Akashvani: 2026 | May

A cynical lexicon, reports on tech, a layoff tracker, funny commercials and a revised EBITDA definition

May 31, 2026

Hi -

I hope you’re well [and not yet replaced by an agent]. Six tech reports all circling the same question (is enterprise AI a real platform shift, a bubble, or both at once?), bookended by two books, two tools, and four tangents. Let’s get going.

📚 READING

Let’s start with two books: one a 150-year-old cynical lexicon. The other, a poor attempt to package the ancients as travel writing.

First a blast from the past. Ambrose Bierce, The Unabridged Devil’s Dictionary¹, is a ~~cynical~~ delightful collection of aphorisms from ~150 years ago that are still uncannily relevant. A few from the As to whet the appetite:

· Abject: Innocent of income.

· Abominable: The quality of another’s opinions.

· Academe: An ancient school where morality and philosophy were taught.

· Academy: A modern school where football is taught.

· Acclimated: Secured against endemic diseases through having died of one. 🤣

· Accuracy: A certain uninteresting quality carefully excluded from human statements.

· Adamant: A mineral frequently found beneath a corset. Soluble in solicitate of gold.

· Adam’s apple: A protuberance on the throat of a man, thoughtfully provided by Nature to keep the rope in place.

· Affianced: Fitted with an ankle-ring for the ball-and-chain.

· Affirm: To declare with suspicious gravity when one is not compelled to wholly discredit himself with an oath.

· All: Every single cent, except the one you have kept out for yourself.

· Anoint: To grease a king or other great functionary already sufficiently slippery.

· Apathetic: Six weeks married.

· Apologize: To lay the foundation for a future offence.

· Architect: One who drafts a plan of your house, and plans a draft of your money.

· Armor: The kind of clothing worn by a man whose tailor is a blacksmith.

· Arrested: Caught criming without the money to satisfy the policeman.

· Auctioneer: The man who proclaims with a hammer that he has picked a pocket with his tongue.

And that’s just a sprinkling of As. Incredible. I couldn’t stop laughing. If you liked this, you may want to try The Etymologicon by Mark Forsyth² - one of my favorite books of all time. Would be good to buy these books, before they ban humor. Or books. Or thinking.

Next, How to Travel³, is not really a travel book. Or a how to guide. But really pickings of some travel writings from the ancients (e.g. Herodotus, Tacitus, Pausanias, Egeria, Lucian, Seneca, Homer, Palladius..). They’ve tried to make it too cute -- the original Latin, or ancient Greek on the left-hand side, and then English on the right-hand side (as if half the planet reads Tacitus at breakfast. Reality check: We are in the age where even our 🍊”leaders” can barely string three monosyllabic words together into something coherent. Or is that covfefe?). That said, there were a few interesting things in there that caught my eye:

· The Germans are about the only barbarians content with one woman per man (Tacitus - “Germanica”)

· When a woman has given birth to two children, the husband no longer has intercourse with her (Palladius - “On the life of the Brahmans”)

· [To Alexander] “The philosopher is not a subject but a ruler, since no human beings exerts power over him”, “Be poor, desire nothings, and everything will be yours, and you will lack nothing”. “Besides, to kill men is not manly. It’s the work of a thief. Manliness is to battle the element with our naked body, to destroy the cravings of the belly, and to gain victory over internal battles, not wealth and pleasure.” (Palladius - “On the life of the Brahmans”)

· The map of the world per Herodotus - alternatively the “father of history” (Cicero) or the “father of lies” (Plutarch). The map is killer isn’t it?

Despite that, not a book I would recommend. What this book really gave me a hunger to do is to read Herodotus. [Apparently the edition to get is this one with all the maps and footnotes – at ~1000 pages, some light reading awaits!]

🤖 TECH

Six sources, all triangulating one question: is enterprise AI a real platform shift, a bubble, or both at once? I don’t think anyone knows!

AI Eats the World⁴, Benedict Evans’s spring deck is the clearest framing of the AI question. Big Four capex hits $700B in 2026, more than global telecoms (page 8). Capex/sales runs ~55% at Meta and Microsoft. OpenAI signed deals for 30GW+ at $1.4T, roughly 1GW per week. Evans calls it “Other People’s Balance Sheets, circular revenue and a lot of plate-spinning” (page 15). Demand looks thinner than the spend. ChatGPT has 900M weekly users but only 5% pay, and 80% send fewer than 1,500 prompts a year (page 29). Frontier models cluster tightly on benchmarks, so “models seem to be commodities, no network effects” (page 25). Closing slide: “Presume radical uncertainty” (page 77).

