Perception-arbitrage M&A

The claim

A new class of transaction has emerged that uses the legal form of M&A and resists its economics. Capital changes hands, equity is issued, a press release names both parties, and there is often a genuine strategic relationship underneath. But the dominant source of value at announcement is not synergy, cash flow, or control. It is reclassification: the market moves the target from one category to another, and reprices it accordingly.

Call this perception-arbitrage M&A. It rents a counterparty's credibility to accelerate a re-rating that disclosure alone would have taken quarters, or years, to achieve. The stake buys recognition of something that may already have existed. It does not fabricate a relationship. It compresses the time the market needs to believe one.

That compression works on two clocks, and confusing them is the first error available. The price moves in a day. The classification takes quarters.

Why classic diligence does not apply

Diligence exists to underwrite economic assumptions: cash flows, liabilities, integration cost, synergy realism. In a perception-arbitrage deal, none of these move materially. The capital at risk is trivial against the re-rating achieved. That triviality is not a design flaw. It is what lets the deal clear in weeks instead of quarters, with no integration risk for a board to underwrite and no synergy case to defend.

Classic M&A buys economics and hopes the market notices. Perception-arbitrage M&A buys a market reaction and treats the underlying economics as a downstream bet, not the point of the transaction.

The mechanism

Large, diversified companies routinely carry real exposure to a hot category inside one business unit, fully disclosed, and the market still prices the parent on its legacy business. The market does not aggregate segment truth into parent perception on its own. Reclassification needs a trigger, and disclosure rarely supplies one: the market was not reading the segment closely enough to reprice it on the numbers alone.

Nokia proves the point twice over, because it ran the honest experiment first. It closed its acquisition of Infinera, the optical specialist, on 28 February 2025, integrated it, and reported the results.³ Analysts called the deal sound. The optical business it created is the same one now driving Nokia's data-centre revenue. The multiple did not move. Eight months to the day later, Nvidia bought 2.9 percent of the company and the market repriced it. A real asset, conventionally acquired and fully disclosed, sat in plain sight for two reporting cycles and changed nothing about the category the market filed Nokia under.

A marquee counterparty's association does the work disclosure cannot. Call it a category stamp: a signal that lets the market relabel the target without repeating the underlying analysis, because it trusts the counterparty's judgment more than it was willing to read the target's own filings.

Why this is not signalling

Classic signalling theory holds that a costly, hard-to-fake action reduces information asymmetry. A company takes on debt, or an insider buys stock, and the market updates because the action would be irrational if the private information were unfavourable.

Perception-arbitrage M&A does not primarily reveal new information. Nvidia's association with Nokia told the market nothing about Nokia's optical business that segment reporting had not already disclosed. What changed was not what the market knew. It was which category the market filed Nokia under.

Nor is this a halo effect in the marketing sense, where association with a trusted brand lifts consumer perception of a product's quality. A halo effect operates on demand. This operates on the multiple a public market assigns to an entire equity, which is a capital markets phenomenon, not a marketing one. The audience is analysts and portfolio managers deciding which comparable set to price the company against. The mechanism is reclassification into a higher-multiple category, not improved sentiment toward a product.

Case: Nokia and Nvidia, October 2025

Nokia's exposure to AI infrastructure was not new. Optical and data-centre switching technology relevant to AI workloads sat inside one business group, disclosed in segment reporting for years. The market still priced Nokia as a legacy telecom equipment vendor.

On 28 October 2025, Nvidia agreed to subscribe for 166,389,351 new Nokia shares at $6.01 each, a $1.0 billion investment for a 2.9 percent stake.¹ The two companies framed the deal as a partnership on AI-RAN, integrating Nvidia's compute platforms into Nokia's radio access network and data-centre portfolio. Nokia shares rose approximately 22 percent that day, touching as much as 26 percent intraday, the company's largest single-day gain in nearly a decade.²

Set against that re-rating, $1 billion is immaterial to both balance sheets. Nvidia did not buy Nokia's economics; distribution into carrier networks Nvidia does not have was worth more to it than the capital risked. Nokia did not sell equity for capital. The Nvidia investment acted as a category stamp: it sold a sliver of the cap table for legitimacy the market would not extend on its own filings.

What the market actually reclassified

The day-one move proves nothing. A 22 percent pop is consistent with a dozen readings, most of them mundane. The durable evidence sits in what the stock now moves with.

