IBM's Q2 miss: enterprise Infrastructure fell while AI physical capex crowded out the stack

International Business Machines $IBM did not wait for its July 22 earnings call. On the morning of July 14 it furnished a letter from CEO Arvind Krishna with selected preliminary second-quarter results — and shares plunged more than 20% in the session (premarket print and early trade).

If you only read the headline ("IBM missed; AI spending hurt"), you will miss the useful distinction inside the filing. IBM's own Infrastructure segment declined. Its Distributed Infrastructure line — Power and Storage — posted its best reported quarter. The same letter says clients redirected late-June capex toward servers, storage and memory to lock in supply-constrained kit ahead of expected price increases.

That is not a story in which "infrastructure spending died." It is a story in which one enterprise stack lost the budget fight to another physical stack in the same AI buildout cycle. The research question for July 22 is whether the gap is a timing slip on large deals — or a more durable shift in how enterprises rank IBM's Z/software complex against the GPU server, HBM and fabric bill that hyperscalers and large buyers are currently securing.

Research, not a price target.

Primary source

Primary: IBM Form 8-K filed July 14, 2026, accession 0000051143-26-000070, Exhibit 99.1 (Letter to IBM Investors) and the GAAP↔operating and cash-flow reconciliations in that same exhibit package. Figures are preliminary; IBM warns finals may differ slightly.

Research assembled with Aether: the July 14 exhibit is in Aether’s press/transcript corpus (IBM:2026:QX:press:000005114326000070). Z-cycle context below comes from Aether-indexed 10-Qs — the Q1 2026 Form 10-Q (accession 0000051143-26-000038) and the Q3 2025 Form 10-Q (accession 0000051143-25-000064).

What the July 14 letter actually says

Measure	Preliminary Q2 2026	vs Q2 2025 (same exhibit)
Revenue	$17.2B, +1% YoY	—
Software	+5%	—
Consulting	flat (+1% constant currency)	—
Infrastructure	−7%	—
GAAP gross profit	$9,907M (margin 57.7%, −100 bps)	$9,977M / 58.8%
Operating (non-GAAP) gross profit	$10,194M (margin 59.4%, −70 bps)	$10,202M / 60.1%
GAAP pre-tax income	$2,479M (margin 14.4%, −90 bps)	$2,597M / 15.3%
Operating (non-GAAP) pre-tax income	$3,290M (margin 19.2%, +30 bps)	$3,197M / 18.8%
GAAP diluted EPS	$2.27, −2%	$2.31
Operating (non-GAAP) EPS	$2.93, +5%	$2.80

Operating EPS bridge from the exhibit (Q2 2026): GAAP $2.27 + acquisition-related $0.58 + retirement-related $0.08 = operating $2.93.

Year-to-date cash (six months ended June 30, 2026), per the free-cash-flow reconciliation in Exhibit 99.1:

Cash bridge ($ millions)	Amount
Net cash from operating activities (GAAP)	$7,766
Less: change in IBM Financing receivables	2,264
Op. cash excl. Financing receivables	5,503
Capital expenditures, net	(743)
Free cash flow	$4,760 (~$4.8B)

Consensus comparisons in the press (Street around ~$17.9B revenue / ~$3.01 operating EPS) matter for why the stock moved. For filing-first work, the more durable facts are the segment mix, the reconciliations above, and Krishna's stated mechanisms.

Call / webcast: IBM's regular quarterly call is Wednesday, July 22, 2026 at 5:00 p.m. ET (investor event page).

Why Q2 was always going to look soft — then looked softer

Mainframe programs are cyclical. IBM's own Q3 2025 10-Q (Aether-indexed) shows how strong the z17 launch was: Infrastructure revenue +17.0% (+15.1% constant currency), with IBM Z +61.1% (+59.0% cc) in that quarter, described as "early success of our z17 platform."

The subsequent Q1 2026 10-Q (also Aether-indexed) shows the cycle still hot into March: Infrastructure revenue $3,326M, +15.3% (+11.7% cc); Hybrid Infrastructure +28.1%; IBM Z +50.9% (+48.3% cc); Distributed Infrastructure +16.7% (+13.1% cc).

