Newsletter · · Ashutosh Agarwal

IBM's Profit Warning Rattles the IT Services Thesis as AI Reprices Billable Work - IT Services vs AI - Week of Jul 18, 2026

IT Services vs AI newsletter for the week of July 18, 2026, drawing on podcasts published July 11–18. IBM's profit warning becomes the services bears' best macro data point as operators and analysts quantify AI-driven billable-hour deflation across Accenture, Infosys and Wipro.

IT Services vs AI

Week of Jul 18, 2026: IBM's Profit Warning Rattles the IT Services Thesis as AI Reprices Billable Work

Podcasts published Jul 11–Jul 18, 2026.

TL;DR

IBM had its worst day since 1968, down roughly 25% in a single session, about $72 billion of value gone in two days, after warning that customers are yanking money out of software and mainframes to buy AI servers and memory chips. The line that should worry every services investor came from Bloomberg Intelligence's Anurag Rana, quoted on Tech Brew Ride Home: "Discretionary IT spending is worsening and will likely be the main theme across most software companies when they report results."

The billable-hour deflation story stopped being abstract. On The Data Exchange, a former head of IBM's AI division said clients are now demanding 30–50% price cuts on AI-augmented consulting work, and named Accenture, Wipro and Infosys as the firms under pressure to reshuffle staff and change who they hire.

The clearest operator proof yet that AI compresses implementation revenue: on [Un]Churned, two Deltek executives said Claude Code cut their custom-integration build cycles by 50–60%, cut documentation/training time by 80%, cleared a large backlog, and let them repackage an eight-month implementation as a four-month one, "and we actually sold our first."

What's new

A quiet week on the calendar turned loud on Tuesday. The single biggest event, IBM's profit warning, wasn't about consulting at all, but it handed the services bears their best macro data point in months. Underneath it, the podcasts that actually cover IT services delivered the sharpest, most specific commentary we've heard on how AI reprices billable work. Ranked by how much they should move your thinking on the names:

  1. IBM's pre-announcement is a discretionary-spend siren, not just an IBM problem. On July 14, IBM issued its first profit warning since the early 2000s. Per the numbers read out on TraderMerlin and Valuetainment, preliminary Q2 revenue came in at $17.2 billion, up only ~1% year over year, versus about $17.86 billion expected, with adjusted EPS of $2.93 against a $3.02 estimate, a miss of roughly $600 million on the top line. The stock fell about 25% in a day, its worst since 1968, and host Patrick Bet-David tallied roughly $72 billion of market value gone in 48 hours. CEO Arvind Krishna's own words, read verbatim on Tech Brew Ride Home: "What played out was worse than our expectations… These conditions require our teams to execute perfectly, and this quarter we faltered. We did not adapt and move quickly enough, and numerous large deals failed to close on the timelines we expected." The cause, in plain terms: in the last weeks of June, clients diverted capital toward servers, storage and memory to lock in supply before prices rose further, starving IBM's mainframe and software lines. Why it matters for services: the same budget reallocation that hit IBM's software hits deal cycles everywhere. Jefferies' Brent Thill, on The Exchange, gave three reasons deals stalled and coined the phrase of the week, "AI paralysis": "you effectively have companies freezing, trying to figure out where they're going with AI, and they're in a test phase, so that can stall purchases." He argued this is more IBM-specific than industry-wide, but conceded software has been "a source of funds" as dollars migrate to AI. (Krishna is an operator/insider; Thill and Rana are sell-side analysts; the price and revenue figures are podcast-reported previews and should be treated as preliminary until the full print.)

  2. "Neoconsulting" and 30–50% price cuts, with Accenture, Wipro and Infosys named. The most valuable services-specific commentary of the week came from Evangelos Simoudis, a former head of IBM's AI division who now runs an advisory practice, on The Data Exchange. His clients, and his own firm, are being told to cut rates because everyone knows AI tools are doing the work: "Customers are asking for lower per hour prices, rates, because they know that all of us are using these tools… They're saying, if you're using Gemini, Anthropic, whatever, you cannot be charging me X, you have to charge me 30% of X, 50% of X." He expects legacy firms like Accenture and KPMG to end up running a two-track model, reduced-rate time-and-materials for traditional work, plus a separate outcomes-based model for AI delivery, and warned that "a company like Accenture, 400,000 employees plus," faces "a lot of reshuffling, maybe even some reductions." On the offshore names specifically: "When we talk to some of the companies we collaborate with in India, which provide either engineering services or IT services, they are feeling a lot of pressure to reduce their costs," and for the public ones "you will start seeing missing revenue targets and also how they hire, who they hire, how many they hire." His kicker: the outcomes-based model "plays much better in the hands of hyperscalers like Microsoft" and the vendor-run "neoconsulting" units at OpenAI and Anthropic than in the hands of the traditional integrators. Why it matters: this is the bear thesis sourced to a practitioner, with a magnitude attached to the price give-back for the first time.

