Nvidia never disclosed the CentML acquisition price. The Logic reported in July 2025 that the deal was worth well over US$300 million to shareholders and could pass US$400 million with performance earn-outs. CentML, a Toronto startup that made AI models cheaper to run on chips, shut its own service on July 17, 2025, and its founders moved to Nvidia.
Updated October 28, 2026. Corrected the price to the reported range, since Nvidia never disclosed it, and added the shutdown date and FAQ. Nvidia’s move also raises the case for owning your stack, which is the bet behind open-source AI in Canada.
The price is a vote of confidence in Canadian deeptech, and also a reminder that the best Toronto lab spinouts still tend to end up inside American companies.
What CentML Actually Does
CentML does not build models. It makes them faster and cheaper to run, the layer between raw GPU power and model performance. Its software finds smarter paths through compute traffic so workloads get done quicker.
Inside the Deal and the CentML Acquisition Price
The Logic reported that Nvidia won a competitive bidding process. Its sources put the value at well over US$300 million to shareholders, with earn-outs tied to undisclosed performance targets that could lift it past US$400 million. Neither company has confirmed a figure.
CEO Gennady Pekhimenko, a University of Toronto associate professor and Vector Institute faculty member, became Nvidia’s senior director of AI software. The other co-founders took management roles, and LinkedIn data cited by The Logic showed at least 18 more technical staff moving in June 2025. Not every employee made the switch.
Why This Hits Different for Canada
CentML is a clean example of the Canadian pattern. University research, a Vector Institute connection, a Toronto incubator in Radical Ventures, a US$27 million seed round in October 2023 led by Google’s Gradient Ventures with Nvidia in it, and then a US buyer within two years. Total venture funding was reported at about US$30.9 million.
The research was Canadian. The intellectual property now sits with a California company. Some of the people stay in Toronto, on Nvidia’s payroll. That is good for those engineers and for the investors, and it does little for Canadian ownership of the AI stack. One exit does not make a pipeline, so watch whether the next founders build here or sell here.
What to Watch Next
Our read is that CentML’s optimization work will show up inside Nvidia’s own software stack. Nvidia has not said so. If it happens, the efficiency gains reach customers only on Nvidia hardware, which tightens the lock-in that made CentML useful in the first place.
What this means for a Canadian business buying AI compute
CentML’s customers lost their service on July 17, 2025, with a few weeks’ notice. That is the practical lesson for any firm that relies on a small AI infrastructure vendor. Before you sign, ask what happens to your workloads if the vendor is bought, get your data and configurations in a portable format, and keep a second route to run the same models. Inference cost is the line most firms will feel first, and the tools that cut it are exactly the ones big chipmakers want to own.
Frequently asked questions
How much did Nvidia pay for CentML?
The price was not disclosed. The Logic reported it at well over US$300 million, rising past US$400 million if performance earn-outs are met.
What did CentML do?
CentML built software that sat between AI models and the chips running them, squeezing more work out of under-used GPU capacity so models ran faster and cheaper.
Who founded CentML?
CentML was co-founded in 2022 by University of Toronto professor Gennady Pekhimenko with Akbar Nurlybayev, Shang (Sam) Wang and Anand Jayarajan. Pekhimenko now leads AI software work at Nvidia.
When did Nvidia acquire CentML?
The acquisition was reported on June 27, 2025, and CentML ended its customer service on July 17, 2025.
What does the CentML deal mean for Toronto?
It confirms Toronto research talent has global buyers. Whether it builds a pipeline depends on the next founders scaling here instead of selling early.
More in the same thread on our Canadian AI companies reference.
