Jevons Weekly #003: Big Tech burns cash, Serval at $3B, and who funds the buildout

Week of Oct 26 to Nov 2, 2026 · 6-min read · Startups, AI and frontier tech through an investor's lens

Tokens got 26.8% cheaper in four weeks, and volume rose 37.1%. That is Jevons again. The bill for serving that demand now sits with Big Tech. Alphabet and Meta burned cash in the third quarter. To keep building, Meta plans a $95B stock sale and Microsoft sold $25B in bonds.

Sequoia led a $150M Series C at $3B post-money. It also led the $75M Series B in December, at $1B. Fortune counts three rounds in 12 months and $277M raised. Headcount went from about 20 in January to about 200. Serval says ARR doubled every quarter since the Series B, with no dollar figures. The same week, ServiceNow guided Q4 revenue to $4.25B to $4.26B, below the $4.36B consensus. Its stock fell more than 26% on Thursday.

CEO Jake Stauch on the Jevons paradox: "now I want 100 things done instead of just the 10."

Source: Fortune

Itaú agreed to buy 100% of NeoSpace, a Brazilian AI startup founded in 2024, for R$ 2.925B. Part is paid at closing and the rest over five years, subject to conditions. The deal still needs CADE and Central Bank approval. NeoSpace builds foundation models on Itaú's data and has served the bank for nearly two years in credit, CRM and fraud prevention. Itaú already held 15% after investing in January 2025. CEO Bruno Pierobon cofounded Zup, which Itaú bought in 2019.

Source: Itaú, material fact · Reuters via UOL · Bloomberg Línea

Wellington and Sequoia co-led a $1B Series G at $40B post-money. Kalshi reports more than $1B in annualized non-sports revenue and more than $1T in annualized volume. It counts over 30M accounts and says it holds over 75% of prediction-market volume. Next on the list: compute markets, Latin America and Europe. Cofounder and COO Luana Lara is Brazilian. In Brazil, the CMN and the Finance Ministry blocked these markets.

Source: Kalshi · CNBC · Bloomberg Línea

For 2026, Amazon now guides about $220B of capex and Alphabet holds $195B to $205B. Meta guides $135B to $145B. Microsoft trimmed its calendar-year figure to about $170B from $175B, citing third-party GPU delays into early 2027. Alphabet's free cash flow was -$9.07B, against -$2.7B expected. Its stock fell 13% on Oct 29, its worst day on record. Meta's free cash flow was -$2.12B, and Amazon's over 12 months is -$42.4B. Microsoft stayed positive at $18.63B, down 27%. Demand held: AWS grew 39.3%, its fastest in nearly five years, with a $578B backlog. Azure grew 48%.

Source: CNBC, capex · CNBC, cash flow · CNBC, Alphabet · Amazon

Meta plans to raise up to $95B in equity: $35B in underwritten common and preferred stock plus a $60B at-the-market program. Nvidia, AMD and Broadcom showed interest in up to $5B. On Oct 29, Microsoft priced $25B of notes, its first US bond deal since 2017. The ten tranches run from 2028 to 2066, with coupons from 4.85% to 6.35%. Goldman estimates $7.6T of AI capex from 2026 to 2031.

Source: CNBC · Microsoft

Cohere announced an initial close of a Series E targeting up to C$7B at a C$35.7B valuation. Schwarz Group leads. The Government of Canada commits up to C$1.6B in stages, through CDIC. NVIDIA, Inovia, Bell Ventures, Bullhound and Prosperity7 also joined. Cohere reports third-quarter ARR up 165% year over year. It plans C$1.7B for Canadian compute by 2030.

Source: Cohere via PR Newswire · BetaKit

Apollo's Daily Spark published no new VC-share data between Oct 26 and Nov 2. We carry the Oct 22 reading: 80.4% of the top 1% of AI spenders are VC-backed, against 13.1% of median spenders. Change vs #002: none.

Ramp adds context. In the week to Oct 23, token volume rose 37.1% against Sep 25 and spend rose 8.9%. The blended price fell 26.8%, to $0.56 per million tokens. In September, the top 1% of companies spent $9,567.53 per employee on AI, up from $8,568.17 in August. The median company spent $15.70, up from $14.74.

How we read it: by our math, the top 1% grew 11.7% in a month and the median 6.5%. Ramp counts spend, not who funds it. AWS says Fortune 500 spending on Bedrock grew nearly 7x in a year, a company figure.

Source: Apollo, Oct 22 · Ramp AI Index · Amazon

The South Korean chipmaker raised a $620M Series D at about $2.9B, four times its prior $735M valuation. Korea's National Growth Fund put in $280M. Its RNGD inference chip draws 180W and is made by TSMC. Furiosa plans 20,000 chips this year and more than 40,000 in 2027. A third-generation chip, built with Broadcom, will use 2nm and HBM4.

Source: The Next Web

UK-based Salience Labs raised a $125M Series B to build all-optical switches for AI data centers. Bullhound led, with AMD Ventures, Intel Capital, SBI, Hudson River Trading and the British Business Bank. Earlier rounds were $11.5M in 2022 and a $30M Series A in 2025.

Source: Optica OPN

Agility and Churchill Capital Corp XI expect about $465M in gross capital. That is $164M from trust, a $201M PIPE at $10 led by Foxconn and $100M already received. Shareholders vote on Nov 2, with closing expected Nov 4 and trading as AGLT on Nov 5. Agility cites more than $300M in multi-year Digit 5 orders, which are not current revenue. It counts a pipeline of more than 30 customers.

Source: Agility Robotics

In Carta nº 1, I wrote that in 2025 Big Tech capex was judged by the demand it served, and in 2026 by the cash that funds it. The demand is there: Ramp shows tokens 26.8% cheaper and volume up 37.1%, and AWS grew 39.3%. The cash ran short: Alphabet and Meta burned it in Q3.

So the funding moved outside. Meta plans $95B in stock, Microsoft sold $25B in bonds and Anthropic wants up to $100B in its IPO. Money that goes into new paper comes out of something, and so far it has come out of silicon.

For investors in startups and frontier tech, my read is this: the buildout continues, and the price fell first at the supplier end. Founders building chips, optics and inference now raise against cheaper public comps. I still see the suppliers as the latent opportunity, and I'm starting to look at the buyers.

The full argument, with the numbers, is in Carta nº 1 of Cartas J. Piccioni.

João Piccioni, editor

Every item links to its source. Nothing here is investment advice. Spot an error? Reply and we'll fix it.