Google, Sambanova and Cerebras.

What Wafer is offering may be especially helpful for businesses wanting to run AI models on non-Nvidia chips, which are more readily available but lack the shared knowledge and software optimizations developers have built up over years of using Nvidia chips for AI, Andere said.

In July, for example, Wafer said that it was able to optimize Z.ai’s popular GLM-5.2 model to run on AMD’s Instinct MI355X. In doing so, the model was able to process requests around 80% as fast as an Nvidia B200 chip but at less than half the cost, the company said.

That sort of technology is valuable. Wafer’s revenue has grown to around $8 million in annualized revenue—or $666,000 a month—in the span of 12 weeks, according to Andere. That reflects businesses’ growing concern with the costs and speed of running AI.

Moreover, the startup has received multiple acquisition offers from companies including larger inference providers and cloud providers, which it turned down, according to a person with knowledge of the offers. (Andere also declined to comment on those offers.) You can imagine why such companies would be interested in buying a startup that would help to unlock underutilized sources of compute, given the ongoing compute crunch.

Currently, Wafer serves startup customers including Vercel, which uses its tech to serve ultra-fast coding models, and voice AI startup Inworld AI, Andere said. Like other inference providers, Wafer has gross margins of around 50%, which are lower than the typical software business, due to the costs of having to rent chips, he said.

Here’s what else is going on…

Big Number

OpenAI’s advertising business has reached $1 billion in annualized revenue run rate, the company said Monday, suggesting it is likely to fall well short of the $2.4 billion in ad revenue it had projected for 2026. The ChatGPT creator launched ad sales in the chatbot in February, as a way to generate money from the hundreds of millions of people who use ChatGPT without paying.

Overheard

Communities that don’t want data centers will end up being “backwards and poor,” President Donald Trump said Monday on TruthSocial, signaling his backing of the politically-charged industry ahead of the upcoming midterm elections.

“If we kill the Golden Goose, you will only have yourselves to blame,” Trump added in a social media post. Last week, the National Republican Senatorial Committee warned that Democrats were using voter opposition to data centers successfully in a campaign against a sitting Ohio Senator, Axios reported.

People on the Move

After 15 years leading Apple, Tim Cook stepped down as chief executive on Monday. And while he is committed to a smooth transition and continuity to the tenure of John Ternus—senior vice president of hardware engineering, who takes over on Tuesday—changes are already happening. Notably, long-time Apple executive Phil Schiller is stepping down as head of App Store and product events, Bloomberg reported. The App Store team will be moved under services lead Eddy Cue, while events will be switched to public relations lead Kristin Huguet Quayle.

Court Watch

Apple filed new “shocking evidence” in its lawsuit against OpenAI alleging that former employees had stolen trade secrets for the ChatGPT-maker. Apple alleged that former employee Chang Liu, who now works at OpenAI, used a confidential Apple circuit schematic in his work at OpenAI, as well as a tool that shares a name with an internal Apple engineering application.

ChangXin Memory Technologies (CXMT), China’s leading memory chipmaker, is suing the U.S. Department of Defense as it seeks a removal from a Pentagon blacklist identifying it as a Chinese military-linked entity. CXMT denied any affiliations with the military and said it produces its DRAM chips for civilian and commercial uses. Its lawsuit, filed in the U.S. District Court for the District of Columbia, also names Defense Secretary Pete Hegseth among the defendants.

The Federal Trade Commission and attorneys general from 22 states sued Amazon for overcharging its advertising customers by allegedly manipulating ad auctions. Advertisers promoting their products on Amazon’s marketplace buy ad spots where the price is set in auctions. In the complaint, the FTC and the states allege Amazon added an extra charge to bids in order to raise the price that advertisers paid above what one employee said could be achieved by “advertiser competition” alone.

Deals and Debuts

See The Information’s Generative AI Database for an exclusive list of private companies and their investors.

Anthropic has signed a cloud-computing deal worth $35 billion with Lambda, the Nvidia-backed company that rents out AI chips, the Wall Street Journal reported.

Clay, whose software helps sales and marketing teams find and research potential customers, is raising a funding round led by Wellington at a $7 billion pre-money valuation, up from $5 billion in an employee share sale in January, Axios reported.

Nvidia invested $3.5 billion in convertible bonds issued by Taiwanese chipmaker MediaTek, Nvidia said, the latest sign of collaboration between the two companies.

SLB has agreed to acquire Kelvion, a century-old maker of the heat exchangers and cooling equipment that keep data centers from overheating, for $3.4 billion in cash plus the assumption of about $700 million of debt—a total transaction value of roughly $4.1 billion.

Andreessen Horowitz said Monday it had raised an additional $1.75 billion for its fifth growth fund, bringing the total raised to $8.5 billion. Andreessen Horowitz raised $6.75 billion for that fund back in January.

Enflame Technology, a company that designs AI chips and is backed by Tencent, priced its initial public offering on Shanghai's STAR Market at 142.18 yuan (about $21.16) a share, selling 43.04 million shares to raise about 6.12 billion yuan (roughly $911 million).

Carrick Capital Partners put $255 million of new money into Saviynt, whose software controls who is allowed to touch a company‘s systems, as part of the final close of Saviynt’s $700 million Series B funding round.

Chinese AI firm Z.ai’s revenue in the first half jumped nearly fivefold from a year earlier to 954 million yuan ($142 million), boosted by skyrocketing sales of cloud-based AI model services.

Physical Super Intelligence, which is using AI for physics research, raised $58 million in seed funding led by Breakthrough Energy Ventures with participation from Dragon Global.

Adaptyv Biosystems, which develops AI for protein design, raised $40 million in a Series A funding round led by Highland Europe.

Easy Aerial, a company that makes drones which stay tethered to a base station so they can hover over a site for long stretches watching it, raised $20 million in a Series B funding round from Insight Partners, Entrée Capital and other investors.

Clipto, which indexes videos, audio, images, meetings and other files on a user’s computer, raised $15 million at a $250 million post-money valuation from investors including HSG (formerly Sequoia China), GL Ventures, EnvisionX Capital, Palm Drive Capital, Hans Tung, Lu Zhang and 522 Ventures.

Blue Voice, a Boston-based AI startup that provides real-time policy guidance to officers in the field, raised $6 million in funding led by SignalFire and Las Olas VC.

Aila, a Riyadh company whose software works out where a student's understanding has gaps and then develops practice around them, raised $3 million in a pre-Series A funding round led by Rua Growth Fund.

Humain, the Saudi state-owned AI company founded last year, announced a series of partnerships with U.S. startups on Monday tied to data centers in Riyadh and Dammam. Together AI, which rents out computing power to developers who train and run AI models, will share a portion of the revenue from each customer with Humain, which plans to provide the startup with 250 megawatts of electricity and 120,000 Nvidia, AMD and Qualcomm chips.

The Pentagon launched versions of OpenAI’s ChatGPT and xAI’s Grok called “ChatGPT Mil” and “Grok for Government,” to its civilian and military personnel.

Deloitte has acquired substantially all of the assets of Wavicle Data Solutions, a data engineering firm whose specialty is developing the clean, organized data foundations a company needs before its AI will work. Terms were not disclosed.

Athena Agentic, which sells software that runs a company's security operations largely without human staff, bought the Omni Cyber Solutions Agentic Platform, an AI security product, in an asset purchase. Terms were not disclosed.

Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: [email protected].

If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.