Krishna's July 14 letter explicitly says that in April — wrapping that launch — IBM told investors it expected Infrastructure revenue to decline low-single digits for the year, beginning in Q2. So a soft Infrastructure print was telegraphed. What was not telegraphed at this magnitude was a −7% outcome driven by Z + Transaction Processing shortfalls and slipped large deals.

The miss is concentrated: Z + transaction software + delayed large deals

Krishna's letter is blunt about attribution:

After the strongest mainframe-program start in IBM history (z17), management expected Infrastructure to decline low-single digits for the year, beginning Q2.
Q2 was worse than that plan, driven by a shortfall in Z performance and the associated software stack, primarily Transaction Processing.
In the last few weeks of June, clients shifted quarterly capex toward servers, storage and memory to secure supply-constrained infrastructure ahead of expected price increases.
IBM "anticipated some supply-chain-related impact" but "did not anticipate the magnitude of the capex reprioritization."
Clients were also "distracted with rapidly-evolving, industry-wide cybersecurity concerns."
Execution: "we faltered… numerous large deals failed to close on the timelines we expected, driving the majority of our shortfall."

That last sentence matters. Management is not claiming the entire miss is a secular AI story. It is claiming deal timing + imperfect execution under a real budget reallocation. July 22 is when that claim gets stress-tested with backlog, book-to-bill language and full-year guidance.

Two infrastructures, not one

"AI infrastructure" in market vernacular usually means:

Accelerators and HBM — NVIDIA $NVDA, AMD $AMD, Broadcom $AVGO ASICs, Micron $MU / SK Hynix $SKHY memory
Servers, storage arrays, networking and optics (the physical data-center bill of materials) — names such as Dell $DELL, HPE $HPE, Super Micro $SMCI on the server side of that scramble
Foundry and packaging capacity — TSMC $TSM and OSATs — that make that kit scarce

IBM's segment called Infrastructure is a different animal. In this letter it includes the Z mainframe franchise and related software attach, with Distributed Infrastructure (Power, Storage) called out separately as a positive.

Inside the same quarter (Exhibit 99.1 highlights):

Line IBM highlighted	Direction	Detail from Exhibit 99.1
Infrastructure (segment)	−7%	Z + transaction software shortfall
Distributed Infrastructure	+37%	"best performance in reported history"; Power + Storage; ~$500M backlog exiting quarter
Red Hat (within Software)	+11%	sequential acceleration
Recent acquisitions (HashiCorp, Confluent)	"strong performance"	called out by name in the letter
z17 program-to-program	~130%	still ahead of z16 (prior record); clients representing 85% of installed MIPs maintaining or growing capacity
Consulting signings	growth	"strong GenAI contribution"
Operating (non-GAAP) PTI margin	+30 bps	productivity initiatives cited

So the paradox that moved the stock is:

Buyers are still buying IBM-adjacent physical kit (Distributed Infrastructure up sharply).
Buyers delayed or starved IBM's high-ASP enterprise Z + transaction software machine at quarter-end.
The same buyers, according to Krishna, redirected spend into the commodity-like AI physical stack (servers / storage / memory) that is short and expected to reprice higher.

That is crowding out inside the IT budget, not proof that enterprise software demand has vanished.

How this differs from the broader data-center buildout narrative

The TSMC / accelerator narrative that dominates FinTok is about building capacity for AI training and inference: wafer starts, N3/N2, CoWoS, HBM, rack power. Capex is rising because physical constraints bind — the same theme as our TSMC Q2 preview.

IBM's letter describes a second-order effect of those same constraints on enterprise quarterly budgets:

When memory, servers and storage become supply-constrained and expected to get more expensive, CIOs protect that spend first.
Large enterprise deals that are long-cycle, committee-gated and software-attached — classic mainframe / transaction processing — slip to the next quarter when the budget is raided in the last two weeks of June.
IBM's Distributed hardware line can even benefit from the same scramble (Power/Storage +37%, $500M backlog) while the Z franchise has a worse outcome than IBM's already-cautious low-single-digit decline plan.

In other words:

Hyperscaler / AI DC buildout = spend more on scarce physical capacity.
IBM Q2 Infrastructure miss = that same scarcity reordered mid-market and large-enterprise quarterly closing, and IBM's Z/software complex was on the wrong side of the reorder in late June — plus deal slip and cybersecurity distraction.

