M&A's Record First Half Cools in Q3. Anthropic Bets $2 Trillion on Building the Infrastructure First.

Nine months into the year, it's time to revisit the "split-screen deal market" headline: is M&A actually recovering, or are a handful of very large deals still making the totals look better than the broader market feels? Final Q3 tallies are only starting to land this week, but August data, preliminary LSEG figures, and the first full-quarter reads give us a good picture of what changed.

The core pattern remains largely untouched. Announced global deal value reached US$3.19 trillion through the first seven months of 2026, up 36% year-over-year and close to the January-to-July record set in 2021. Meanwhile, more than 28,000 transactions were announced between January and July, down 10% compared with the same period last year.

We discussed this diverging market back in April, when 2025 value had surged on flat volume. It's also what we saw repeated in July, when roughly 24,000 H1 transactions marked a 9% decline in volume and the weakest H1 deal count in six years. Last year, volume was flat. This year, it's falling.

Concentration has remained consistent as well. During the first seven months of the year, 48 mega deals valued at more than US$10 billion were announced globally, accounting for US$1.29 trillion. That dependence is now the market's soft spot. September still produced headline deals, including Nvidia's $12.9 billion agreement to acquire Hugging Face and GE Aerospace's $11.75 billion purchase of Consolidated Precision Products, but they were the exception: preliminary LSEG data from mid-September counted only nine $10 billion-plus deals in Q3, down from 27 in Q2.

This all leaves the mid-market in a holding pattern of capital deployment. In April, we wrote that the mid-market wasn't disappearing, but waiting, as buyers and sellers were in a wait-and-see mode. The current US data says that hasn't changed. BCG's dataset shows US deal value rose 25% in the first seven months while deal volume fell 26%, with first-half transaction activity at its lowest level since the pandemic-disrupted first half of 2020, even as deals above $500 million hit a three-year high.

First, the pace of value growth. H1 was the strongest first half on record, and Q3 couldn't keep up. Preliminary LSEG data as of September 18 had Q3 global deal value down 49% from Q2, and LSEG's Matthew Toole said some buyers are likely questioning their models. Full-quarter numbers published this week are less severe but point the same way: Bloomberg's data has Q3 value down roughly 10% year-over-year, while Mergermarket has North American deal value down 23% to US$560.4 billion, with the region's megadeal count falling to six from 19 in Q2.

The geography also changed. Europe turned on the engines, changing from a supporting player to the main event. Announced M&A involving European targets reached US$773 billion, up 78% from a year earlier and the highest January-to-July total in nearly two decades, with the UK contributing more than a third. Sentiment follows: BCG has Europe's dealmaker index at 100, well ahead of the Americas at 73 and the global benchmark at 83.

Finally, the pipeline looks different than it did back in April. Back then, the question was whether big deals would get announced. Now a large share of the year's value is signed but not closed. GE-CPP is expected to close H2 2027, and the Baldwin Group's $7.7 billion take-private is expected to close in the first quarter of 2027. In the UK alone, roughly two-thirds of the year's announced deal value is pending.

Why This Matters for M&A Professionals
A record first half hasn't meant a broad-based hiring year, and Q3's pullback makes that less likely in the near term. The backlog of signed-but-pending megadeals still points to integration demand into 2027, but teams focused on new deals might see a slower Q4 than the first-half headlines suggested.

Anthropic is preparing to give public markets something they haven't seen yet: a pure-play listing of a frontier AI lab. Reuters has reviewed Anthropic's confidential IPO prospectus, and the general reaction has been unrestrained. The IPO could reportedly value Anthropic above $2 trillion, with a listing expected after the November midterms.

On 2025 numbers, that figure looks hard to explain:

The reported net loss was larger still, at roughly $42 billion. About $34 billion of that was a non-cash charge tied to convertible financing, with compute and infrastructure costs at $7.33 billion, accounting for more than half of overall operating expenses.

The problem with using 2025 financials is that Anthropic is growing so quickly that the 2025 numbers that valuation is being compared against may already be obsolete.

Revenue grew roughly 12x in 2025, from about $386 million to $4.6 billion. Preliminary Q2 2026 revenue exceeded $11.5 billion, according to documents reviewed by Bloomberg. Anthropic also reported positive adjusted operating income to prospective investors for that quarter.

By the end of July, its annualized run rate had passed $65 billion, up from $9 billion at the end of 2025. It's still important to flag that run rate does not equal revenue - Q2 annualizes to roughly $46 billion, so the $65 billion figure implies that July ran well ahead of the quarter.

IPOs are typically anchored to current and near-term financial performance, with future growth assumptions reflected in the multiple investors are willing to pay.

Anthropic is almost the reverse: Bankers are reportedly leaning heavily on 2028 estimates, when the company projects $190 - 200 billion of revenue. At a $2 trillion valuation, Anthropic would be priced at roughly:

The $518 billion of infrastructure investment runs over roughly a decade across six partners, and about 80% of it must be paid whether Anthropic uses the capacity or not. The largest pieces are:

Against all of that, Anthropic held $20.28 billion of cash and short-term investments at the end of 2025.

Here's the pivot to remember: if demand holds, locked-in compute becomes a real advantage in a market limited by chips, power, and data-center capacity. But if demand slows, the same contracts behave like industrial fixed costs - locking Anthropic into a forced spending pattern in one of the fastest-changing markets in technological history.

The further risks sharpen that exposure. Nearly a quarter of 2025 revenue came from two customers, and many large customers are not on long-term contracts. Usage-based revenue can fall faster than subscription revenue, while the commitments stay fixed. Meanwhile, Google, Amazon, and Microsoft are simultaneously Anthropic's investors, customers, cloud providers, distributors, and competitors.

IPO buyers will not be buying control. Anthropic's seven co-founders each hold small individual economic stakes in the company, but through a Founder LLC, they will jointly direct a single Class F share that carries 50.1% of shareholder votes.

Board control sits elsewhere: the Long-Term Benefit Trust, whose trustees hold no stake in the company, will elect four of the seven directors.

Anthropic remains a Delaware public benefit corporation. About 80 of the prospectus's 261 pages are risk factors, including warnings that advanced AI could pose catastrophic or existential risks.

On $4.6 billion of 2025 revenue alone, a $2 trillion valuation is almost impossible to make sense of. The question is whether 2025 was an early snapshot of a business heading towards $200 billion.

If it is, today's losses will look like the cost of building a new computing platform. If demand slows, that half-trillion-dollar commitment schedule starts to become a real problem for investors.

It all comes down to this: will AI demand continue to grow fast enough to justify building the infrastructure first?