Chip stocks sold off last Thursday as the midterm results raised doubts about how fast data centers would get built. The Nasdaq closed down 5.5% — its worst day since April 2025. But by Monday afternoon, the Nasdaq had bounced back nearly 3%, which shows that the market isn’t settled on how to think about things.
For weeks I’ve been tracking one finding: the savings from AI aren’t showing up, in the budget or in the work. Last week’s reading added more evidence, plus a controlled experiment on a fix. Investors are still weighing a slower buildout, which raises the stakes on what organizations do with the capacity they already pay for.
This week’s coverage:
Still No Savings. One Fix Has Data Behind It.
Another round of evidence that saved time isn’t becoming savings, plus the one controlled experiment in the pile on what changes it.
Whoever Takes the Blame Won’t Let the System Decide
A dispatcher explains why a system that works still doesn’t get the final call, and Accenture’s recruiting agents raise the same question.
Workers Are Paying for Savings That Haven’t Arrived
Slower recoveries after AI layoffs, engagement at a 15-year low, and a shift in who makes the workforce calls.
AI Capacity Got More Expensive and More Political
Thursday’s selloff and Monday’s rebound, surcharges and renewal hikes at 56% of surveyed organizations, and Copilot Chat caps Microsoft won’t publish.
Here’s what I was reading.
Still No Savings. One Fix Has Data Behind It.
A CIO.com piece added ActivTrak’s latest numbers. Across 9,140 knowledge workers at 467 enterprises, employees who use AI worked 8.42 to 9.08 hours a day, against 7.58 for those who don’t. Heavy AI users had focus sessions 22% shorter than light users (37.8 minutes versus 48.2) and switched apps 15% more often than non-users. An analyst in the piece noted that the data is limited by what the tool can instrument, but directionally, it still matches what I’ve been reading on this. (Does AI actually save employees time? Data is contradictory)
A Fortune piece on Stellantis found real speedups: a Ram drag simulation that took 14 to 24 hours now returns an estimate in about three minutes. It also found people checking the output. One engineer said he checks 10 to 15 sources and doesn’t think the tools save him time, and a manager asked, “So why am I working so much harder?” Engineers got faster, and Stellantis still reported a second-quarter adjusted operating margin of 1.8%, against a 7% target for 2030. The margin has plenty of causes besides AI, but the speedups haven’t reached it yet. (‘The best thing that you can do for your career is to make yourself redundant’: Inside Stellantis’ AI push after the ‘starting gun went off’)
Week 36 put the share of enterprises seeing real AI ROI at 6%. Last week, I reported Gartner’s finding that 97% of organizations can’t cut expenses through productivity gains alone. ActivTrak and Stellantis back that up — measured from the employee’s side. Saved time doesn’t become savings until someone decides where it goes. (Companies are not saving money with AI, Gartner finds)
An HBR piece added a controlled experiment on a fix. Researchers split 515 early-stage companies and showed one group examples of how others had redesigned their workflows around AI. That group identified 44% more use cases, was 18% more likely to land paying customers, and generated nearly double the revenue. The sample is early-stage companies, and the piece doesn’t show the result lasts or scales. Still, the prescription is concrete: map a workflow end to end, redesign the whole process instead of one step, and expect the bottleneck to move once you fix it. That’s exactly what I’ve laid out in my Cognitive Operating Model framework for AI. If you haven’t read that yet, head over there after you finish this week’s brief! (For Real Performance Gains with AI, Start by Mapping Your Workflows)
Whoever Takes the Blame Won’t Let the System Decide
A dispatcher at a global logistics firm told an HBR author why he keeps adjusting routes from a system he admits often finds better options: if the system makes the call and it goes wrong, “it is my name on the report.” Employees who carry the blame for a system’s decisions will keep overriding it, and they’re reading the incentives correctly. The piece’s fix is four questions to answer before a system scales: who loses discretion if it works, who gains authority and over which decisions, who explains outcomes when the system is followed, and who absorbs the risk while people learn to trust it. For systems already live, testing it is simpler. Turn the system off tomorrow and see whether any decision changes. (Why Even Your Best Employees Override AI Systems That Work)
A Register piece ran the question the other way. Accenture’s chief leadership and HR officer said agents now screen resumes and schedule interviews, time-to-hire fell 28%, and a human always makes the hiring decision. The same piece noted a February memo making regular, tracked AI adoption a requirement for promotion to top roles. The HBR UPS example, doesn’t agree: the move that worked there was tracking whether drivers followed the system, not whether they simply used it. Measuring usage tells you someone has the dashboard open. It doesn’t tell you who made the decision, or why. (Accenture AI agents are sifting through applicants)
Workers Are Paying for Savings That Haven’t Arrived
A Fortune piece on a Goldman Sachs note said workers laid off from AI-exposed jobs took about a month longer to find new work and were roughly 10 percentage points more likely to still be unemployed a year later. Workers 35 and older took about five weeks longer, against about three for younger workers. Goldman inferred AI displacement from occupation exposure, since no survey asks whether AI cost someone the job. Challenger counted 120,136 AI-cited cuts through September. (Goldman says AI layoffs are starting to leave a ‘long-lasting’ scar, just like the last 40 years of history)
An HR Dive piece on Gallup’s third-quarter survey put U.S. employee engagement at 29%, the lowest since 2011. The sharpest declines came from hybrid and remote employees at organizations adopting AI. Only 46% of employees strongly agreed they knew what was expected of them, down from 49%. The piece doesn’t establish that AI caused the drop. (Employee engagement hits 15-year low, Gallup finds)
An HR Dive piece on an LHH survey of 600 leaders found 66% of HR leaders expect layoffs in the next 12 months, nearly double the 2024 share. HR led outplacement decisions at 58% of organizations in 2024. It leads them at 30% now, while the CFO’s share went from 6% to 18%. LHH’s prescription is continuous workforce planning, with reskilling, redeployment, and career transition built into one strategy, and 99% of leaders said they already look at alternatives to layoffs. (Nearly two-thirds of HR leaders expect layoffs in 2027)
Redeployment only works if someone knows where the work went. Cut headcount against savings that haven’t shown up, and the cost lands on workers who take longer to recover.