Coatue Public Market Updates⁵, the bull’s rebuttal to Evans. AI is the biggest tech wave ever, bigger than mainframe, PC, internet, mobile, or cloud (page 19). Unlike 1999, the buyers are profitable and self-funding. Hyperscaler P/Es have actually compressed since Jan-25 (Meta 23x to 17x, Microsoft 30x to 22x). 2026E hyperscaler capex stacks to $680B, of which $525B flows as FCF to the “sellers of shortage” (NVDA, Samsung, SK Hynix, Micron, TSMC) (page 17). That basket returned ~107% YTD; the buyers just ~4%. OpenAI + Anthropic combined ARR hit ~$55B in April 2026, past Salesforce.

UBS enterprise AI survey⁶, the cleanest CIO scoreboard on enterprise AI. Only 19% of enterprises run GenAI in production at scale, up from 17% in Oct ‘25 and 6% in Nov ‘23 (page 9). Three years in, still single-digit growth. Only 8% deploy agents at scale. Top adoption hurdles are organizational, not technical: 53% cite unclear ROI, 48% compliance, 45% complexity. 42% now say AI will reduce hiring (up from 31% six months ago). Model layer is consolidating: OpenAI 54%, Anthropic 46%, Gemini 38% (down from 46%). Anthropic ramped from $9B to $30B run-rate in Q1 (page 2). 70% of enterprises still self-build.

UBS SMB IT survey⁷, 450 decision-makers, 50 to 999 employees, US/Canada/UK/Australia. 85% expect IT budgets to rise in CY26, average growth 10% Y/Y. AI is the top initiative and the second-largest driver of spend declines, particularly in Data & Analytics (37%), Payments (31%), and Work Management (30%) (page 11). Incremental-budget funding for AI fell from 56% to 48%; reallocation is the rule now, with SaaS the most-cut category. Only 9% of SMBs say they need more seats; 55% are over-provisioned (page 12).

Sequoia AI Ascent 2026, “This is AGI”⁸, Sequoia’s annual keynote (plus YouTube video) is the bull case as a manifesto. Software is a $350B market (of which only ~$6B is AI); services another $650B (~$400B AI-touched); the prize underneath is a $10T labor pool of which only ~$75B has been captured (page 5) - i.e. they want your job, not your SaaS contract. That gap is what they’re chasing. Best framework: Get MAD (Moats, Affordance, Diffusion) (page 15). The METR chart is the interesting visual: task length an LLM can complete with 80% reliability went from seconds (GPT-2) to ~2 hours (GPT-5.4) (page 35). The four-stage agentic ladder runs from Tab autocomplete (”Little Helpers”) to Agentic dev (”Interns”) to Background agents (”Managing Interns”) to Dark Factories (”Mastery”) (pages 41-46). Pricing-unit shift: salaries become tokens, headcount becomes API calls (pages 49-50). [I couldn’t find the deck, so it’s basically a screenshot of some of the slides that the video shows.]

The InfraRed Report 2026⁹, Redpoint’s take on AI infrastructure. Data-dense companion to the Sequoia manifesto. Thesis: the next 12 months favor infra over apps. Token consumption across AWS, Google, and Microsoft went from ~50T in Q1’24 to ~5,000T in Q1’26, a 100x jump (page 10). OpenAI + Anthropic ARR went from $0.2B end-2023 to ~$69B by April 2026 (page 9). Coding agent revenue: Claude Code >$5B, Cursor $2.7B, Codex >$2B. SWE-bench Verified went from 4% to 87.6% in two years. Data wall hits in 2028, GPU demand is 2.9x supply, reasoning queries use 30-100x the compute of chat (pages 12, 25, 40). Public-market split is brutal: app SaaS -39% LTM vs. infra +38%.

🔧 TOOLS

Two tools this month – a layoff tracker, and a MacApp builder.

LayoffHedge¹⁰, a running ledger of 2026 tech layoffs that tags each cut by stated reason (AI, restructuring, M&A, performance), so you can see the AI share separately. As of late May: 47 tech companies, ~170k jobs cut, averaging 11% of workforce per company. AI-attributed cuts alone are ~100k (~63% of the total). [If anyone believes all these cuts are “because of AI”, I have a bridge near Brooklyn that I’d like to sell…]

Glaze¹¹, a natural-language desktop app builder from the Raycast team. Describe what you want, get a Mac app that runs locally (no servers, no internet). OS-level access to files and hardware, easy publishing to team or public stores, free tier with daily credits then $20/mo. Mac-only for now. [Sorry to ask, but you do use Raycast as an app launcher++ don’t you? If not, stop and install it. Now!]

🎲 TANGENTS

Four tangents this month. Amazing commercials from the past, a nerdy delight, a podcast that reframes your kitchen, and EBITDA alternatives.

Joe Sedelmaier’s commercials¹², the official portfolio of the man who basically invented 80s comedy advertising [back when ads were funny instead of 90-second pharma disclaimers about your liver]. If the name doesn’t ring a bell, the work will: Wendy’s “Where’s the Beef?”, the FedEx fast-talker, Mr. Coffee, Little Caesars, Sprint, Alaska. Deadpan characters, odd faces, absurdist scenarios, perfect comic timing. Three of my favorites: FedEx Fast-Talking Man (still the funniest 30 seconds about overnight delivery), Southern Air Steerage (I think United got their marching orders on how to design their experience from the second cabin…), and Bad Flight (ditto).