  3. "The money is moving, not shrinking." Dave Sobel's Business of Tech reframed the IBM warning as a budget relocation story and stitched it to the labor question better than anyone: "Software renewals traded for AI projects. Senior wages traded for Amplified Juniors. Engineering hours traded for platform subscriptions. One dynamic, three markets." His data points, all attributed: a RAMP study built with Revio Labs (vendor research, so weigh accordingly) found the heaviest AI adopters grew total headcount about 10% over two years and entry-level hiring 12%; CompTIA's read of June data had employers adding nearly 15,000 IT workers, pushing IT unemployment to 2.9%, below 3% for the first time this year, even as tech firms announced another 15,000 cuts. And Service Leadership's 2026 profitability report shows top-quartile managed-service providers pulling away on "service multiple of wages," how much a service commands versus what the labor behind it costs, with the separator being early adoption of service-desk automation. Why it matters: it's the bull-and-bear tension in one place. The winners aren't cutting people; they're re-pricing the hour and moving up into "accountable AI operation." The losers are "defending the only part of the budget that's genuinely shrinking."

  4. Deltek: Claude Code turned an eight-month implementation into four. The week's best operator evidence for SI disintermediation came, oddly, from a customer-retention podcast. On [Un]Churned, Deltek, which sells project-based ERP to government contractors and professional-services firms, walked through what AI did to its own delivery organization. VP Jason Goldsmith: custom integrations that "would take 6, 8, 10, 12 weeks to go through a cycle" now collapse "dramatically, like 50%, 60%, depending on the complexity," and "the custom work is often the long pole in the tent of an implementation." CCO Margo Martin added that they cleared a "huge backlog" ("all of a sudden Jason said, yeah, we're all caught up"), cut new-version documentation and training by 80% ("from months to weeks… to days"), and repackaged a benchmark implementation: "it used to take eight months… now we're taking that down to four. And so we've packaged that. We actually sold our first." Why it matters: implementation and integration hours are exactly what systems integrators bill for. When a software vendor's own services arm can halve them and still sell the shorter engagement, that's the linear-headcount-to-revenue model breaking in real time. Note the nuance: Deltek is running a "blended model," buying agents where it can, building where its ERP data structures are too complex for anyone else, so this is compression, not disappearance.

  5. The counter-current: AI can create more work, and teams can get bigger. Not everyone thinks the pie shrinks. On Odd Lots, Gary Wingens, chair of law firm Loewenstein Sandler, described a due-diligence project whose cost fell 70% with AI, from about "$10 million" to "$3 million," which turned a job the client had refused to pay for into revenue that actually happened (the Jevons paradox: cheaper work means more of it). His pricing tell for anyone modeling services deflation: top-firm hourly rates rose 10.1% in 2025 against roughly 3% inflation, "because the billable hour has actually become more productive and more valuable because of AI tools," even as the total price of a finished solution falls. Firms defend margin by keeping the human quality-control layer that clients won't do themselves. The macro skeptic's version came from Chamath Palihapitiya on All-In, relaying a customer whose "token costs are doubling every 45 days" for "maybe 5% max" productivity gain, and pegging the actual AI-driven EPS lift across the S&P 493 at "somewhere between 0 and 2%." Why it matters: if AI mostly expands the volume of knowledge work rather than deleting it, the demand-destruction case for integrators is weaker than the bears claim, and the ROI reckoning could throttle the AI capex that's cannibalizing IT budgets in the first place. (All operator/practitioner-adjacent commentary; the 0–2% figure is Chamath's on-air characterization, not an audited number.)

The debate

Bull: AI grows the pie and lets providers earn more per hour. The optimistic read has real evidence this week. The Jevons dynamic on Odd Lots, a 70% cost collapse unlocking previously uneconomic work and rising hourly rates, is exactly what an integrator bull wants to see: cheaper delivery expands the addressable universe of projects while the productive hour re-prices upward. Business of Tech shows the heaviest AI adopters hiring more people, not fewer, and top-quartile providers earning a widening premium by moving into "accountable AI operation." Even the Data Exchange bear conceded the shift "is still to be determined… it is not clear how far and how much the target customers will embrace it." And on Tech Lead Journal, a former Gojek CTO argued this is "a cognitive industrial revolution" that redistributes rather than destroys roles: you need "fewer execution-level developers but substantially more business analysts and product managers," because "judgment becomes expensive." Someone has to define the specs, own the guardrails, and be accountable for the output, and that someone can be an integrator.

