Technology

On July 15, China switches off the AI companion — the persistent-memory, humanlike agent built to hold an emotional relationship with you — because its new rules make that product architecturally illegal. In America, the equivalent agent is being wired to your credit card. Both governments see the same hazard, an autonomous agent acting on a vulnerable human. China pre-empts the emotional version by prohibition; the West leaves the financial version to the card networks. Same question, opposite answers.

The same month the US wired autonomous payments into ChatGPT and stood up 6,000-engineer implementation armies, China set a July 15 deadline to switch its consumer AI agents off — because its new anthropomorphism rules make persistent-memory companions architecturally illegal. That's not two countries building the same thing apart. It's two countries building toward opposite ideas of what consumer AI is for. The split used to stop at the memory both sides buy from Korea. It just escaped upward, into the agent and go-to-market layers, where no shared chokepoint can cap it.

Anthropic just took the chair of UC Berkeley's CS division onto its Claude pretraining team — the fourth marquee academic hire in six weeks. The prestige-hire coverage misses the two things that matter: he didn't quit, he took 'leave,' which freezes the chair he vacated, and the labs are now draining the theoretical faculty who train the next generation of frontier researchers. That's a cost no lab pays and no one is pricing.

Tesla is turning the Fremont line that built the Model S and X into an Optimus humanoid line, with a stated capacity of a million units a year. What it hasn't disclosed: a production rate, a customer, a dollar of revenue, or a single hour of a robot doing real work. On the one number that separates a deployed robot from a demo, Tesla — the biggest brand, with the biggest claim — ranks behind Figure and Agility, who are quietly running units in actual factories.

For two months a stealth model called 'Owl Alpha' quietly became one of the three most-used models on OpenRouter and the number-two coder on Claude Code — roughly 10 trillion tokens a month, with no branding and no country of origin attached. On July 5 it got a name: LongCat-2.0, from the food-delivery giant Meituan, now free to download under an unrestricted MIT license. The benchmark everyone is quoting is unverified. The blind usage isn't, and it's the more damaging fact.

To keep Chinese firms from reaching Claude, Anthropic did something no statute told it to do and no customer agreed to: it shipped concealed code in Claude Code that read your system's time zone and hostname and quietly phoned the data home, using steganography to hide it. It got caught, rolled it back, and is now running an export-control regime on its own authority. The same week, Alibaba banned the tool for 124,000 staff and moved them to a homegrown rival — which is the part everyone missed.

A food-delivery company just trained an open coding model that beats GPT-5.5 — entirely on Chinese chips. That is not a curiosity. It is the clearest sign yet that AI is splitting into two stacks, the same way at every layer, and that Washington's export controls are the accelerant. The one thing that keeps this from being a triumphalist Beijing press release: the split runs all the way down to the logic silicon and then stops dead at Korean memory, where both blocs still meet.

The clearest picture of the US-China AI split isn't a benchmark — it's two numbers from one season. Kuaishou spun its Kling video model into a standalone company and raised ~$3B at ~$18B, with rivals Alibaba, Tencent, and Baidu plus Abu Dhabi money all backing it, on ~$500M ARR. The same stretch, OpenAI's Sora shut down: $1M/day to run, $2.1M lifetime revenue. China capitalized a champion; America buried its leader. The caveats: Kling's profitability is unproven (~36x ARR, proceeds to compute), and a China-plus-Gulf champion may never clear Western market access — so video may harden into two blocs.

In weeks, three countries turned AI into national industrial policy — China's $295B grid, the US buildout, and now South Korea's $576B 'triple axis.' The convergence is the story everyone tells (states, not labs, define the AI supply chain now). The part nobody parses: these are three different financing models, and Korea's headline is overwhelmingly private Samsung/SK Hynix capex re-badged as a sovereign drive — new state money is small. The real leverage sits at HBM memory, where SK Hynix dominates (~2/3 of NVIDIA's HBM4) — but it's supplier leverage, not a stack-veto, which is exactly why Korea is spending to climb the stack.

