1. Anthropic Takes the Lead (AI)
On May 28, Anthropic announced a $65 billion Series H at a $965 billion post-money valuation, led by Altimeter, Dragoneer, Greenoaks, and Sequoia. Capital Group, Coatue, D1, GIC, ICONIQ, and XN co-led, with Blackstone, Brookfield, Baillie Gifford, D.E. Shaw, Fidelity, and strategic infrastructure partners including Micron, Samsung, and SK hynix also participating. 1 The raise closed the loop that Log 013 caught in rumor form, when Anthropic was reported to be talking about at least $30 billion above $900 billion.
The important number is not only the valuation. Anthropic said run-rate revenue crossed $47 billion in May, up from about $14 billion when it closed Series G in February. The company's disclosed trajectory now runs from $183 billion at Series F in September 2025, to $380 billion at Series G, to $965 billion today. Anthropic also said the round includes roughly $15 billion of previously committed hyperscaler capital, including $5 billion from Amazon. The compute side is just as visible: five-gigawatt commitments with AWS and Google, plus the Colossus lease Aelium covered in Log 012.
Then Anthropic moved onto the public-market track. On June 1, the company confidentially submitted a draft S-1 to the SEC for a proposed IPO. 2 That makes this more than a private-market mark-up. OpenAI's $852 billion valuation was the benchmark Aelium covered in Log 008; Anthropic is now above it and formally preparing to test the public market. 3
The talent signal arrived first. On May 19, Andrej Karpathy joined Anthropic's pre-training team. 4 Karpathy was a founding OpenAI researcher, led AI at Tesla, returned to OpenAI in 2023, then left to build Eureka Labs. His move matters less as a single hire than as a direction of travel: one of the most visible OpenAI alumni joined its closest rival right as the valuation gap flipped.
Why it matters
Anthropic is no longer the safety-focused challenger with unusual enterprise traction. It is now the most valuable private AI company in the world and a near-term IPO candidate. The capital-market signal, talent-market signal, and customer-demand signal all landed together. The fight with OpenAI is no longer only about model quality. It is about who owns the enterprise channel, the compute supply, the developer workflow, and the public-market narrative.
Reality check
A $965 billion private valuation is not a public-market price. The S-1 is confidential, so revenue quality, gross margins, compute obligations, customer concentration, and hyperscaler circularity are still hidden. The $65 billion headline includes capital already committed by infrastructure partners, which makes the "new cash" impression larger than the actual financing delta. And the OpenAI lead is narrow enough that one disclosure can change the ranking again.
2. Saylor Blinks (Crypto)
Strategy disclosed on June 1 that it sold 32 BTC between May 26 and May 31, raising about $2.5 million at an average price around $77,135 per Bitcoin. 5 The stated purpose was narrow: funding preferred-stock distributions. The behavioral meaning was larger. Strategy had spent years making itself the corporate expression of one promise: buy Bitcoin and do not sell.
The sale is tiny against Strategy's holdings. After the transaction, the company still held 843,706 BTC, worth more than $64 billion at the disclosed sale price. But the reversal matters because Log 007 framed Strategy as the buyer of last resort during crypto capitulation: miners were selling BTC to fund AI infrastructure while Strategy expanded its capital-raising plan and pushed toward one million Bitcoin. The largest corporate holder has now shown that even a never-sell vehicle has obligations that can force a sale.
The timing created a second market test. A Polymarket contract worth more than $50 million on whether Strategy sold Bitcoin before a May 31 deadline moved into dispute after the disclosure landed on June 1. 6 The underlying transaction happened before the deadline. The public proof arrived after it. That is exactly the sort of edge case prediction markets are supposed to settle cleanly, and exactly the sort they still struggle with.
The macro backdrop did not help. Digital-asset investment products saw $1.67 billion of weekly outflows, the second-largest drawdown of 2026, with Bitcoin products taking the heaviest hit. 7 Strategy's sale did not cause the risk-off move, but it landed inside it.
Why it matters
Strategy selling is more important as a broken posture than as a balance-sheet event. Saylor's company gave corporate Bitcoin treasury firms their operating myth: permanent accumulation. A 32 BTC sale does not invalidate the thesis, but it does introduce a constraint. The market can now ask what else might force a sale, how preferred-stock obligations interact with BTC volatility, and whether "never sell" was always a slogan rather than a mechanism.