Bear: AI absorbs 5–25% of billable work and breaks the headcount-growth model. The pessimistic read got more concrete than ever. The Data Exchange put a number on the deflation, clients demanding 30–50% rate cuts on AI-augmented work, and named Accenture, Wipro and Infosys as exposed. The Deltek case shows implementation timelines, the core of SI revenue, halving in practice, and the client happily buying the shorter, cheaper engagement. IBM's warning (The Exchange, Bloomberg Surveillance) proves the demand side is already deferring big deals: "AI paralysis" plus a 25% component-price shock forcing the Global 2000 to scrutinize "software renewals… seat counts." And the disintermediation threat is structural: the Data Exchange argues the new outcomes-based model "plays much better in the hands of hyperscalers like Microsoft" and the vendor-run neoconsulting units at OpenAI and Anthropic than in the hands of the traditional SIs.

The swing factor: whether AI-unlocked new demand (more projects, more litigation, more patents, more agents to build and operate) outruns the deflation of existing billable work, and whether clients keep paying integrators to be the accountable owner of AI outcomes, or route that money straight to the platform vendors. This week the bears landed the harder punches on specifics; the bulls have the better long-run structural argument. Both agree the linear headcount-to-revenue model is over.

Stocks in play

Accenture (ACN). Not discussed on a dedicated podcast this week; the read-through is from the neoconsulting thread. The Data Exchange named Accenture directly as the archetype under pressure: 400,000-plus employees facing "reshuffling, maybe even some reductions," a likely two-track pricing model, and rate give-backs of 30–50% on AI-augmented work. Bull: Accenture is the most likely traditional firm to build a credible vendor-neutral AI practice and capture the "accountable AI operation" budget that Business of Tech says is filling as other lines drain; scale and multi-vendor neutrality are assets when clients want a partner not tied to one hyperscaler. Bear: it is the poster child for the pyramid model the bears say is breaking, and its size makes rapid margin/mix repricing hard. Next catalyst to watch: the fiscal Q4 print (September), specifically new bookings, managed-services growth, and any commentary on GenAI bookings versus pricing give-backs. (One tangential sighting: Accenture ran an ad on Valuetainment touting its Spotify ad-operations automation work, a reminder that Accenture Song is itself selling the automation that compresses creative billable hours.)

IBM (IBM). The most-discussed name of the week by a wide margin. The July 14 warning dominated dozens of shows. Bull: diversification is the defense, as TraderMerlin put it, "if AI all of a sudden comes crashing back down… they'll still be making money off of consulting… infrastructure and software." Consulting is roughly a third of the company; per TBPN, software is ~44% of revenue at ~80% gross margin, consulting ~31% at under 30% margin, and infrastructure ~23% at just under 60%; consulting was guided roughly flat (up ~1% in constant currency, per Schwab Network commentary this week). IBM is also pitching genuinely AI-era products: the Deep View featured IBM's "Sovereign Core" governance software for regulated industries. Bear: the CEO admitted "we did not adapt and move quickly enough"; Watsonx growth of only "19% to 22%" (TraderMerlin) lags the hyperscalers badly; the mainframe/software base is precisely what the AI-capex reallocation is starving; and Wall Street Unplugged flagged rising memory-driven price hikes plus new coding-model competition from Meta and SpaceX as fresh margin threats. Valuetainment aired unverified, Reddit-sourced claims of a morale problem and a US-to-India headcount shift (host's own on-air check: "135,000 employees" in India vs. "40,000" in the US; "2,840 job openings for IBM in India… 376 available in the USA"), treat as anecdote, not fact. Next number to watch: the full Q2 report, whether the failed "elephant deals" rebound in the second half, and whether consulting bookings hold as discretionary budgets tighten.

Infosys (INFY). No direct podcast coverage this week, despite mid-July being India IT earnings season. The only read-through: the Data Exchange named Infosys among the offshore firms "feeling a lot of pressure to reduce their costs" and warned public Indian IT names will start "missing revenue targets" and changing "how they hire, who they hire, how many they hire." Bull: offshore cost arbitrage plus AI could let Infosys deliver AI transformation cheaper than Western firms and win share on price. Bear: the linear-headcount growth model, the heart of the Indian IT engine, is the single most exposed structure to AI productivity, and price-led competition cuts both ways. Next catalyst: Infosys's own quarterly result and, critically, fresher-hiring guidance and any change to its headcount-to-revenue trajectory. We flag the absence of primary commentary as a real coverage gap, not evidence of calm.