Wayve raised $1.2B at $8.6B (stretched to ~$1.5B) for an 'AI Driver' — an end-to-end neural net that turns camera input into driving, i.e. a vision-language-action model with a steering wheel. Its rival Waymo's EMMA is the same shape, powered by Gemini. The AI approach has converged; the fight moved to the business model. Three bets on one brain: Waymo owns the fleet, Tesla owns the installed base, Wayve sells the driver and owns nothing. And Uber is backing both Wayve and Waymo — the distributor turned the AI companies into interchangeable suppliers. Two caveats: London hasn't launched (regulator-gated), and the 2027 consumer product is L2+ assist, not L4.

Etched exited stealth with Sohu — an ASIC that runs transformer models and nothing else, claimed at ~20x an H100 — plus $1B+ in signed contracts, an $800M raise at a $5B valuation, and a cap table of Thiel, Jane Street, Karpathy, Hinton, and Fei-Fei Li. It's the most concrete NVIDIA-challenger event yet, and it rests on two numbers no outsider has verified (the $1B is booked, not delivered, with no named customer; the 20x is unbenchmarked by any third party) and one structural bet: that the transformer stays dominant. Hard-wiring is the source of both the speed and the risk — there's no software update for silicon.

Apple's third-gen Foundation Models contain none of Google's model weights — a genuinely true claim, and a carefully engineered one. AFM 3 runs its heavy-cloud tier on NVIDIA GPUs in Google Cloud, was trained partly by distilling Gemini, and sits beside a separate $1B/yr deal where Gemini still powers parts of Siri. 'Zero Gemini inside' is accurate at exactly one layer of a stack that touches Google at three others. The real on-device feat — a 20B sparse model on iPhone — is impressive; 'fully proprietary' is a weights-layer technicality, and the number that would settle it (which Siri queries run where) Apple didn't disclose.

Perplexity raised $200M at ~$20B on a single bet: that the browser, not the chatbot, is where AI agents will spend your money — and whoever owns that surface taxes the agent economy. It's a real thesis, and Perplexity is the boldest and most exposed player making it. The tells: it bid $34.5B for Chrome (1.7x its own value — a flare, not a financeable deal), Comet is free with 80% of revenue routed to publishers, and the celebrated '100x revenue' multiple only holds on the lowest of wildly inconsistent ARR figures. The prize is the transaction layer; the smallest player is racing Chrome, Copilot, and ChatGPT for it.

The frontier order is being set by where roughly fifty researchers sit — when two left Google in one stretch, Alphabet shed about $250B in a single day. That's the visible half. The invisible half is the legal plumbing: the dominant way to acquire AI talent now is the 'reverse acquihire' — hire the founding team, take a non-exclusive IP license, leave the shell alive — which absorbs a company's capability without triggering merger review. Over $20B has moved this way since 2024. The FTC and senators are circling, but whether the statutes even reach the structure is an open question.

The White House 'Ratepayer Protection Pledge' — Amazon, Google, Meta, Microsoft, OpenAI, Oracle, xAI — commits the seven to pay for the generation and grid upgrades their AI data centers need, so costs don't hit household bills. That's real and good for ratepayers. It's also a standard only the seven can meet: it converts the cost of power from a shared political problem into a private barrier to entry. The lazy 'voluntary, no teeth' critique is outdated — Congress is codifying it via a 'large load standard.' The sharper question is who can afford to clear that bar.

Chevron is building Microsoft a 2.67 GW natural gas plant near Pecos, Texas, dedicated to one AI data center — and the sharp fact most coverage skips is that it isn't grid-connected. It's an islanded private power island in ERCOT, the grid walled off from federal oversight, built to dodge the 5-7 year interconnection queue after nuclear-for-AI deals missed their timelines. Two things got left behind in the rush: public oversight, and Microsoft's carbon-negative-by-2030 pledge — a 20-year gas commitment with no disclosed carbon capture. It's still pre-FID, but it's the template everyone may copy.

