Kyndryl, the IT infrastructure services company spun out of IBM in 2021, has opened an AI Innovation Lab in Frisco, in the Dallas-Fort Worth area, and expects to create up to 300 skilled jobs there over four years in AI, technology consulting and design engineering. It joins the company's existing AI labs in the UK and Luxembourg.
The lab runs clients through a four-stage process of immersion, co-design, build and roadmap, with workshops and rapid prototyping intended to take a customer from an exploratory idea to a working prototype in days. It also houses a team working on systems for motor vehicle agencies, the state bodies that handle driver licensing and registration, and a demonstration of Kyndryl's cyber defence operations centre.
"The future of AI will be built through collaboration," said Jamie Rutledge, president of Kyndryl US. The lab's partners include the nonprofit technology training organisation NPower and the local groups Dallas AI and Tech Titans.
The Research Published The Same Day Explains Why
Kyndryl released a survey on the same day that sets out the problem the lab is meant to address. Across 2,000 business and technology leaders in 12 industries on five continents, 98% said their modernisation projects had suffered delays and negative business effects.
The reason, in the survey's findings, is that most organisations do not know what they are running. Only 9% said they fully understand the dependencies between their applications, and a quarter reported large numbers of undocumented applications. A company that cannot say which systems depend on which cannot safely change any of them, which is the practical barrier between an AI pilot and a production deployment.
AI has now overtaken both cost reduction and legacy replacement as the main reason companies modernise at all, and 63% expect their use of sovereign cloud to grow.
Agentic AI Is Still Rare In Production
The same survey found that only 10% of organisations have deployed agentic AI in production for modernisation work. Among those that have, 67% reported being on track or ahead on their modernisation goals, against 51% of everyone else.
Kyndryl frames that as evidence the technology works, though the comparison more likely reflects which organisations are capable of deploying it. Companies that already document their systems and run disciplined engineering are both more likely to put agents into production and more likely to hit their targets.
The Gap Between Deployment And Results
Kyndryl's earlier research this year sharpened the same point. Its People Readiness Report in June, based on 1,100 senior leaders across eight countries, found AI embedded in core business processes or deployed broadly at 57% of enterprises, while just 11% had achieved both of their top two AI objectives.
Workforce readiness was falling rather than rising. Only 23% of leaders said their workforce was fully prepared for AI, down from 29% a year earlier, 52% reported difficulty hiring people with AI skills, and only a third of organisations had clear policies on which decisions AI may and may not make.
That last figure matters most for anyone deploying agents, since an agent without a defined decision boundary is an open liability, a question that runs through every deployment where AI agents run devices and systems on a company's behalf.
What Kyndryl Gets Out Of It
The lab also serves a commercial need. Kyndryl reported first-quarter revenue of $3.6 billion for the three months to June, down 3% on a year earlier, as older infrastructure contracts inherited from IBM continue to run off.
The growth sits elsewhere. Kyndryl Consult revenue rose 10% year on year, hyperscaler-related revenue passed $530 million and grew 34%, and signings over the trailing twelve months reached $14.2 billion. A lab that pulls customers into short, intensive projects is a route into exactly that higher-value consulting work, and a way to start conversations about modernisation that lead to longer engagements.
Dallas, Not Silicon Valley
The location says something about where enterprise technology work is going. North Texas has drawn corporate headquarters and technology operations for years on lower costs and a large labour pool, and Frisco has been among the fastest-growing cities in the US.
For a company selling to banks, insurers, airlines and government agencies, proximity to those customers matters more than proximity to AI research. The lab's dedicated motor vehicle agency team points the same way, since state agencies running decades-old systems are a natural market for a company whose business is keeping and modernising exactly that kind of infrastructure.
What Companies Should Take From The Numbers
The useful lesson for technology buyers is in the survey rather than the ribbon-cutting. If 98% of modernisation projects run late and only 9% of organisations fully understand their application dependencies, then the first step towards useful AI is usually documentation and discovery rather than model selection.
Before committing to an AI programme, companies should know which systems hold their data, what depends on what, where undocumented applications sit, and which decisions they are willing to let software make. Those answers determine whether a prototype built in days can ever reach production, and getting real value from existing operational systems is a discipline in itself, as we found in how Datanomix turns machine data into manufacturing decisions.
The Bottleneck Is Not The Model
Kyndryl's lab is a modest commitment by the standards of AI spending, and 300 jobs over four years will not reshape the Dallas economy. Its interest lies in what the company is betting on: that the constraint on enterprise AI is no longer the capability of models but the state of the systems they have to work with.
The survey behind the announcement supports that view, and it is a harder problem than buying access to a frontier model. If Kyndryl is right, the companies that get the most from AI over the next few years will be the ones that spent the time mapping what they already run.