Wipro (WIT). No direct podcast coverage this week. Same read-through as Infosys: named on the Data Exchange as an offshore firm under cost/pricing pressure and a candidate to build a vendor-neutral "neoconsulting" practice. Bull: if Wipro can pivot to outcomes-based AI delivery faster than peers, it protects relevance even as the hour deflates. Bear: smaller scale and a historically growth-challenged book make it more vulnerable to both discretionary-spend cuts and AI-driven rate compression. Next number to watch: Wipro's bookings/large-deal TCV and margin commentary on its next print, and any explicit quantification of AI's revenue contribution versus cannibalization.

Read-throughs

TCS, Cognizant (CTSH), Capgemini, EPAM. Zero direct podcast coverage this week; no host or operator named them. The only signal is the sector-wide pressure described on the Data Exchange (30–50% rate give-backs, offshore cost pressure, two-track pricing). Given mid-July is peak India IT reporting, the silence is a corpus gap in English-language podcasts, not a sign these names are unaffected; treat the read-through as directional only.

Enterprise-software vendors whose implementations feed services (CRM, NOW, WDAY, SAP). The clearest data point is negative: on the day of IBM's warning, Tech Brew Ride Home reported "Workday and ServiceNow falling about 6%" on the discretionary-spend read-through, and The Exchange flagged ServiceNow earnings expectations in the single digits. No podcast this week produced fresh, named evidence on Agentforce, ServiceNow AI agents, SAP Joule or Workday agents adoption, a notable quiet after last week's KeyBank Agentforce-doubt story. If enterprise-agent platforms stall in "proof of concept," the SI implementation-revenue tailwind tied to them stalls too; if they scale, they may disintermediate the very integrators that sell the setup. Watch both vendors' next prints for agent-adoption metrics.

Microsoft / GitHub Copilot. Framed this week less as a coding tool and more as the winner of the services reallocation. The Data Exchange argued the outcomes-based model "plays much better in the hands of hyperscalers like Microsoft," i.e., Microsoft's own services and partner motion may capture budget that once went to independent SIs. Bull for MSFT's services optionality; bear for the integrators sitting downstream of it.

Build-vs-buy and in-house AI. The Deltek case is the template: enterprises are running a blended model, buying agents for commodity tasks and building only where proprietary data/complexity demands it. On 20VC, Glean co-founder Arvind Jain offered a striking in-house data point, a triage agent "costing $1M/month that replaced 15 on-call engineers' work," while arguing teams will get bigger, not smaller. The through-line: every dollar a client spends building or renting its own agents is a dollar it isn't paying an integrator to do the work, but the complexity of enterprise data still leaves a large "help me actually deploy this" gap that favors whoever can own the outcome.

What changed vs last week

Last week's issue (Week of Jul 06) left several open threads. Here's the movement:

  • Billable-hour deflation escalated from thesis to numbers. Last week it was practitioner-flavored (TSIA, WSJ, Deloitte). This week the Data Exchange attached a magnitude (30–50% rate cuts) and named ACN/INFY/WIT directly, a clear intensification of the same story.
  • The vendor-funded services land grab deepened. Last week's Microsoft "Frontier"/Amazon/OpenAI-consultant-army thread gained conceptual support: the Data Exchange argues the economics of outcomes-based AI delivery structurally favor the hyperscalers and vendor neoconsulting units over the legacy SIs. No new dollar commitments were disclosed this week, but the "why it works" got clearer.
  • NEW hard macro data point: discretionary IT spend is visibly slowing. Last week had no equivalent. IBM's warning, "AI paralysis," slipped "elephant deals," a 25% component-price shock forcing budget scrutiny (The Exchange, Bloomberg Surveillance), directly advances the deal-cycle-elongation theme with a real, dated example.
  • Agentforce POC-stall: no fresh corroboration. Last week's KeyBank Agentforce downgrade had no follow-up this week; ServiceNow and Workday only appeared as collateral damage in the IBM sell-off (down ~6%). The POC-stall pattern is neither confirmed nor refuted, keep chasing.
  • The pyramid-inversion question got new (if not identical) evidence. Last week's standout was Pat Casey's ServiceNow 7:1→1:1 engineer-to-PM ratio. No one re-quoted a single clean ratio this week, but Tech Lead Journal ("fewer execution developers, more BAs/PMs; judgment becomes expensive") and Deltek's 50–60% cycle compression both point the same direction.
  • Accenture Song: quiet. Last week's Song deep-dive had no sequel; the name surfaced only as an advertiser (Spotify ad-ops automation) on Valuetainment. No update on the unverified breach claim from two weeks ago.
  • Indian IT remains under-covered. Despite mid-July earnings season, TCS, Infosys, Wipro, HCL, Tech Mahindra and LTIMindtree drew zero direct podcast coverage this week, the same chronic gap flagged last week.