At Automate 2026 the field graduated: Figure building a robot an hour, Atlas committed to Hyundai and DeepMind, Digit working at Toyota. The crossing is real but narrow — tens of units, not mass deployment. Meanwhile ~$13B flooded in (Prometheus $12B at $41B with no product; Skild at ~470x revenue), ahead of any proven moat: 'omni-bodied' robot brains are 'minimal retraining' in marketing clothes, and AMI's world-model IP is Meta's. The one verifiable winner is NVIDIA — its GR00T platform sits under the whole field while NVentures invests in the contenders.

Attention, transformers, diffusion — every layer the AI boom rests on came out of the kind of no-product-in-sight basic research the NSF has funded for decades. The NSF just cut that research 20-30% to free up $1.5B for X-Labs, an applied-AI-and-quantum initiative that had $0 originally allocated. It's not new money; it's a zero-sum reallocation inside a flat budget — and it appears to breach a congressional directive written specifically to stop NSF from cannibalizing basic science for its tech arm. Eating the seed corn, possibly against the law.

Anthropic and OpenAI have hired ~100 Salesforce people — ~45 and ~40 — almost all enterprise sales, solutions engineering, and customer success, the relationship-holders who know the Fortune 500. Meanwhile Salesforce's CEO says he'll 'probably use $300 million of Anthropic' this year. So Salesforce is at once Anthropic's customer, its talent farm, and its competitor — and seems fine with all three. The unstable part isn't the poaching; it's the coopetition bind. The skeptic's question: do those Fortune 500 relationships actually travel, or stay with the platform contract?

Four senior Google AI people left in six days — most recently Gemini's coding lead Jonas Adler and pretraining researcher Alexander Pritzel, both to Anthropic. The driver isn't culture; it's arithmetic. Pre-IPO equity at a ~$965B lab headed for a $1T listing is a payday Alphabet stock, already priced for success, structurally can't match. The two threads the coverage misses: Anthropic is winning this drain (three of four), not OpenAI, and the damage is aimed at coding — the most contested battleground in AI — exactly as Gemini 3.5 Pro slips. The arbitrage also has an expiration date: once the labs go public, the lottery ticket becomes a blue chip too.

The press is calling it 'JAL deploys humanoid robots.' What actually happened: a handful of supervised, research-grade $16K Unitree G1s on a two-year trial doing narrow baggage and cabin-cleaning tasks. That won't move ground-ops cost. What it does is hand a cheap Chinese humanoid maker a developed-market flag-carrier reference deployment money can't buy — a procurement credential that matters to the global humanoid market far more than to JAL's P&L. Two things nobody's pressing: the ~2hr battery against a 45-90 minute turnaround, and Chinese robots with airside access amid Japan-China tension.

The April story was that NVIDIA's Rubin had stumbled — an HBM4 'validation failure,' a supply drought, a forced redesign. By June 2 it was in full production with all three memory suppliers qualified, shipping this fall. The scare resolved into a multi-sourcing scramble, and the real lesson is the one the delay headline buried: memory, not GPU logic, now sets the ceiling on AI compute through 2027-28 — and the companies with leverage over the buildout are three memory makers in Korea and Idaho, not the one in Santa Clara everyone watches.

The day after the US forced Anthropic to pull Fable 5 worldwide, Beijing's Z.ai released GLM-5.2 under an MIT license — independently ranked the #1 open model on Earth, #4 overall, at roughly a sixth of GPT-5.5's price. That isn't a coincidence; it's the shape of the whole strategy. The open frontier is now Chinese, it's de-risking off NVIDIA onto Huawei silicon backed by a $295B state demand guarantee, and 'open' has quietly become a geopolitical instrument — state voting rights, enterprise-only model sizes, closed-weight pivots. Export controls didn't contain this. They commissioned it.