Conflating the two into "AI ruined IBM" or "AI is fake because IBM missed" is category error. The letter supports a narrower claim: AI physical procurement cannibalized IBM's planned Z/software closes this quarter.

Lightwell and quantum — disclosed, but not the Q2 miss

The letter also unveils narrative-heavy initiatives that should not drown out the Z scorecard:

Lightwell — a $5B commitment, "more than 20,000 engineers", framed as an enterprise clearinghouse for open-source vulnerabilities; GA announced July 8. Early adopters named include Bank of America $BAC, BNY $BK, Citigroup $C, Goldman Sachs $GS, JPMorgan Chase $JPM, Mastercard $MA, Morgan Stanley $MS, RBC $RY, State Street $STT, Visa $V, Wells Fargo $WFC.
Quantum — letter of intent for the Anderon quantum wafer foundry with $1B CHIPS incentives and $1B IBM cash; plans to invest more than $10B in quantum over five years; on track for a large-scale fault-tolerant machine by 2029.

These are real disclosures. They do not explain the Q2 Transaction Processing shortfall. Keep them in a separate mental bin on July 22.

Did the market overreact?

A >20% session move after a preliminary miss of this narrative weight is extreme for IBM historically (press notes it may be among the stock's worst one-day declines). Extreme ≠ automatically "wrong." Filing-first checklist:

Arguments that the move prices something real

Revenue and operating EPS missed Street by a non-trivial gap in a late-cycle AI tape that has little patience for enterprise timing stories.
Management already guided for Infrastructure to decline — and then printed −7%, worse than the communicated low-single-digit plan, after Q1's Z +50.9% wrap.
Krishna framed execution failure ("we faltered") next to external causes. Markets punish that combination.
Software only +5% with transaction processing soft is the growth narrative investors paid for post-Red Hat / HashiCorp / Confluent.

Arguments that one day may overshoot the durable thesis

Operating (non-GAAP) EPS still +5% YoY ($2.93 vs $2.80); operating PTI margin expanded 30 bps to 19.2%.
Red Hat +11%, Distributed Infrastructure +37% with backlog, z17 still ~130% program-to-program, Consulting GenAI signings growing.
Cash flow year-to-date remains substantial ($4.76B FCF after the Financing-receivable bridge).
IBM schedules a full call on July 22 to update full-year expectations — today's letter is deliberately incomplete.
If the "majority" of the shortfall is delayed large deals, some revenue may be timing, visible in Q3 bookings language — if management is right. That is a hypothesis until July 22, not a fact.

The useful stance is not "buy the dip" or "dead company." It is: the tape repriced the duration risk on Z/software closes under AI physical scarcity. Whether that duration risk is one quarter or many is exactly what the July 22 transcript must clarify.

What to listen for on July 22

Scorecard item	Why it matters
How much of the miss was pull-forward delay vs lost demand	Separates timing from franchise damage
Q3 Infrastructure and Transaction Processing outlook	Tests whether late-June was a one-off scramble
Distributed Infrastructure backlog conversion (~$500M)	Shows if the "other" infrastructure win is durable
Red Hat trajectory after +11%	Software quality of earnings
z17 program metrics vs 130% / 85% MIPs language	Franchise health independent of one quarter's close
Full-year FCF / margin guide vs prior	Whether management absorbs or resets the year
Lightwell / quantum commentary size vs Z fundamentals	Do not let narrative distractions crowd out the miss

Run the financials on IBM and the physical-stack names — NVDA, AMD, AVGO, MU, TSM — on EvidInvest (free).

Sources

IBM Exhibit 99.1 to Form 8-K, July 14, 2026 — accession 0000051143-26-000070 (primary for all Q2 preliminary figures + reconciliations)
IBM Form 8-K cover, July 14, 2026
IBM Form 10-Q, quarter ended March 31, 2026 — accession 0000051143-26-000038 (Q1 Infrastructure / Z / Distributed figures)
IBM Form 10-Q, quarter ended September 30, 2025 — accession 0000051143-25-000064 (z17 launch-quarter Z growth)
Consensus comparisons as reported in financial press; not a substitute for the exhibit tables

Research assembled with Aether from IBM's furnished Exhibit 99.1 and filed 10-Q segment history. Preliminary Q2 results may be revised on the July 22 call. Research, not investment advice.