In partnership with
In Todayโs Issue:
๐ Asian tech stocks tumble as the AI rally cools
๐ฐ AlphaSense raises $350M at a $7.5B valuation
๐ Chinese models surge across OpenRouter token usage
๐งช Meta reportedly tests a premium AI agent called Hatch
โจ And more AI goodnessโฆ
โก The Signal
The selloff hit the narrowest part of the whole AI boom.
Seoul opened the week with a circuit breaker and Tokyo followed it down, and the trigger traces back to one earnings call: Broadcom refused to raise its AI-chip guidance, and the customers it named are Anthropic, Google, Meta, and OpenAI. So the same handful of labs whose compute hunger underwrites these valuations are the ones whose spending now decides whether the KOSPI holds. One of them, Anthropic, spent the same week calling for a brake. Monday was about what investors will finally pay for that story, after a year of paying almost anything.
All the best,
Kim Isenberg
๐ Anthropic Wants an AI Brake Pedal
Anthropic is calling for a global coordination mechanism that could slow or pause advanced AI development if risks escalate. The company argues that frontier labs and governments are still mostly operating with a gas pedal, while the verification machinery for a credible shared slowdown barely exists.
๐ tl;dr: Anthropic is turning recursive-self-improvement anxiety into a governance proposal, but the politics in Washington and Silicon Valley are brutal.
๐ฑ Google Shrinks Gemma 4 for Local AI
Google released new Gemma 4 checkpoints trained with quantization-aware training to cut memory requirements for local deployment. The release includes Q4_0 checkpoints and a mobile-specialized format; Google says the Gemma 4 E2B text-only model can require less than 1GB (!) of memory.
๐ tl;dr: The headline fights are about frontier benchmarks. The quieter contest is making small models cheap enough to run on a phone, and Google just pushed Gemma 4 under 1GB.
๐งฌ AI-Designed Vaccine Reaches Humans
Cambridge scientists say they have run the first human trial of a vaccine whose key component was designed by AI. The Phase 1 trial in 39 volunteers was safe with no serious side effects, though it produced only a modest immune response, so this is a safety milestone and not yet proof the vaccine works.
๐ tl;dr: The useful AI-in-science question is shifting from โcan it suggest molecules?โ to โwhich AI-designed candidates survive clinical reality?โ
๐ AI Rally Stress Test
Use this when an AI stock, startup, or infrastructure bet looks obvious but the valuation already assumes perfection.
Try this: โRead this AI finance story and separate the case into five buckets: 1) real revenue or usage evidence, 2) future growth already priced in, 3) financing or rate sensitivity, 4) customer concentration or supply-chain risk, and 5) what would have to happen for the market to re-rate the asset down even if the technology keeps improving.โ
๐ฌ Watch This
If you build or buy internal tools, watch this one. OpenAI's Sites lets Codex go from writing code to shipping the whole app: describe a dashboard, a review workspace, or a quick internal tool, and Codex builds it, wires up auth, storage, and a database, and deploys it to a live URL you can share. The real point lands fast: the coding agent is turning into app-delivery infrastructure, the layer that quietly decides who owns enterprise software next.
If you build or buy internal tools, watch this one. OpenAI's Sites lets Codex go from writing code to shipping the whole app: describe a dashboard, a review workspace, or a quick internal tool, and Codex builds it, wires up auth, storage, and a database, and deploys it to a live URL you can share. The real point lands fast: the coding agent is turning into app-delivery infrastructure, the layer that quietly decides who owns enterprise software next.
โ Demis Hassabis, CEO, Google DeepMind
Meta is reportedly testing Hatch, a consumer-facing AI agent that could create tools, execute tasks, and plug into Meta-owned services such as Instagram. Indian Express, citing reports, says Meta is also considering a paid Hatch Plus tier with higher usage limits, putting the product closer to premium agent plans from OpenAI and Anthropic than a free chatbot.
The dustup is distribution. Meta may not need the best standalone assistant if it can put an agent in front of billions of Instagram users, which raises the stakes for safety, pricing, and lock-in before the product is even public. One detail worth flagging: reports say Hatch has so far been built on Anthropicโs Claude Opus 4.6 and Sonnet 4.6, with Metaโs own Muse Spark meant to take over at launch.
The AI Rally Finally Met a Sell Button
The Takeaway
๐ Asian markets sold off as investors rushed out of AI-linked shares after one of the strongest runs in global tech.
๐ Koreaโs chip-heavy KOSPI plunged as much as 9% intraday and triggered a circuit breaker, Japanโs Nikkei fell more than 3%, Taiwanโs TAIEX held up better at around 1%, and the Nasdaq had already dropped 4.2% on Friday (Bloomberg, Reuters).