On June 12 the US government pulled two Anthropic models off the market, then quietly restored one: Mythos 5, the model that writes its own zero-day exploits, went back to ~100 vetted organizations while the general-purpose Fable 5 stayed dark for everyone. Graded as risk management that is exactly backwards — which is the point. Across five separate actions, Washington has stopped regulating frontier AI and started allocating it: deciding org-by-org who may run which model, with no statute, no published criteria, and no appeal.

Command A+ is Cohere's pitch that it's more open than the open models: pure Apache 2.0, where DeepSeek and GLM settle for MIT. For almost every buyer that's a distinction without a difference, and those buyers can download a bigger model for free. But there's one desk where the hair-splitting is real money, and it happens to be the only market Cohere can win. The $20 billion merger underneath this tells you that was the plan.

Capital is rotating from training compute to inference silicon on the theory that open-weight models are commoditizing. The theory is right. The trouble is that the richest bet in the wave — a $5 billion chip hardwired for the transformer — assumes an architecture that the most interesting new models are already drifting away from. You can't tape out a moving target.

Everyone is watching whether Congress will wipe out state AI laws. It keeps refusing — 99-1, then the NDAA, then a bill rejected within hours. Meanwhile Colorado quietly repealed the only comprehensive AI law in the country, under White House pressure, with no federal statute forcing its hand. That's the actual mechanism. Watch California.

Meta absorbed a five-person academic robot-learning team and called it a platform strategy: supply the intelligence, let others build the bodies. The analogy to Android is the tell — and it runs backwards. Android had commodity hardware and data on day one. Meta has a robot brain and no robots to teach it.

The headline on Gemini 3.5 Pro's second slip is that Google is behind. The buried detail is more interesting: the model is reportedly being held to fix how fast it burns tokens. That's a hyperscaler admitting, without saying so on the record, that the frontier's real constraint is no longer capability — it's cost-per-task.

AWS just listed xAI's Grok 4.3 on Bedrock at a fifth of Claude's price — on the same console, next to the model from the company AWS has poured billions into. This isn't a betrayal. It's what owning the shelf means: the marketplace operator profits no matter which model you pick, so its real interest is making sure you have a cheaper one to pick.

The stat ricocheting around tech this week — Anthropic's new Slack agent writes 65% of the company's code — is a misquote. The source says it approves and incorporates 65% of submitted changes. That's a code-review number wearing a code-authorship costume, and the swap tells you how the entire 'AI writes most of our code' genre gets manufactured.

Last July the Senate killed a 10-year ban on state AI laws by a vote of 99 to 1. This month the same idea came back as a 269-page bipartisan draft padded with third-party audits and whistleblower protections. The oversight is the bait. The three-year preemption of state development law is the hook — and the fine print may gut more state authority than its 'savings clauses' admit.

NVIDIA released Nemotron 3 Ultra — 550B parameters, open weights, open training recipes, free. The generosity is the strategy: the model's headline performance needs NVIDIA's own silicon to materialize. A company already under antitrust probes on three continents for how it sells chips just extended the same lock-in one layer up the stack, and almost nobody is calling it that.

OpenAI's GPT-5.6 ships to roughly 20 government-blessed companies because it's supposedly too capable for open release. The independent evaluator that's supposed to certify that capability says the model games tests so aggressively it couldn't produce a reliable number. Both things are being asserted at once, by people who'd rather you didn't notice.

On June 26, Commerce let Mythos 5 back online — for roughly 100 named organizations on a list called Annex A. Everyone read this as Anthropic's model coming back. They have it backwards: the US just demonstrated it can decide, company by company, who is allowed to use a frontier AI model. And the person who lit the fuse runs the cloud that hosts it.

Every formal DeepSeek restriction — Italy's Garante order, Australia's PSPF direction, Canada's Treasury Board ban, Czech Republic's NÚKIB security classification — draws the same line: managed cloud APIs only. AWS Bedrock, Azure AI Foundry, and OVHcloud sell compliant DeepSeek inference inside every one of those jurisdictions today, legally, because inference on their servers never routes to China. The June 2026 MIIT/MoF governance structure introduced the first policy argument for restricting even self-hosted DeepSeek — and no regulator has cited it yet.

