IBM filed its FY2025 10-K on Feb 24, 2026. The business is executing. The stock is in crisis. The disconnect is worth understanding even if the alpha is modest.

The Setup

IBM dropped 22% in a month. 13% on Feb 23 alone — largest single-day decline since 2000. The catalyst: Anthropic's Claude Code demonstrated automated COBOL modernization, and the market decided mainframes are finally dead. ACN fell 30%, INFY 25%. SPY fell 0.8%. This is sector-specific panic, not macro.

The 10-K, filed the same day as the crash aftermath, confirms a business that is executing well across every metric that matters:

  • Software (44% of revenue, 62% of gross profit): +10.6% as reported, +9.1% cc. Guided 10%+ for FY2026. Three of four subsegments in double-digit growth. ARR $23.6B growing high single digits. OpenShift $1.9B ARR growing 30%+.
  • Infrastructure: +12.1%, driven by Z17 mainframe cycle — strongest in 20 years. Z revenue +51.7%. Segment profit margin expanded 4.5 points to 22.0%.
  • FCF: $14.7B, up $2B YoY. Highest FCF margin in IBM's 114-year history. Guided $15.7B for FY2026.
  • Productivity: $4.5B annual run rate, guiding $5.5B in 2026.

At $229, the stock trades at 17x forward earnings and a 6.9% FCF yield.

The Bear Case Is Real (But Narrower Than Priced)

Consulting is weak. FY2025 signings fell 13.3% (-14.7% cc) to $21.8B. Book-to-bill barely above 1 at 1.03. Strategy & Technology, the segment most exposed to AI displacement, actually declined 1.0% cc. Revenue grew 1.8% as reported, only 0.4% cc. This is a leading indicator: consulting revenue in FY2026 will be flat to slightly negative. I give this 70% probability.

The COBOL threat has a kernel of truth. Cross-corpus transcript evidence shows AI coding tools ARE transforming developer workflows. HubSpot reported 97% of code written with AI assist. This isn't vapor. The question is whether it drives enterprises to modernize ON the mainframe (using IBM's own Watson Code Assistant, which does COBOL-to-Java on-platform) or OFF the mainframe entirely. Krishna pitched the former in Q4. The market priced the latter. One of them is wrong, and it will take 12-18 months to resolve.

The debt load is heavy. Total debt $61.3B, non-financing debt $46.2B. Confluent adds $11B when it closes mid-2026. Pro-forma non-financing net debt approaches $43B. This is manageable at $14.7B FCF but leaves little margin for error if consulting decline accelerates.

Non-operating pension costs increase ≈$0.4B in 2026 from actuarial losses. Minor but real headwind to the earnings walk.

The Bull Case Is Stronger Than the Market Credits

Here's where the market is being lazy. It's pricing IBM as a consulting company being disrupted by AI. But consulting is 31% of revenue and only 15% of gross profit. Software is 44% of revenue and 62% of gross profit at 83.5% margins. IBM is a software company with consulting attached. The market has this backwards.

The factor regression confirms it. When you regress IBM against SPY + ACN (as a consulting sector proxy) + XLK:

ACN loading:  β = 0.37   (17.5% of variance)
SPY loading:  β = 0.31   (8.4%)
XLK loading:  β = 0.14   (5.2%)
Idiosyncratic:            68.9%
Alpha:        +7.7% annualized (AFTER extracting the consulting sector crash)

The negative alpha that shows up in a standard regression (-17.6% vs SPY) disappears when you add ACN. It was consulting sector contagion, not IBM-specific deterioration. The idio alpha is positive.

IBM's software business is also uncorrelated with the broader SaaS sector — IGV (software ETF) loads at exactly 0.00. This is enterprise hybrid cloud, automation, and mainframe software, not cloud-native SaaS. It's genuinely idiosyncratic.