๐ The pressure came from Broadcomโs disappointing outlook, a strong US jobs report that raised rate-hike bets, and renewed Iran-Israel strikes that pushed oil higher.
๐ CNBC framed the Asia leg around investors souring on AI-linked names, with SoftBank down more than 7%.
Monday repriced the AI trade without breaking the underlying demand for it. Asian markets opened the week with investors dumping some of the yearโs most crowded AI-linked winners, especially chip and infrastructure names. The BBC framed it as a plunge after a record AI rally and renewed Iran attacks; Reuters put the immediate trigger on Broadcomโs outlook and the strong US jobs report that pushed traders toward a possible rate hike this year.
The selloff was concentrated where the AI story had become most stretched. Koreaโs KOSPI, loaded with chip exposure and one of the worldโs best-performing markets this year, plunged as much as 9% intraday, triggered a 20-minute circuit breaker, and sat about 14% below last weekโs record high. Japanโs Nikkei fell more than 3% and Nasdaq futures were trying to recover after Fridayโs 4.2% drop, while Taiwanโs TAIEX held up far better, down only about 1% because TSMC alone is more than 40% of the index and barely moved. CNBC adds the single-name proxy: SoftBank fell more than 7% as investors questioned AI-linked valuations.
The macro backdrop made the move sharper. Fresh Iran-Israel strikes lifted oil, US jobs data pushed yields higher, and Reuters noted that traders were again pricing a Fed hike this year. That combination hits long-duration growth stories first: the more a stock depends on AI profits arriving years from now, the more sensitive it becomes to rates, energy shocks, and momentum unwinds.
Monday separated AI demand from AI pricing. Data centers still need chips, memory, networking, power, and software. The question after Monday is how much future growth investors are willing to pay for before the revenues show up.
Why it matters: AI is now big enough to move indexes, currencies, oil-sensitive risk sentiment, and IPO planning in the same week. That makes the AI boom less like a sector theme and more like a market structure problem.
What happens when you throw out the GTM playbook
That investor was wrong. Gamma is now worth $2B, with 50M users and more than half their growth driven by word of mouth.
They're one of 6 AI-native startups in HubSpot for Startups' free Bold Bets Playbook. Replit grew revenue 50x after half the team pushed back on the strategy. Ramp generated 100M+ views from a single stunt. Clay's co-founder wouldn't hang up a sales call until the prospect DMed him in Slack.
Each one took a GTM risk most founders would never greenlight. Each one paid off.
The chart: Chinese models are driving the recent explosion in OpenRouter token usage.
The lesson: The AI market is becoming a price/performance battlefield, not just a frontier-model race.
The caveat: OpenRouter is only one developer-routing platform, so this does not represent the whole AI market, and token volume is not the same as revenue, users, or model quality.
AlphaSense Raises Into the Finance-AI Budget
โก Bottom line AlphaSense raised $350 million at a $7.5 billion valuation, nearly double its prior $4 billion valuation from 2024.
๐ก Why it matters Even as public AI stocks wobble, private capital is still chasing AI tools with measurable finance and enterprise workflows.
๐ What it means The strongest AI-finance pitch is moving from generic copilots to proprietary data, premium content, and decision workflows that banks, asset managers, and large companies already pay for.
AlphaSenseโs new round is a useful counterweight to the market selloff. Reuters reported that the market-intelligence platform raised $350 million at a $7.5 billion valuation, with Vitruvian Partners, Accenture Ventures, and J.P. Morgan Asset Management leading the round. The company says annual recurring revenue passed $600 million in Q1 2026 and total funding is now above $1 billion.
AlphaSense packages AI around research reports, regulatory filings, earnings-call transcripts, expert interviews, news, and internal company content, rather than selling a general chatbot. That matters for finance because the value is in faster triage across the documents that already shape investment, strategy, and diligence work, where producing a fluent answer was never the hard part.
The investor list also says something about where AI budgets are still moving. Accenture wants agentic workflow exposure, J.P. Morgan has both strategic and customer-side reasons to watch market intelligence, and AlphaSense says its customers include Adobe, Amazon, Microsoft, Nvidia, Pfizer, and JPMorgan Chase. Public-market investors are questioning AI multiples this week. Enterprise buyers are still writing checks for systems that compress research labor.
The open question is whether AI search becomes a feature inside every enterprise suite or remains valuable as a specialized platform with trusted content rights. AlphaSense is betting that finance will pay for the specialized version.
Hiring in 8 countries shouldn't require 8 different processes
This guide from Deel breaks down how to build one global hiring system. Youโll learn about assessment frameworks that scale, how to do headcount planning across regions, and even intake processes that work everywhere. As HR pros know, hiring in one country is hard enough. So let this free global hiring guide give you the tools you need to avoid global hiring headaches.