Before the Nvidia stake, Nokia's daily returns carried almost no relationship to the semiconductor complex: a correlation of 0.25. In the nine months since, that correlation has doubled to 0.54. Against Coherent, the optical component maker, it went from 0.19 to 0.55. Against Ciena, from 0.26 to 0.46. Each of those shifts is statistically significant, and each survives conversion into euros and the removal of broad market movement.⁴

Ericsson is the control, and it did not follow. Ericsson's correlation to the same semiconductor complex moved from 0.23 to 0.29, a difference indistinguishable from noise. Strip out the broad market and Ericsson's residual relationship to the complex stays at roughly zero while Nokia's climbs from negative to 0.40. Two companies in the same industry, selling to the same carriers, and only one of them changed what it moves with.

The nuance that matters most is what Nokia did not lose. Its correlation to Ericsson went from 0.47 to 0.51: unchanged. Nokia did not swap comparable sets. It added one. The market kept pricing Nokia as a telecom equipment vendor and started pricing it as an AI infrastructure name at the same time, and the second loading arrived on top of the first rather than in place of it.

Categories are not exclusive. A company can be filed under two at once, and carry the volatility of both.

The timing is the part that repays attention. The new loading did not appear on 28 October. For the first five months after the deal, Nokia's correlation to the semiconductor complex sat at 0.27, statistically where it had always been. It reached 0.64 only from March 2026, after Nokia reported first-quarter revenue from AI and cloud customers up 49 percent and an optical order book of roughly €1 billion.⁵

The obvious reading is that the results, not the stamp, did the work. Look at the quarter and that reading collapses. The first quarter is Nokia's seasonal trough: €4.5 billion of revenue against €6.1 billion three months earlier, a pattern that has held every year for a decade. Group revenue grew 2 percent year on year. Reported operating income was €63 million on that €4.5 billion, a margin of 1.4 percent. The AI and cloud business everyone was excited about was under a tenth of sales.⁵

No quarter that size reclassifies a €50 billion company. Optical revenue growing 20 percent in the softest three months of the year would once have read as a decent print for a telecom equipment vendor and nothing more. It counted as confirmation of an AI infrastructure business only because the market had already agreed to file Nokia under that heading. The stamp did not supply the evidence. It supplied the frame the evidence was read through, and then five months passed in which nothing arrived to contradict it.

That is the shape of the mechanism. The stamp proposes a category. The market holds the proposal open. Absent disconfirmation, the loading consolidates. The re-rating was instant and the reclassification took two quarters, and anyone reading the price move as the moment of recategorisation read the wrong instrument.

Beyond Nokia

The pattern recurs, though not always in pure form. Microsoft committed up to $5 billion to Anthropic in November 2025, helping push Anthropic's valuation toward a reported $350 billion.⁶ Amazon has committed up to $33 billion to Anthropic⁷ and up to $50 billion to OpenAI for a stake of about 5 percent, alongside tens of billions of dollars in compute purchase commitments running the other way.⁸ Each investment carries a genuine commercial relationship, chip and cloud capacity changing hands for real money, and each also lets the investor claim a stake in the AI infrastructure category, not only a supplier position. These are hybrid cases. Real economics and reclassification both operate, and separating their contribution is harder than in Nokia's case, where the capital committed was trivial against the re-rating achieved.

The contrast cases matter as much as the confirming ones. Salesforce's 2020 acquisition of Slack for approximately $27.7 billion and Cisco's 2023 acquisition of Splunk for $28 billion were priced, negotiated, and diligenced as integration plays.^{9 10} Both required the acquirer to absorb a large target's product, customers, and cost base. Neither company was reclassified into a new category at announcement; both repriced slowly, on diligence and integration progress. Nokia did not. That distinction, not a shared label, is what makes perception-arbitrage M&A a class of transaction rather than a description of one deal.

What the drawdown revealed

Nokia's stock peaked on 2 June 2026 and has since fallen 46 percent. That looks like the reversal a perception-only stake invites. It is not.

Ericsson fell 29 percent over the same window. Ciena fell 40 percent, Coherent 38, Lumentum 32. Nokia did not fall like a telecom equipment vendor caught in a sector unwind. It fell like an optical name in an AI drawdown, which is precisely what the correlation data says it has become. Year to date, Nokia is up 41 percent against Ericsson's 3 percent.⁴

The category assignment survived a violent drawdown intact. It strengthened through it. Whatever else is fragile here, the reclassification is not.

