A patent-infringement lawsuit involving chips for AI workloads has made its way into court almost five years after it was filed, with the founder of Singular Computing demanding Google pay him $1.67 billion for allegedly copying his company’s innovations.
Joseph Bates, a computer scientist who founded Singular Computing in 2005, is claiming that Google used some of his company’s technology to use in its Tensor Processing Units (TPUs), the chips the IT giant created to accelerate AI workloads. In particular, Bates is alleging that Google used Singular Computing’s innovations for versions 2 and 3 of its TPUs.
A jury was selected last week in federal court in Boston, with attorneys for both companies making their opening arguments, and it expected to continue this week. The trial is scheduled to last at least two weeks.
According to a Reuters report, Bates’ attorney, Kerry Timbers, said that between 2010 and 2014, he shared the patented technology used to create his processors with Google, which rather than licensing the innovations instead copied them and used them in its TPUs.
Timbers also told jurors about internal Google emails, in which no-chief scientist Jeff Dean wrote to colleagues that Bates’ innovations could be “really well suited” for the chips Google was developing. In another email, another Google employee said the company was “quite corrupted by Joe’s ideas.”
Reuters wrote that Timbers argued that “this case is about something we all learned a long time ago: respect for others, don’t take what doesn’t belong to you, and give credit where credit is due.”
Google Is Pushing Back
Google has denied the allegations, with its lawyer, Robert Van Nest, telling jurors that the Google scientists responsible for developing the TPUs never met Bates and were not helped by the employees who did. Van Nest referred to Bates as a “disappointed inventor.”
Bates, he argued, had met with a number of top tech companies – including Meta, Microsoft, Amazon and OpenAI, which ignited what is now a rapidly expanding generative AI space with its ChatGPT chatbot – in hopes of using technology in their AI chips.
The lawyer said the problem was Singular’s use of approximate math in its technology, which can generate “incorrect” calculations.
“Google’s chips are fundamentally different, fundamentally different, than what is described in Singular’s patents,” Van Nest reportedly told the jury.
In court filings, Google also claims that Bates initially demanded $7 billion from Google for the alleged copying of Singular Computing’s technology, although the demand at trial is for $1.67 billion.
Google’s TPUs underpin the AI features used in a range of its services, including Gmail, Google Translate and Google Search, so the case will have a significant impact – beyond the monetary payout – on the tech giant if the jury sides with Bates.
Both Google and Singular Made Mistakes: Analyst
Rob Enderle, principal analyst with The Enderle Group, told Techstrong.ai that while the case against Google involves a trendy technology like AI, this kind of situation – where a smaller company presents its strategic technology to a larger vendor without a contract in place– is not uncommon and needs to be done carefully.
“And it is a dual edged sword because not only can the company presenting lose its IP, but the recipient company, in this case Google, can be compromised even if they didn’t take the technology because a similar internal development is now effectively compromised,” Enderle said.
The fact that AI technology is involved is what’s driving the high damage request. That said, such lawsuits are becoming less common than in the 1990s and early 2000s, now that vendors are doing a better job of protecting their IP and protecting against such litigation. Still, “the fact that Singular Computing is able to maintain the litigation suggests both they and Google screwed up here,” the analyst said.
There are lessons learned for both sides. Companies like Singular Computing should hire a law firm specializing in IP before presenting their technologies to a huge company like Google. For Google and others, the lawsuit is another argument for buying processors from chip makers that have more robust IP defenses rather than building them.
“This is one of the reasons why you work with companies like Nvidia, Intel, Qualcomm, or AMD,” Enderle said. “They can protect against this class of litigation and are far less likely to make the mistake that Google likely made here, thus better avoiding the negative publicity and financial risk.”
A Booming AI Chip Market
Still, Google isn’t the only top-tier IT company developing its own AI-optimized chips to run machine learning workloads. Microsoft, Meta, IBM, Apple, Qualcomm, OpenAI and Amazon Web Services (AWS) all are developing or have developed their own AI processors.
Meanwhile, major chips makers like Intel, AMD and Nvidia continue building out their AI processor portfolios.
The global market for AI chips is expanding quickly, with Allied Market Research analysts expecting it to increase from $14.9 billion in 2022 to $383.7 billion by 2032, an average annual growth rate of 38.2%.
In a report, the analysts noted a broad spectrum of use cases for AI chips, such as AI-centric products like smartphones that include AI-powered cameras and smart speakers with voice-activation helps.
“Enterprises utilize AI chips within data centers to tackle tasks such as training expansive machine learning models and refining business operations through predictive analysis,” they wrote. “Research institutions and academia leverage AI chips to push the frontiers of AI research, delving into novel algorithms and architectures.”
Consumers – the end users of AI chips – get improved products and services that call for accelerated AI processing, the analysts wrote, pointing to tailored streaming recommendations and intelligent home automation.