AI Capacity Got More Expensive and More Political
The Nasdaq fell 5.5% on Thursday, the S&P 500 fell 4.4%, and the Dow lost 1,833 points. Chip stocks led the drop on fears that Democratic gains in the midterms would slow the data center buildout, and the Philadelphia chip index fell more than 12%, its worst day since 2020. (Wall Street’s AI chip selloff, via AP)
Monday brought a rebound. The S&P 500 was on pace for its best day since April, and the Nasdaq was up about 2.8% in afternoon trading, helped by a drop in oil prices, though the S&P 500 and Nasdaq were still slightly below their Election Day closes. (Dow rallies 1,200 points, Nasdaq jumps nearly 3% as stocks climb out of postelection slump) The rebound doesn’t answer the capacity question, because the ballot results and the price increases below haven’t changed.
Voters sided against data centers in nearly every local measure on the ballot. VentureBeat counted 37 of 38 with votes in, CIO Dive reported all but one of 41. Both tallies are unofficial. A Stratechery piece read the result as a new constraint on AI: the last limit is political, which is harder to solve and interferes with solving the rest. The argument was that AI capacity may stay scarce far longer than anyone planned. A CIO Dive piece added UBS’s estimate that nearly one-third of expected 2028 data center additions are at risk. VentureBeat’s midterm coverage cautioned that the election doesn’t establish that existing capacity will disappear or cloud prices will rise. Separate what’s already operating from what’s still awaiting approval. (AI on the Ballot) (What the ‘Blue Wave’ in the 2026 U.S. midterm election means for the American AI industry and businesses) (What does a midterm power shift mean for AI policy?)
The bills are already rising. A VentureBeat survey of 156 organizations (self-selected, so it describes that group only) found 56% had a vendor add an AI surcharge, raise a renewal citing AI, or move features they already had into a pricier tier in the past year. Microsoft raised select Microsoft 365 list prices on July 1, by 5% to 13% on enterprise suites with Teams. Cost overruns delayed, scaled back, or canceled an AI project at 38%, and only 12% named a compute shortage. Of 128 respondents running GPUs, 73 put production utilization at half or less and 42 don’t measure it. (The AI race is raising companies’ bills, from GPUs to memory to Microsoft 365)
A Register piece covered Microsoft’s decision to apply usage limits to more Copilot Chat features for users without a paid Copilot license. Its own guidance says hitting a limit “might indicate interest in additional Copilot capabilities.” A license request is a purchasing decision triggered by friction, and it’s showing up before the org can prove a license would improve the work. The rollout began Nov. 2 for tenants with 2,000 or fewer seats and Nov. 9 for larger ones. As a Microsoft partner, I joined a partner call that confirmed this directly: the limits cover things like image generation, they’re unpublished, and admins can’t see or change them. The same call put usage-based billing for new Microsoft 365 Copilot business subscriptions off to the first quarter of 2027, with no date — so metered pricing is still ahead. (Copilot Chat has usage caps, and Microsoft would rather you found them by hitting them)
Measure utilization before buying more, attach a finance-reviewed return to every AI renewal, and decide before users hit a cap whether they can buy licenses themselves. In the VentureBeat survey, only 34% of respondents tracking infrastructure returns had a figure finance had reviewed.
The evidence that AI savings don’t arrive on their own isn’t new, and last week added more of it. What changed is who’s making the workforce calls: HR led outplacement at 58% of organizations in 2024 and leads it at 30% now, while the CFO’s share tripled. The one experiment in last week’s reading that moved revenue gave teams a clearer picture of their work, not a bigger budget. Someone has to own that redesign before the next renewal or the next layoff round, and the people deciding on both are increasingly the ones furthest from the work.
That’s it for this week’s BeAIReady brief!
If you appreciate the depth of reporting and how I connect the dots, please like, share this post, and subscribe (or share the Brief with a friend!). Thanks!
~erick