Articles

Open Weight AI Models:
Mozilla's CTO on Who Is Ready for Adoption

Mozilla's CTO says large companies with well-staffed IT teams are ready to adopt open-weight AI models, but most smaller businesses are not. He cites complexity and a lack of talent as obstacles to adoption.

Key Points

  • Mozilla's CTO reports that enterprises whose open model projects did not reach production cited complexity, a lack of talent, and trouble with hardware and software at scale.
  • Companies do not have to host open weight models themselves, Mozilla's CTO notes, because outside providers are starting to run them, but he still sees usability as an obstacle.
  • The time to test cheaper models is when a workflow reaches production and before it scales.

Raffi Krikorian, the CTO of Mozilla and an advocate for open weight AI models, considers large companies ready to adopt them today and advises small and medium-sized businesses to wait. An open weight model is a file, or a set of files, that a company can download and run on its own computers, and it is what most people now mean by an open model. Anthropic's Fable, one of the frontier models from the big AI labs, runs on Anthropic's servers, and customers can't run it on their own. Running a model in-house gives the company what Krikorian calls "full control over what's happening," at the cost of operating the computers and paying the power bill.

In Krikorian's view, success depends mostly on the talent on a company's team, the time the team has, and access to the company's own data for evaluation. He considers Fortune 500 and Fortune 1000 companies ready now, citing their well-staffed IT teams and relationships with good vendors. He places a mid-sized organization "in the realm of people that can pull this off" if it has a good vendor or consulting agreement. To small and medium-sized businesses, and to companies whose business has nothing to do with technology, he said, "This is not the year to go try the open models yet."

Three reasons enterprises gave for stalled projects

Krikorian reported three reasons enterprises gave Mozilla when their open model projects did not reach production: the work was too complicated, the team lacked the talent, or the company could not find the right hardware and software combinations to run at scale. The answers come from the survey behind the first version of Mozilla's State of Open Source AI report, and in his view "all those things are still at play right now." He conceded the gap when a CXOTalk viewer asked about organizations that lack the skills and experience to run these models: "I am preaching a dream, but we do need to close this implementation gap."

Hosting is becoming optional

Companies do not have to host a model themselves, Krikorian noted: "There are now starting to be NeoClouds and API providers who are running these open models for you under pretty good license terms." A company pays one of them by credit card, as it would pay OpenAI or Anthropic, at a price he put at "like a 50th to a 35th of the cost."

Usability remains a problem, as Krikorian acknowledged when the host and viewers kept returning to it. He put the fault on technologists like himself: "The failure here is that we the nerds need to do a better job of explaining it, making runbooks, making double-click installers and stuff like that."

Start with a backup plan

Krikorian advises most CTOs to start evaluating a backup plan for a provider shutdown, an outage, or runaway costs, and he ties that plan to open models. He wants people to choose the right model at the right time and price, with open models as one option. Switching now is another matter: "My practical advice for most CTOs is never switch today."

At the enterprises Krikorian talks to, the pressure usually comes from more than one office. Some CFOs raise the cost of AI usage, in his telling: "Our token spend is out of control. We spent $1 million this month alone on tokens." At others, he reported, IT and security teams want confidential business IP to stay on-premises.

CEOs usually ask about resilience because, by his account, they watched Anthropic temporarily turn off its newly released Fable model in June after a U.S. government order. All of those concerns, Krikorian concluded, "point to, in my opinion, just having a good backup plan." He led infrastructure at Twitter, and he advises companies to think about model providers the way a data center treats power: "You would never just run a single power line into a data center because the day a squirrel chews through that line, the squirrel's going to have a bad day, but your data center's offline."

Test cheaper models on routine work

Open weight models suit, in Krikorian's view, "the boring stuff that a business needs to do, like summarize an email or like manage a calendar." He would keep scientific research on frontier models. A frontier model may do better on a single attempt, he acknowledged, but a cheaper model allowed several attempts for the same money may reach similar quality.

Krikorian puts the time to test at "when you get to production and before you hit the scale button." Earlier in a project he would stay on a frontier model, because "you don't exactly know what you're doing," and he makes an exception for what he called "a superpowered engineering organization." When Mozilla's finance team started using Fable for financial workflow analysis, his reaction was, "There is no reason that you need something that's basically trying to drive a Ferrari to Whole Foods." He would first try something simpler than Fable "before you even try moving to an open weight model."

The next tier of companies

Krikorian expects to be talking soon about the next tier of companies, after two earlier examples. About a year ago, by his account, Pinterest, a Silicon Valley company with ample engineering staff, moved to open models that it hosts itself. Thomson Reuters, in the same account, is not a technology company but has strong engineers, and it has since adapted a model to its own internal data.

Krikorian told host Michael Krigsman, "Today, it's hard, but it's easier than yesterday, and tomorrow will be easier than today."

Watch the full conversation with Raffi Krikorian and read the complete transcript on the episode page.

CXOTalk prepared this article with AI assistance from the verbatim transcript of episode 933. Quotations are unedited from the broadcast.

Related

Mozilla CTO: Open Source and Open Weight AI, Rent or Own?

Mozilla CTO: Open Source and Open Weight AI, Rent or Own?

Mozilla CTO: Open Source AI Agents and the Fight for Control

Mozilla CTO: Open Source AI Agents and the Fight for Control

Deloitte CTO on the AI Investment Trap: CIO Advisory 2026

Deloitte CTO on the AI Investment Trap: CIO Advisory 2026

Box CEO Aaron Levie: CIO Advice on Agentic AI and the Enterprise

Box CEO Aaron Levie: CIO Advice on Agentic AI and the Enterprise

Enterprise AI Adoption: A Board Member’s Perspective

Enterprise AI Adoption: A Board Member’s Perspective

Inside AI Strategy with Google Cloud's CTO

Inside AI Strategy with Google Cloud's CTO

Inside AMD’s AI Strategy with EVP and CTO Mark Papermaster

Inside AMD’s AI Strategy with EVP and CTO Mark Papermaster