PNR de-coded¹³, A mench took his own Air India e-ticket apart character by character. Fun ensues: The six-character PNR (DDTCIV in his example) is not globally unique; it only identifies your booking inside one reservation system. The real primary key is the e-ticket number. IATA standardized PNRs in the 1960s around five mandatory fields: name, itinerary, contact, ticketing status, audit trail. Fares are quoted in NUC (Neutral Unit of Construction), a 1970s synthetic currency that doesn’t exist in the real world. Lovely piece of digital archaeology.

The Columbian Exchange¹⁴, Misha Glenny’s In Our Time on the biological and cultural swap that followed Columbus’s 1492 landfall. Three historians walk through what moved which way and what it did to the planet. Highlights: no chillies in Indian food before this (WTF!), no tomatoes in Italian cooking (Mama mia!), no potatoes anywhere in the Old World. Going the other way: horses, wheat, cattle, and the diseases that wiped out an estimated 90% of indigenous American populations. [Wait, MAGA wasn’t the first to oppose immigration??] The reforestation that followed (so much land left untended) may have contributed to the Little Ice Age.

Maybe we need something (darkly) funny to end with: Wall Street already adjusted EBITDA into meaninglessness. Enter stage left, EBITTDA¹⁵ - the updated version in the age of handwringing about token usage: Earnings Before Interest, Taxes, Tokens, Depreciation, and Amortization! If you prefer the classic one, but with a twist courtesy 🍊 :Earnings Before Iran, Tariffs, and Donald Announcements. [Not to worry – we have ~1000 days to go: Plenty of time to end humanity. Besides, ~300k years is good run for homo sapiens, no?]

Sources: ¹ Ambrose Bierce, The Unabridged Devil’s Dictionary (Univ. of Georgia Press, 2000; orig. 1906). ² Mark Forsyth, The Etymologicon (Icon Books, 2011). ³ Maria Pretzler (ed.), How to Travel: An Ancient Guide for Aimless Wanderers (Princeton Univ. Press, 2025), https://press.princeton.edu/books/hardcover/9780691259062/how-to-travel ⁴ Benedict Evans, “AI Eats the World” (Spring 2026 deck, ben-evans.com). ⁵ Coatue, Public Market Updates (May 2026). ⁶ UBS, Evidence Lab Enterprise AI Survey (March 2026 wave). ⁷ UBS, “Under the Hood of SMB IT Spend Trends” (2nd Annual SMB Survey, May 2026). ⁸ Sequoia Capital, AI Ascent 2026 Keynote, “This is AGI” (75-slide deck + YouTube: youtube.com/watch?v=LRo33rnv6rQ). ⁹ Redpoint Ventures, “The InfraRed Report 2026” (77-slide deck, May 2026). ¹⁰ LayoffHedge, Tech Layoffs 2026 (layoffhedge.com/industry/tech-layoffs-2026). ¹¹ Glaze (glaze.app), Raycast Inc. ¹² Joe Sedelmaier, commercial director portfolio (sedelmaier.com/work2). ¹³ Ajitem S., “Iron Core, Part 2: Six Characters” (ajitem.com/blog/iron-core-part-2-six-characters). ¹⁴ Misha Glenny, “The Columbian Exchange,” In Our Time, BBC Radio 4 (open.spotify.com/episode/6WHUGVYjKVbOWfSXkP6HWv). ¹⁵ Brandon Carl, EBITTDA tweet (tinyurl.com/272mc5dm) and Navin Uttam variant (tinyurl.com/25zvafme).

A few that didn’t quite make the cut but may be worth a skim if you have come to the end of the internet and have nothing else to do: (i) Lazard on AI, a sober buy-side framing of where value accrues across the stack. (ii) Jefferies’ Q2 2026 software earnings materials, the quarterly bull-case scorecard on GenAI revenue ramps. (iii) Anthropic vs. Microsoft, a sell-side compare on the model+platform race. (iv) Sparkline Capital’s “AI Disruption,” a quant lens on which incumbents the market thinks are most exposed. (v) Jefferies’ Monthly Software Valuation Update (May 2026), a market-cap-weighted scoreboard of every public software vertical — useful if you want the up-to-date EV/revenue multiples in one place. Saved in the cutting-floor folder.

Hope this helps.

[Two requests: One, if you find this useful, please let others know. Second, if you find something – a book, a podcast, a report, a tool, a quote, advice, etc. - that others may find useful that you would like me to share, I’m all ears! It’s wicked hard and lonely trying to find this stuff.]

Thanks,

Akash