t Sierra Ventures’ CXO Summit, Rob Thomas, SVP of Software and Chief Commercial Officer at IBM, sat down with Tim Guleri, Managing Partner at Sierra Ventures, for a conversation on “Crossing the AI Divide.” Speaking to a room of Fortune 2000 CIOs and CTOs, Rob said enterprise AI’s biggest obstacle isn’t the technology. It’s fear, inside companies and in how the industry talks about AI.
Rob’s first message to the industry was simple: stop scaring the customer. Software marketing used to be about landing the top 10 banks, then about having the best cloud. “Now software marketing is our products are so dangerous they’re going to kill you.” Safeguards matter, but fear slows adoption for everyone. He added that past technology waves, from steam to electricity to the internet, kept running for a long time, and by his math AI is about a trillion dollars into a run with roughly six trillion still to go.
Inside large enterprises, Rob sees a barbell. The C-suite wants to move on AI, and so do developers. The people in the middle are the ones who hesitate. At Deutsche Telekom, IBM pitched AI to run network operations: more resilience, less repetitive work. The head of network loved it, and so did the engineers who were glad to hand off work they didn’t enjoy. The real work was convincing the operations leaders in between that the change wouldn’t add risk and would leave the network more resilient than before.
Rob is surprised by how many companies start AI programs without a clear reason. “They’re just saying we want to do AI.” In 2022, IBM set a specific goal: take $6 billion out of the back office and reinvest it in R&D. The work came from automating repetitive tasks across HR, supply chain, procurement, and product management, not from cutting-edge AI.
With a clear goal set from the top, teams can be trusted to find the ideas. Without one, Rob warned that grassroots experiments mostly lead to “a massive increase in token cost.”
IBM is building for a multi-model world, with heavy investment in an open inference layer because Rob expects enterprises to run many more open-weight models. His reasoning is practical. Left to choose, developers will default to the most expensive model, and that doesn’t work at scale. Just as a database needs a query optimizer, enterprises need something that routes each task to the right model.
Rob is skeptical that companies will replace their software stacks with homegrown apps. Writing the code is about 1% of the cost. The rest is maintenance, upgrades, security fixes, and keeping everything else from breaking.
He sees a different risk for application vendors. IBM built an AI layer called Ask HR on top of its HR system and asked every employee to use it for everything from job verifications to promotions. Adoption took 18 months. Once everyone was on it, IBM swapped out the HR system underneath and no employee noticed. When users only touch the AI layer, the system of record behind it becomes much easier to replace, and that puts pressure on price.
Rob thinks quantum is closer than most people realize, with fault-tolerant systems around 2029. He compares it to GPUs, which went from a small slice of computing to a dominant one. Quantum probably won’t follow the same curve, but even a middle share would be significant. He called China the least visible and possibly biggest competitor.
When an audience member raised “harvest now, decrypt later,” Rob pushed back on the urgency. He doesn’t expect encryption to be broken before about 2032 or 2033, give or take two years, and noted that many of the loudest warnings come from vendors with something to sell. Preparation is good practice, but “this is not Y2K. There’s not a precise date.”
Asked how leaders close the gap between C-suite ambition and the rest of the organization, Rob pointed to IBM’s own biggest problem. The company had become so risk averse that every process was built to stop something from going wrong, “and it’s impossible to grow without risk.” Much of the past six years has been about bringing back risk-taking in a culture of “checkers and preventers,” and he said that work isn’t finished.
The other lesson is focus. Over 26 years at IBM, Rob has seen good decades and bad ones, and the bad ones came from believing the company could do too many things at once. IBM is still willing to make long bets, and its first check for quantum was written in 2014. But it only makes those bets where it has a real advantage.