Reality check
Thirty-two Bitcoin is rounding error for Strategy. This was a distribution-management sale, not a directional call on Bitcoin, and the company still holds one of the largest single BTC positions on Earth. ETF and product outflows are meaningful but not existential against cumulative institutional inflows. The Polymarket dispute is also a market-structure story, not a Bitcoin-fundamentals story.
3. IBM Answers Quantum Industrial Policy (Quantum)
IBM followed last week's federal quantum intervention with a corporate commitment of its own. On June 2, the company said it plans to invest more than $10 billion over five years in quantum computing, spanning R&D, capital expenditure, manufacturing, ecosystem work, and possible acquisitions. 8 The stated goal is to deliver IBM Quantum Starling in 2029, a large-scale fault-tolerant system designed to run 20,000 times more operations than today's quantum computers.
This is the sequel to Log 014. On May 21, the Commerce Department announced preliminary CHIPS agreements putting more than $2 billion behind quantum companies and supply-chain partners, including proposed federal equity stakes in IBM and GlobalFoundries. IBM was the anchor. The company's answer is Anderon, a proposed standalone subsidiary to build what it calls America's first purpose-built quantum chip foundry in Albany, New York. IBM said it would contribute $1 billion in cash, plus IP, physical assets, and workforce, alongside a proposed $1 billion CHIPS award. 9
The two announcements belong together. The federal government is treating quantum as strategic infrastructure. IBM is responding by turning its research lead into a manufacturing and roadmap claim. The $10 billion pledge, the Albany foundry, and the 2029 Starling target make the bet legible: own the stack from chip fabrication to fault-tolerant architecture before quantum becomes commercially broad.
Why it matters
Quantum has spent years rich in claims and poor in industrial structure. The past two weeks changed the shape of the sector. Government equity stakes put the state on the cap table; IBM's foundry plan puts manufacturing at the center; the 2029 Starling target gives investors and policymakers a date to judge. This is not yet quantum computing as a market. It is quantum computing becoming an industrial base.
Reality check
The CHIPS terms are preliminary and still require definitive agreements, milestones, and compliance. A $10 billion five-year pledge can be stretched across many buckets, including ordinary R&D, partnerships, and acquisitions. Fault tolerance remains a target, not a delivered machine. IBM has credibility, but the field has a long history of timelines moving right.
4. NVIDIA Picks the Humanoid Body (Robotics)
At GTC Taipei, NVIDIA announced Isaac GR00T N1.5 and its first open humanoid robot reference design. The reference system pairs NVIDIA's humanoid foundation models and robotics software with Unitree's H2 Plus humanoid body, Sharpa robotic hands, and two Jetson Thor T5000 modules using Blackwell architecture for on-robot AI compute. 10 The system is aimed first at researchers at places including Ai2, ETH Zurich, Stanford, and UC San Diego, with availability from Unitree expected in late 2026.
The choice is the story. Aelium has covered humanoid robots repeatedly: China's scale in Log 002, battlefield deployment in Log 005, Figure's work metrics in Log 013. NVIDIA is not trying to win the body race directly. It is trying to make the brain, simulation layer, and compute module the part everyone builds around.
Unitree gives NVIDIA a pragmatic reference chassis. The company has already shipped consumer, research, and industrial robots at a scale Western rivals mostly have not matched. It is also moving toward public markets, with Chinese state media reporting its STAR Market IPO review has passed. 11 That makes NVIDIA's decision politically loaded as well as technically sensible: the reference humanoid stack for Western researchers is being placed on a Chinese-made body.
Why it matters
NVIDIA won AI by owning the compute layer, then wrapping it in software that made the hardware easier to adopt. The humanoid move follows the same pattern. Let Tesla, Figure, Boston Dynamics, Unitree, and others fight over bodies. NVIDIA wants the default brain, simulator, training pipeline, and edge-compute box. If humanoids become a platform category, the reference design matters.
Reality check
A research system is not a factory deployment. Unitree's scale is strongest in research, education, and lower-cost platforms, not proven general industrial labor. The U.S.-China dimension is a regulatory risk for universities, suppliers, and procurement. And NVIDIA's reference design does not settle the hard robotics problems: dexterity, safety, uptime, maintenance, and useful work outside controlled demos.