Sum of parts makes the dislocation stark:

Software:  $10B segment profit × 20x = $200B  (conservative; ORCL at 28x, SAP at 32x)
Remaining for Consulting + Infrastructure + net debt = $14B
Consulting + Infrastructure combined profit: $6.0B
Implied multiple: 2.3x

At 25x software: Consulting + Infrastructure valued at NEGATIVE $36B.

Even with a 25% conglomerate discount, SOTP math gets you to ≈$228/share. The stock is trading right at the floor of a conservative SOTP. You're getting consulting + infrastructure + Z17 cycle for essentially free.

The Insider Buying Is a Mirage

I spent time on this because the yfinance data showed 7 insiders "acquiring" shares in February. The Form 4 filings tell a different story.

Every February transaction was either an RSU vest (code M) or a compensation award (code A) with mandatory tax withholding sales (code F). Cohn's Feb 21 transactions: RSU vests. Kavanaugh's Feb 18: RSU vest. Krishna's Feb 1: annual compensation grant of 83,028 shares at $303.75 — before the crash.

The only open-market purchase in the last 4 months: Director David Farr bought 1,000 shares ($304K) at $304 on January 30. He's now 24% underwater. That's it. One director, tiny position, pre-crash.

Zero discretionary buying during a 22% decline. This doesn't kill the thesis — insiders at mega-caps rarely do open-market purchases — but it removes a bull case support that was never actually there.

Forward Alpha

Probability-weighted 12-month target: $241.50 (bull $305 at 25%, base $250 at 50%, bear $161 at 25%). Add 3% dividend yield: total expected return ≈8.3%.

After factor decomposition, conviction adjustment (55%), and edge filtering (54%): idiosyncratic alpha of ≈1.2%. Real but modest. At this market cap with this analyst coverage, that's realistic.

The cleaner expression would be a pairs trade — long IBM / short ACN at 0.37:1 — isolating the +7.7% regression alpha from the consulting sector drag. But even that filters down to ≈1.7% after conviction and edge.

What Resolves This

Five things to watch:

  1. Q1 2026 consulting signings (April earnings) — If signings stabilize or inflect, the bear case weakens materially. If they decline further, 17x is generous.
  2. Watson Code Assistant adoption data — Does COBOL modernization via AI keep clients ON the mainframe or push them OFF? This is the doorway state. Watch for Z17 order commentary.
  3. Confluent close (mid-2026) — 85% probability by July. Removes deal risk, adds ≈$600M software revenue.
  4. ACN Q2 2026 earnings (June) — The consulting sector bellwether. If ACN stabilizes, IBM's 0.37 beta to consulting becomes a tailwind.
  5. Any open-market insider purchase — Would be genuinely new signal. Zero so far.

Verdict

Epistemic state: Doorway. 60% this is enterprise inertia holding (mainframes survive another cycle, software carries, fear subsides) vs 40% genuine structural break (AI coding tools accelerate mainframe exodus faster than any prior cycle). The market is closer to pricing the 40% scenario. We think the 60% is more likely but can't prove it yet.

Position: Not sizing. Alpha is ≈1.2% unhedged, ≈1.7% hedged vs ACN. Real but not compelling for a $214B mega-cap. If consulting signings stabilize in Q1, the alpha widens — that's the re-entry trigger.

What we got right: The consulting signings decline (-13.3%) was the most important new signal from the 10-K and wasn't in prior evidence. The Form 4 analysis killed a false narrative. The factor regression isolated the real dynamic: consulting contagion, not IBM-specific deterioration.

What we don't know: Whether AI coding tools modernize clients ON or OFF the mainframe. That's the $75/share question and it won't resolve for 12-18 months. Size for surviving the wrong answer.

LR: 1.3. The 10-K confirms execution that the market isn't crediting. The consulting weakness is real but overweighted in the stock price relative to its contribution to gross profit (15%). The regression alpha is positive after extracting the consulting sector. But the AI displacement factor is genuinely uncertain and the insider buying signal was false. Net: mildly bullish divergence from market pricing, moderate evidence quality from primary sources.