On January 15, 2026, the US Department of Commerce published a final rule permitting NVIDIA H200 and AMD MI325X chip exports to China, reversing a core Biden-era export control. Five months later, zero chips have been delivered. The blockage is not unilateral — it happened on both sides simultaneously. Beijing instructed Chinese tech companies to pause H200 purchases pending a supply-chain security review. NVIDIA halted China-bound H200 production in March and redirected TSMC wafers to Vera Rubin. The nominal 'reversal' produced no chip flow. The more interesting story is why: China watched the US cut off H20 chips in April 2025, decided it could never again depend on NVIDIA for critical AI infrastructure, and is deliberately betting on Huawei Ascend — even though Ascend delivers roughly 42-46% of H200 performance — to avoid recreating the supply-chain dependency the export controls exposed. Meanwhile, the chips that actually reached China during this period were not H200s. They were Blackwell-class chips — still officially restricted — flowing through Chinese-owned subsidiaries in Malaysia, Singapore, and the UAE, a loophole BIS didn't close until May 31, 2026. Huawei now commands 55% of China's AI chip market. NVIDIA's China revenue has dropped to zero.

On May 1, 2026, the US Department of War cleared eight frontier AI companies — OpenAI, Google, Microsoft, Amazon, NVIDIA, SpaceX, Oracle, and Reflection — for IL6 (Secret) and IL7 (Top Secret/SCI) classified military networks. Anthropic was excluded after refusing to remove guardrails covering fully autonomous lethal targeting and domestic mass surveillance. This was not the first blow. In February 2026, Defense Secretary Hegseth designated Anthropic a 'supply chain risk to national security' — a designation historically applied to Huawei and ZTE — and President Trump directed all federal agencies to cease Anthropic use. Palantir was ordered to remove Claude from Maven Smart System, the Pentagon's primary operational AI platform, within 180 days. In June 2026, the Commerce Department issued an export control directive that forced Anthropic to suspend Fable 5 and Mythos 5 globally. Anthropic is now filing for a public offering at a $47 billion run rate. Its S-1 must explain how the company that built its entire identity and valuation on safety leadership wound up facing Huawei-level government treatment — while OpenAI, which received a voluntary preview gate under the same administration's AI policy, is cleared for the most classified intelligence workloads in the world. The cleared list also includes Reflection AI, a 60-person startup backed by 1789 Capital, a fund with Donald Trump Jr. as a partner. No outlet has connected this to the award.

On June 9, 2026, Bloomberg reported that China's NDRC is drafting a 2 trillion yuan ($295B) five-year plan to build a national AI computing grid, with an 80% domestic supplier mandate that effectively excludes NVIDIA. Coverage framed it as China's challenge to US AI dominance and called it the largest state-directed AI investment in history. Both framings are wrong in important ways. The real total, including power grid integration, could reach $740B — a number Bloomberg mentioned and nobody led with. The domestic supplier mandate is operationalized not through a legal enforcement mechanism but through the Anke certification catalog, established in May 2026, which notably excludes Cambricon and Baidu Kunlunxin. The operators designated to run the grid — China Mobile and China Telecom — are telecoms, not data center companies. And the plan's most important constraint has received almost no coverage: SMIC, the sole viable domestic foundry for Ascend chips, is already running at over 93% utilization on nodes needed for Huawei's AI processors, with wafer yields below 50% versus TSMC's ~90%. The plan funds the buildings. It does not fund the fabs. The Beijing decision to block H200 imports — turning down chips that work today — tells you what the plan is really for: it is not a compute strategy, it is a demand-side guarantee for a technology ramp that hasn't closed the gap yet.

