Cisco president Jeetu Patel says AI tokens are not value for money today
Cisco's Jeetu Patel said at the Semafor Tech Summit that AI token costs still exceed the value they deliver at scale. He warned that rising spending, network strain and patching gaps could slow enterprise adoption unless companies use AI more efficiently.
AI may be attracting billions of dollars in investments and becoming a key part of enterprise software, but Cisco president and chief product officer Jeetu Patel believes the economics behind the technology still need work. Speaking at the Semafor Tech Summit, Patel said that the cost of AI tokens remains significantly higher than the value they are currently delivering at scale, a situation that could eventually force companies to rethink how they use AI. His comments come at a time when businesses are rapidly adopting AI agents for coding, research, customer service and workplace productivity. While demand for AI tools continues to grow, Patel warned that the industry must find a balance between costs and returns.
"What's happening is the costs of tokens are far higher than the actual value that these tokens are generating at scale," Patel said during the discussion.
According to him, if the industry fails to achieve that balance, organisations may start reducing their AI usage. “If you don't create an equilibrium there then people just pull back on using tokens and that's actually not good for anyone,” he said.
Interestingly, we are seeing several tech companies like Microsoft finding AI cost to be too high because of Claude, which is why the company is working on improving its current GitHub services. The Verge just recently reported that its sources reveal Microsoft "engineers are being encouraged to start transitioning their workflows to GitHub Copilot CLI in the coming weeks, ahead of the (Claude license) cutoff.
Patel further argued that the solution will likely involve a combination of smaller AI models, open-source alternatives and intelligent routing systems that determine which model is best suited for a particular task. He noted that not every query needs to be handled by a massive frontier model containing trillions of parameters.
AI spending could create a budget shock for companies
One of Patel's biggest concerns is the unexpected cost burden AI could create for large enterprises. As organisations deploy AI agents across teams, token consumption can rise quickly and generate bills that many companies have not planned for. He illustrated the challenge with a simple example.
“Here's the crazy math. Think about if you have every employee in your organization just use $200 of tokens every week. In 50 weeks, a company our size would need $900 million extra that's not budgeted,” he said.
He added that some employees could end up using far more than that amount, making the financial impact even larger. The best example of this is what happened at a tech company recently. It received a bill of $500 million (approximately Rs 4,770 crore) for a month after it forgot to put usage limits on Claude for employees. Axios reported this but didn't reveal the name of the company, but the news quickly went viral on social media.
Patel also pointed to infrastructure costs that go beyond token spending. According to him, AI agents place a heavier load on corporate networks than human workers performing similar tasks.
"On average, an agent consumes 450% more network bandwidth than a human conducting that same exact task," he said.
As a result, enterprises are seeing a sharp increase in infrastructure consumption alongside growing AI adoption. Patel believes this trend is unsustainable unless costs become more closely aligned with business value. Apart from economics, Patel also discussed cybersecurity challenges that are emerging as AI systems become more capable. He said the gap between the discovery of a vulnerability and the appearance of an exploit is shrinking rapidly. While attackers can move within minutes, organisations often need weeks to deploy patches across their systems.
“Patching takes about 45 days to go out and patch and only 20% of the vulnerabilities that get found ever get patched,” Patel said.
To address the issue, he suggested that companies should adopt memory-safe programming languages and build additional safeguards that can protect systems while patches are being rolled out.
AI may be attracting billions of dollars in investments and becoming a key part of enterprise software, but Cisco president and chief product officer Jeetu Patel believes the economics behind the technology still need work. Speaking at the Semafor Tech Summit, Patel said that the cost of AI tokens remains significantly higher than the value they are currently delivering at scale, a situation that could eventually force companies to rethink how they use AI. His comments come at a time when businesses are rapidly adopting AI agents for coding, research, customer service and workplace productivity. While demand for AI tools continues to grow, Patel warned that the industry must find a balance between costs and returns.
"What's happening is the costs of tokens are far higher than the actual value that these tokens are generating at scale," Patel said during the discussion.
According to him, if the industry fails to achieve that balance, organisations may start reducing their AI usage. “If you don't create an equilibrium there then people just pull back on using tokens and that's actually not good for anyone,” he said.
Interestingly, we are seeing several tech companies like Microsoft finding AI cost to be too high because of Claude, which is why the company is working on improving its current GitHub services. The Verge just recently reported that its sources reveal Microsoft "engineers are being encouraged to start transitioning their workflows to GitHub Copilot CLI in the coming weeks, ahead of the (Claude license) cutoff.
Patel further argued that the solution will likely involve a combination of smaller AI models, open-source alternatives and intelligent routing systems that determine which model is best suited for a particular task. He noted that not every query needs to be handled by a massive frontier model containing trillions of parameters.
AI spending could create a budget shock for companies
One of Patel's biggest concerns is the unexpected cost burden AI could create for large enterprises. As organisations deploy AI agents across teams, token consumption can rise quickly and generate bills that many companies have not planned for. He illustrated the challenge with a simple example.
“Here's the crazy math. Think about if you have every employee in your organization just use $200 of tokens every week. In 50 weeks, a company our size would need $900 million extra that's not budgeted,” he said.
He added that some employees could end up using far more than that amount, making the financial impact even larger. The best example of this is what happened at a tech company recently. It received a bill of $500 million (approximately Rs 4,770 crore) for a month after it forgot to put usage limits on Claude for employees. Axios reported this but didn't reveal the name of the company, but the news quickly went viral on social media.
Patel also pointed to infrastructure costs that go beyond token spending. According to him, AI agents place a heavier load on corporate networks than human workers performing similar tasks.
"On average, an agent consumes 450% more network bandwidth than a human conducting that same exact task," he said.
As a result, enterprises are seeing a sharp increase in infrastructure consumption alongside growing AI adoption. Patel believes this trend is unsustainable unless costs become more closely aligned with business value. Apart from economics, Patel also discussed cybersecurity challenges that are emerging as AI systems become more capable. He said the gap between the discovery of a vulnerability and the appearance of an exploit is shrinking rapidly. While attackers can move within minutes, organisations often need weeks to deploy patches across their systems.
“Patching takes about 45 days to go out and patch and only 20% of the vulnerabilities that get found ever get patched,” Patel said.
To address the issue, he suggested that companies should adopt memory-safe programming languages and build additional safeguards that can protect systems while patches are being rolled out.
