Evening all

Google just dropped their new quantum computing chip, Willow. It can solve problems in under 5 minutes that would take a supercomputer 10,000,000,000,000,000,000,000,000 years. Sure, it's beating benchmarks, but the real question is: can it compete with the solar-powered Casio Fx-911 calculator? The one that certainly did not improve my calculus grades but did let me write “Hello” during exams.

As I scour wintery Germany for sunlight to revive my own handheld quantum computer, I share today’s musings.

Cheers
Philip

Chart Art

Seed is the New Series A. Right?

Ah, the eternal debate: Is Seed really the new Series A? Is Mo’ Money = Mo’ Problems (Biggie’s wisdom is being put to the test). From Harry Stebbings to Thomas Tunguz the VC scene has pondered since 2021. I like testing the limits of what excel can handle have conglomerated all funding data since 1998 to check this out. Reports are clear: Seed rounds are bigger than ever. I asked two questions:

1. Really?

2. If so, so what?

Here’s a snack of out of the data buffet I prepared:

• Seed valuations up ~4.0x since 2014. Median multiples are now a frothy 16.0x.

• Capital efficiency is sliding. My proxy for ROIC depicts Seed ROICs dropping from a peak of 2.8x in 2018 to 1.9x today.

• Revenue isn’t keeping pace. Startups are raising more but delivering less on a per-dollar basis. Bang vs. Buck — a fight for the ages.

I’m exploring bloated Seed rounds, capital inefficiency, valuation bottlenecks and most importantly why spending $10k on couches may not be a wise use of proceeds. As an appetizer: My key charts below.

Curious for more? Full article here and see why Biggie was right.

Leverage. LMEs. Lies.

Partners Group has shared the following chart

Big claims and big doubt on my end. I do not believe operational improvement makes up 70% or even 50% of the returns. I appreciate other investors, like Apollo, who have acknowledged that operational improvement isn’t all it’s cracked up to be. Especially when we check out the latest BDC PIK Update something becomes crystal clear. Leverage at all costs. Cant pay? No worries. Get that leverage. Boost that yield. PI(C)K perpetual leverage. Q3 Average PIK income at 15.2%.

Why This Discussion Matters (Even If You’re Not Into Levered Loans)

Companies are staying private for longer, which NBIM (Norway’s sovereign wealth fund) says is a growing problem for public market investors of its size. After all, it already owns about 1.5% of every major listed company in the world, and the number of public companies is shrinking.

If companies want to stay private, investment strategies need to adapt. For index fund-style investors, this means missing out on a growing chunk of global economic activity. The public market pie is shrinking, and understanding how PE sponsors generate returns is more relevant than ever.

Breaking Down PE Value Creation

Cutting through the noise

• Between 2012 and 2022, nearly all PE value creation came from revenue growth and multiple expansion. On a side note: Buyout entry multiples fell from 11.9x EBITDA to 11.0x through the first nine months of 2023, outpacing declines in public market multiples.

• Margin expansion—the hallmark of operational improvements—“barely registers.”

More? McKinsey notes that from 2010 to 2021, two-thirds of PE returns stemmed from multiple expansion and leverage. The heavy lifting wasn’t coming from operational wizardry but rather financial engineering.

Multiple arbitrage makes up a 47% of returns. Even if we translate revenue growth as operational improvement at 100%. Without margin expansion, 53% + 0% does not equal 70%. Not in my world. But perhaps my maths is off?

Dividend Recaps

As exit markets have dried and LPs have grown impatient, dividend recaps through August 2024 hit $43bn YTD. A 6x jump YoY. Despite these ATHs: MainFT notes, PE sponsors want more. Now they are pushing for looser loan terms to enable even larger payouts.

Quick contextualisation:

• What: Draw on additional leverage to pay a special dividend

• The appeal: Dividend recaps pull cash flows forward, boosting interim IRRs and allowing GPs to raise new funds faster.

• The risk:

  • Unsustainable debt burdens on the OpCo (Prime example of this below)

  • Fund-level returns suffer: Studies show recaps reduce cash-on-cash multiples and public market equivalent (PME) returns—effectively trading long-term value for short-term optics. This has been shown as a result of behavioural fund management plays: A train of thought:

    • The GP recaps, boosting the IRR

    • On the back of a boosted IRR the GP goes into fundraising mode

    • The prior vintage lacks attention and deals subsequent to the dividend recaps, suffer.

    • Data has clearly shown: Funds with heavy recap activity show lower returns on subsequent LBOs and fewer new deals overall compared to peers

Case Study

Who better than to illustrate the real life effects of excessive leverage to boost early repayment than the hero of LMEs and B2N Junk Connoisseur: Clearlake

• Clearlake bought out Wheels Pros in 2018 for $130m

• Over time, they heavily utilised dividend recaps to return equity

• In 2022, Clearlake sold the business for $2.4bn, generating $1bn in profit (~7x MOIC).

• In 2024, Wheels Pros filed for Chapter 11 bankruptcy.

The result? GPs secured massive profits, but the long-term outcome for later investors was devastation.

Who’s got Capa?

Three little words. The stuff of trauma for any former or current investment banking junior. At the risk of giving me flashbacks I posed that very question. But this time aimed at AI.

The answer to who has got the most computing capacity amongst the big dogs of tech? Google. With a confidence interval of 50%, their bottom range in H100-equivalent GPUs still blows everyone else out of the water. Its like stepping up from a Some context:

The NVIDIA H100 Tensor Core GPU is a tech marvel designed explicitly for training and deploying AI models at hyperscale:

• Each H100 GPU delivers up to 60 TFLOPS of double-precision compute power and 1,000 TFLOPS for AI workloads

• Optimized for transformer architectures the H100 accelerates training by up to 9x and inference by 30x compared to older GPUs like the A100

• 3 terabytes per second of memory bandwidth

• Advanced sparsity and FP8 precision modes ensure maximum compute output while reducing energy costs

That’s a lot of jargon and FLOPS. In short the message is this: The H100 is a GPU that is up to 6x more powerful on an average perspective across cost saving, energy efficiency, training time than its predecessors.

Hindsight is 20/20 AD

Just a chart that puts things into a bit of perspective. 2024 has been a year of comebacks. Old market wisdoms. Oasis. Enron. And China. China’s economic resurgence isn’t a flash in the pan — its a comeback. At least if we take a long view. A economic hangover that lasted a few centuries seems to now draw to a close: A few centuries ago China was responsible for roughly 30% of global GDP. The century of humiliation post-1800s disrupted this dominance — some key events during that century:

• The British led Opium Wars I & II. 1839-42; 1856-60

• The religiously motivated destabiliser of the Qing Dynasty: The Taiping Rebellion; 1850-64

• Sino-Japanese War that led to the independence of Korea and the loss of Taiwan; 1894-95

• Boxer Rebellion that caused the massively punitive Boxer Protocol enforced on China; 1899-1901