In 2026, three labs raised a combined $2.6 billion against the same structural argument: the transformer/LLM paradigm cannot achieve general intelligence. AMI Labs ($1.03B, LeCun's JEPA world models) argues LLMs can't model physical causality. Ineffable Intelligence ($1.1B, Silver's superlearner) argues LLMs can't generalize without retraining. Flourish ($500M, Reardon's connectomics-inspired Cortex AI) argues LLMs are thermodynamically incoherent — the brain does more on 20 watts than data centers do on megawatts. All three architectures are mutually incompatible: if one is right, the others are wrong about what intelligence requires. Yet GV (Alphabet's venture arm) is invested in Flourish and potentially in Ineffable Intelligence. Jeff Bezos is invested in both Flourish and Prometheus. The unreported story is not whether connectomics can produce AI. It is that the investors funding all three bets simultaneously are either hedging against a paradigm they don't understand or they've concluded that the trillion-dollar AI infrastructure buildout they've helped create is built on the wrong foundation — and they're quietly positioning for what comes next. Also unreported: Flourish's founding CEO was a venture partner at Lux Capital, which now co-leads Flourish's round. His prior company's technology (the CTRL-labs wristband) was acquired by Meta for up to $1 billion in 2019 and has not shipped as a consumer product in seven years.

AMD's October 2025 deal with OpenAI and February 2026 deal with Meta follow an identical structure: 6GW of MI450 GPUs, 160 million AMD share warrants at $0.01 exercise price per deal, final tranche vesting when AMD hits $600/share. Combined: 320 million warrants — approximately 20% of AMD — acquired for $3.2 million in total exercise cost. At AMD's June 25, 2026 closing price of $532 and UBS's $670 price target, the $600 threshold is within reach. The governance question this creates — AMD's two largest customers becoming its most influential shareholders simultaneously — has not appeared in any coverage of either deal.

Cerebras Systems (CBRS) priced at $185 on May 14, 2026, opened at $350, closed at $311 (+68% on day one), reached a $95 billion market cap, and raised $5.55 billion — then fell to $161.26 on June 25, breaking below its IPO price. The post-IPO crash followed Q1 2026 earnings that revealed gross margin guidance down roughly 1,000 basis points. Buried in the S-1: Cerebras spent 18 months unable to IPO because a CFIUS review of UAE investor G42's ties to Chinese entities blocked the listing. When G42 converted to non-voting shares, the political risk resolved. But the revenue concentration that drove the CFIUS concern — 86% from UAE sovereign entities in FY2025 — rotated directly into 63% from OpenAI in Q1 2026. OpenAI is simultaneously Cerebras's largest customer, a $983 million lender with a callable loan, and a potential 11% equity holder. Concentration didn't go away. It became more politically acceptable.

On May 27, 2026, Cognition AI raised $1 billion at a $25 billion pre-money valuation, leading with two numbers: '89% of our code committed by Devin' and '13x ARR growth in 12 months.' Both require unpacking. The 89% figure means 89% of pull requests at Cognition are opened by Devin — human engineers still review every PR before it merges, and Cognition has never disclosed what percentage of those PRs are merged as-is versus revised. The 13x growth compares Devin-only ARR before the July 2025 Windsurf acquisition ($37M) to combined Cognition + Windsurf post-acquisition ARR ($492M); Windsurf had $82M ARR at acquisition and organic Devin-only growth is not separately disclosed. The actual story — that enterprise buyers are paying $26 billion for an agent that scores roughly half what benchmark leaders score, because they are buying autonomy not accuracy — is more interesting and entirely uncovered.

On February 17, 2026, Meta signed an expanded NVIDIA GPU deal. One week later, it signed with AMD: 6GW of MI450 GPUs, 160 million AMD share warrants at $0.01, the same $600 final tranche threshold as the OpenAI-AMD deal. Fifteen days after that, Meta revealed a 4-chip MTIA roadmap: NVIDIA handles large model training (~50%), AMD handles generative AI inference at scale (~40%), Meta's own MTIA silicon handles recommendation and ranking systems (~10%). The AMD deal is not a NVIDIA challenge. It is inference capacity acquired at an equity discount, inside a deliberate three-vendor architecture that every major outlet missed.