A perception-only stake has no integration to hold it in place, so the move should be easy to unwind. The tape says otherwise. Once a market has assigned a new comparable set, it prices through the new set in both directions, and the assignment proves stickier than the thing that triggered it. What remains genuinely open is the level, not the label. AI-RAN revenue is still close to zero. The commercial platform launched in July 2026; T-Mobile is a trial partner; the spectral efficiency claims are dated 2028.¹¹ The category is settled. The multiple inside it is not.

Why it does not scale

The mechanism depends on genuine surprise. Nvidia's association was a category stamp because the market had not already priced it in. Ericsson, sitting in the same industry with its own AI narrative and its own carrier relationships, got none of the reclassification. A second telecom operator running the identical playbook gets a smaller re-rating. A tenth gets none. Perception-arbitrage M&A is a scarce move, available to whichever company is first and best-positioned to borrow a category leader's credibility, not a repeatable template.

What follows

Two questions matter, and they are not the same question. Was the transaction economically rational. Was the perceptual reframe correctly read. A board can get the second right while the first remains genuinely open, and boards that conflate them will misjudge both.

But the larger point sits above Nokia entirely. Markets do not misprice businesses so much as they classify them too coarsely for what their economics deserve. Categories are a computational shortcut, cheap to assign and expensive to revisit, and a company can spend years improving its underlying numbers without moving its multiple because the market has had no reason to look again. Nokia bought a real optical business and waited. It sold three percent of itself to Nvidia and was reclassified.

Occasionally the cheapest way to create shareholder value is therefore not to change the business at all. It is to change the category in which the business is understood.

Disclosure

This piece is based entirely on publicly available information: company filings, press releases, published financial reporting, and market price data. The analysis and the views expressed are the author's own and do not represent the views of Nokia Corporation or of any other organisation.

Notes

1. Nokia Corporation, "Inside Information: NVIDIA to make USD 1.0 billion equity investment in Nokia in addition to new strategic partnership" (28 October 2025); Reuters, "Nvidia's $1 billion stake sends Nokia to decade high on AI hopes" (28 October 2025). ↩
2. CNBC, "Nvidia takes $1 billion stake in Nokia, sending the 5G equipment maker's shares up 22%" (28 October 2025); Bloomberg, "Nvidia to Take $1 Billion Nokia Stake, Supply Network AI" (28 October 2025). ↩
3. Nokia Corporation, "Nokia completes acquisition of Infinera to create innovation powerhouse in optical networks" (28 February 2025). ↩
4. Author's analysis of daily log returns, August 2023 to July 2026, using adjusted closing prices. Pre-period 2 August 2023 to 27 October 2025 (n=562); post-period 29 October 2025 to 31 July 2026 (n=189). Semiconductor complex proxied by the VanEck Semiconductor ETF (SMH). Differences in correlation tested by Fisher z-transformation; Nokia's shifts against the semiconductor and optical set significant at p<0.01, Ericsson's not significant. Results replicated on euro- and krona-denominated returns and on market-residual returns with S&P 500 movement removed. ↩
5. Nokia Corporation, "Report for Q1 2026" (29 April 2026); quarterly revenue and operating income on a reported basis from company filings, FY2023 to Q2 2026. ↩
6. CNBC, "Anthropic valued in range of $350 billion following investment deal with Microsoft, Nvidia" (18 November 2025). ↩
7. CNBC, "Amazon to invest up to another $25 billion in Anthropic as part of AI infrastructure deal" (20 April 2026). The new commitment, up to $20 billion of it tied to commercial milestones, sits on top of the roughly $8 billion Amazon had already invested since 2023, taking its total potential commitment to Anthropic to $33 billion. ↩
8. Reuters, "Amazon in talks to invest as much as $50 billion in OpenAI, source says" (29 January 2026); Reuters, "OpenAI clinches $840 billion valuation with mega funding from Amazon, Nvidia, SoftBank" (27 February 2026); CNBC, "How Amazon's massive stake in OpenAI could boost its AI and cloud businesses" (27 February 2026). ↩
9. CNBC, "Salesforce acquires Slack for over $27 billion, marking cloud software vendor's largest deal ever" (1 December 2020). ↩
10. Cisco Systems, "Cisco to Acquire Splunk, to Help Make Organizations More Secure and Resilient in an AI-Powered World" press release (21 September 2023); CNBC, "Cisco makes largest ever acquisition, buying cybersecurity company Splunk for $28 billion in cash" (21 September 2023). ↩
11. Nokia Corporation, "Report for Q2 and Half Year 2026" (23 July 2026). ↩