The Latticework Monday Morning Briefing is our weekly “Guide to the Markets” for long term-oriented investors. It is sent on a separate mailing list (complimentary to members). If you do not wish to receive it, opt out here.
Ideas from our Weekly Inspiration newsletter
We highlight a few stock write-ups from the latest Weekly Inspiration:
Alphabet (GOOGL) now ships two custom AI accelerators a year rather than one, an inference-optimized eighth-generation TPU and a training-optimized sibling, because the workloads have diverged too far to serve with a single design. Tae Kim’s distillation of the Hot Chips presentation is the clearest public account of what that buys. The inference chip carries 384 megabytes of SRAM, double the previous interconnect bandwidth, and performs collective operations inside the IO die rather than reaching across the package, which cuts on-chip latency by 5x and fixes the main deficiency of the prior generation in serving mixture-of-experts models. The training chip runs on a new Virgo fabric that links 134,400 chips in a single non-blocking domain, and customers are already training frontier models on tens of the prior generation’s 9,216-chip super pods. At ~17x trailing earnings the valuation looks undemanding for vertically integrated AI infrastructure, though capital spending compresses the free cash flow yield to 1.3%.
HAL Trust (Amsterdam: HAL) is a family-controlled Dutch holding company whose stated net asset value of EUR 180 per share understates what the assets are worth, by construction rather than by accident. The largest piece is Boskalis, the world’s largest non-Chinese dredger, carried at EUR 5.3 billion, or 6.8x average 2024 and 2025 earnings, against 14x at listed peer DEME; Andrew Brown of East 72 marks it at 6x EV/EBITDA for roughly EUR 8 billion of equity, a EUR 2.7 billion uplift on the carrying value. Vopak sits at 7x proportional EBITDA against 9-10x in comparable transactions. His sum of the parts comes to EUR 228 per share, trimmed ~6% from EUR 242 after a soft half at Boskalis, where EBITDA fell 26% to EUR 553 million on thin salvage work and slow Middle East contracting. Against EUR 161 shares that is a ~30% discount, on a portfolio compounding at 15.5% a year since 2002, and every divestment so far has cleared book value comfortably. He owns the shares.
Universal Music Group (Amsterdam: UMG) trades at EUR 14.24, roughly half the EUR 27.50 at which Vincent Bolloré’s Odet sold its own stake, having bought in at EUR 18 to EUR 19. Jeremy reads that exit price as the controlling shareholder’s own estimate of the asset, and reads Pershing Square’s full sell-down as a removal of distraction rather than a verdict. The case rests on capital allocation now getting the Bolloré treatment: 3.1% of the shares have been retired in 2026 under a second EUR 250 million program, and the remaining Spotify holding, now ~20% above the price used in the interim accounts, is plainly saleable, with half of it capable of retiring another 4.58% of the count at EUR 15. A 4.4% FCF on 6% revenue growth implies a 10.4% total yield on a catalogue whose useful life runs far past its 20-year amortization, with the caveat that three customers account for 42% of sales. He holds no position in UMG itself, owning the wider Bolloré complex instead.
As always, the above theses reflect the linked authors’ views (available here), not Latticework recommendations.
Articles worth your time
The Missing Calculation in the AI Data Center Boom, by Mallika Paulraj, argues that investors modeling data center demand are leaving out how fast the compute required per query is falling. Two specifics carry the piece. Velaura’s Titan Core attacks arithmetic operations, which consume 40% to 70% of XPU power, and claims up to a 50% reduction in consumption. TensorMesh’s KV caching reuses stored context and cuts inference compute costs 80% to 95% on repeated or shared queries. She is careful about where the argument stops: Amdahl’s Law limits how much component-level efficiency turns into proportional infrastructure savings, and the effect is less that demand collapses than that the moat migrates from silicon toward software. Read it against the capex assumptions embedded in current valuations.
The Scaling versus Profitability Trade-off: Venture Capital’s Weakest Link!, by Aswath Damodaran of NYU Stern, takes apart the choice between growing fast and earning money that sits under every valuation of a young, cash-burning company. He identifies six determinants of whether a business can actually scale: market size, market growth, industry structure, capital intensity, customer inertia, and dependence on a key person. Founders trade ownership and control for scale, and staying small preserves both while capping the addressable market, which makes the decision a genuine trade rather than a failure of ambition. His sharpest observation is that market definition is itself strategic: describing Uber as logistics rather than as a car service tripled the market it was valued against, without changing anything the company did.
SaaSpocalyse No, by Harvey Sawikin of Firebird Management, treats February’s software crash as a test of whether the market prices differentiation when it is under stress. It did not. Selling ran sector-wide and indiscriminate, and Via Transportation fell 35% despite economics that are not SaaS economics at all, then rebounded 83%. On the other side, Robinhood margin balances reached $17.5 billion in February, up 98% year over year, and $19.5 billion by May, so the leveraged retail dip-buyers were directionally right and structurally fragile at the same time. He also flags the “Texas hedge,” long AI and short software, which concentrates risk rather than offsetting it in exactly the moment correlations reverse violently.
The Smartest Money, by Amos Lawrence of Verdad, finds that aggregate net equity issuance has predicted forward returns at both the market and the sector level. Issuance peaked at $1.5 trillion in 2000 and again in 2021, and the S&P 500 fell 12% and 18% respectively over the year that followed each peak. Today biotech and pharmaceuticals and industrials sit in the top issuance tercile, while communication services and staples sit in the lowest. He is honest about the noise in it: REIT distribution mandates, funding constraints during crises, and genuine capital-spending opportunities all push issuance around for reasons that have nothing to do with managements timing their own stock.
The deck
The 126-page Monday Morning Briefing is available to members. It spans our weekly scoreboard, idea-generation screens, market valuation and positioning, macro and fixed income, and equity valuation screens across international markets (Canada, the UK, Germany, Australia, Japan, Korea, India and Sweden, alongside a cross-country P/E table and European REIT discounts to net tangible assets). We welcome your feedback.
Feedback on the Briefing
“Most of what I monitor, all in one place. Great value add.” —Brad Lummis
“Loving these Monday briefings!” —Jon Bartel
“Tightly presented and easy to digest. I just spent 20 minutes going through it, and it’s helped to level set me for the week ahead.” —Michael Loftis
“A great piece and thoughtfully assembled.” —Brian Wolf
“I have never seen more valuable content in one place.” —Bill Coleman
“Worth its weight in gold.” —Shree Viswanathan
Featured Events
Latticework 2026, Chicago, Illinois (Nov. 10-11, 2026)
Ideaweek 2027 (FULLY BOOKED), St. Moritz (Feb. 1-4, 2027)
The Zurich Project 2027 (COMING SOON) (Jun. 1-3, 2027)