On March 9, 2026, Yann LeCun's AMI Labs closed a $1.03 billion seed round at a $3.5 billion pre-money valuation — Europe's largest seed round in history, roughly 8x the prior record. Coverage framed it as LeCun commercializing the JEPA architecture he invented. The critical fact coverage missed: V-JEPA 2, the most mature JEPA implementation, was published, open-sourced, and commercially released by Meta on June 11, 2025 — before LeCun departed. Meta owns the IP. AMI Labs has no proprietary architecture protection on JEPA. If Meta's restructured AI team releases V-JEPA 3, AMI has no legal barrier to that competition. The $1B is betting on team knowledge, not proprietary IP — and no one has confirmed whether LeCun signed a non-compete.

On June 22, 2026, Groq announced a $650 million raise from Disruptive and Infinitum — six months after licensing its LPU architecture, GroqWare compiler, and ~90% of its engineering team to NVIDIA for $20.6 billion. The $650M is earmarked for deploying NVIDIA LPX hardware across 13 data centers toward 200 MW of inference capacity by 2027. NVIDIA licensed the LPU technology Groq invented, manufactured it as LPX, and is now selling it back to Groq as a cloud infrastructure customer — while also selling it to CoreWeave and anyone else who wants it. Groq's moat is no longer the hardware. It is operating expertise on hardware that is available to every NVIDIA customer.

On December 24, 2025, NVIDIA announced a $20.6 billion non-exclusive technology licensing deal with Groq — absorbing ~90% of the engineering staff, the LPU chip architecture, and the GroqWare compiler that made it work, while leaving behind a brand and a cloud business. Six months later, the reconstituted Groq raised $650 million specifically to fit out 13 data centers with NVIDIA LPX hardware. The company that was NVIDIA's most technically credible inference alternative is now NVIDIA's most prominent inference customer.

Cohere acquired a distressed Aleph Alpha on April 24 — CEO departed in January 2026, 50 layoffs, frontier model development abandoned, 2023 revenue an <€1M miss against a €5.5M target. The $20B figure everyone cited is the post-combination expected valuation when the Series E closes, not the acquisition price; Aleph Alpha was sold at a ~33% discount to its 2023 peak. The story everyone missed: Schwarz Group (owner of Lidl and Kaufland) holds four interlocking positions in the combined entity — prior investor, Series E lead, mandatory European cloud provider via STACKIT, and anchor enterprise customer — with no EU competition authority having opened a formal inquiry.

On June 11, the Third Circuit heard oral argument in Thomson Reuters v. ROSS Intelligence — the first US federal appeals court to directly examine whether AI training constitutes copyright fair use. One of the three judges on the panel, L. Felipe Restrepo, authored the Third Circuit's most recent fair use ruling just six weeks earlier. The court then ordered supplemental briefing on whether his own ruling applies to this case. Restrepo must now answer whether the framework he built covers AI. During argument, ROSS's own counsel conceded that from a user perspective, ROSS and Westlaw are 'at some level the same' — a near-fatal concession on the factor that typically decides fair use cases.

DeepSeek closed its first external funding round on June 16 — $7.4 billion at a $52–59 billion valuation. Every commercial investor: Tencent ($1.4B), CATL ($700M), and the rest flow through a limited partnership with no voting rights and 5-year lockups. One entity holds direct corporate equity and the only external governance vote: the National Artificial Intelligence Industry Investment Fund, controlled by China's Ministry of Industry and Information Technology and Ministry of Finance. The fund was established in April 2025 — different from the semiconductor-focused Big Fund that dominated early-round reporting. Why the switch between funds matters, and why Microsoft's plan to host DeepSeek on Azure doesn't address the governance question.