The week was heavy on H1 dealflow analysis, but the standout for me was OpenAI’s security overhaul after alleging DeepSeek cloned its models via “distillation”. OpenAI tightened access to sensitive tech (biometrics, offline enclaves, stricter vetting). The move underlines AI’s shift into national-security terrain as US agencies warn of intensified foreign IP targeting.
IPOs / Public
Amazon – considering a deeper multi-B investment in Anthropic to expand joint work on model training, data centres and enterprise sales, countering Microsoft and OpenAI.
Nvidia – first company to reach a $4T market cap.
Microsoft – >$500M AI savings in call centres; AI now 35% of new code; sales cycles 9% faster.
Apple – weighing Anthropic or OpenAI to power a new Siri after internal AI stumbles; trade-off between control/branding vs paying for external capability.
Big Dogs
OpenAI – $3B Windsurf (agentic coding startup) deal collapsed; Google DeepMind hiring Windsurf’s CEO, co-founder and key R&D staff.
Groq (LLM inference chips) – in talks to raise $300–500M at a $6B valuation; “anti-Nvidia” LPU thesis focused on low-latency inference.
Revolut – discussing ~$1B led by Greenoaks at a $65B valuation.
SpaceX – reportedly exploring a raise at a $400B valuation, potentially with an employee tender.
Mistral (AI lab) – said to be seeking $1B equity plus hundreds of millions in debt for data-centre build-out.
LangChain (LLM app dev/observability tooling) – raising at ~ $1B valuation led by IVP.
OpenAI – prepping an AI web browser (Chromium-based) with chat + agents; Perplexity (AI search) launched its own AI browser.
Block – Jack Dorsey unveiled Bitchat (open-source, decentralised chat app).
Venture Capital Insights
Evantic Capital (solo VC fund) – Matt Miller closed $400M for AI/infra at Series B; Sequoia rumoured as LP (despite Klarna tension). Prior bets include Graphcore, Grafana, dbt Labs.
PitchBook Q1 AI – horizontal platforms (OpenAI, Anthropic) captured nearly 70% of AI funding ($50B), while vertical apps led in deal volume (60%), underscoring the capital intensity at the platform layer vs scalability of sector-specific apps.
Q2 AI – $40B raised (45% of global VC), with the US at $34.5B, driven by mega-deals like Scale AI and Thinking Machines. Smaller bets drying up, the winners’ pool shrinking fast. “If you’re not building AI or enabling its infrastructure, tread carefully”.
Dragos Novac H1 – slight rebound: ~2,500 deals vs 2,200 in H2’24; only 1,500 deals <$10M (5-year H1 low). 35 mega-deals >$100M (e.g. Helsing $677M, Isomorphic Labs $600M, XY Miners $300M, Neko Health $260M, Amboss $250M) vs 50 last year. Europe’s growth pipeline steady at ~100 candidates/yr (61 deals in $50–100M range) vs 131 in 2024.
YC Euro intake – 49 startups in H1 vs 62 in all of 2024 (notably higher than the most active locals: CDP, Kima and Fuel). This matters because YC remains a powerful channel into US capital, talent and customers – expect more Euro founders to flip topcos/HQs or build US go-to-market early.
Keith Rabois – resurfaced tweet (“Find a large, fragmented industry with low NPS; vertically integrate and simplify”) hit 5M impressions - timeless heuristics!
Venture Geopolitics
OpenAI – stepped up internal “tenting”, offline environments and security leadership (ex-Palantir CISO Dane Stuckey; board member Paul Nakasone) amid espionage fears.
Liangzhu (Hangzhou) – rising as a state-backed AI hub (Zhejiang University talent) with DeepSeek, Mindverse (agentic AI).
China EVs – of 129 domestic EV brands, only ~15 expected to survive by 2030. Price wars, oversupply, and commoditised tech are crushing margins. Globally, this points to cheap Chinese EV exports flooding markets, worsening trade wars, undermining Western carmakers and cementing Beijing’s grip on EV batteries and supply chains.
Strategic sectors
AI
a METR study found experienced OSS developers using early-2025 AI tools (Cursor, Claude 3.5/3.7) took 19% longer to complete tasks. Developers expected +24%, experts even more. In reality, AI added overhead when used in complex projects (time spent prompting, waiting, verifying outputs), often compounded by over-reliance on flawed suggestions.
Thomas Tungz wrote a post on cost management in AI: when you query AI, it gathers relevant info to answer you. Practitioners believe input = ~20x larger than output. His experiments with Gemini revealed 300x on average, up to 4000x. On OpenAI’s pricing page, output tokens for GPT-4.1 are 4x price of input tokens. BUT when input is 300x more voluminous, input costs are still 98% of total bill. As such, cost management is all about input (i.e. core challenge of building with LLMs isn’t prompting, it’s context engineering). For developers, focusing on input optimisation is critical for controlling costs, reducing latency and building a successful AI-powered product.
Ben Thompson wrote about how in past tech waves firms focused on distinct niches, but the AI race is more universal, intensifying competition for the same resources (specifically, AI talent and GPUs). Apple has been cautious, aligning with its philosophy of computers as tools to enhance humans, rather than agents that replace them, while Meta aggressively hires top talent and invests heavily, seeing AI as central to its mission of doing things for users. These contrasting philosophies — tools (Apple/Microsoft) vs agents (Meta/Google) — shape each company’s strategy and explain why some are willing to pay whatever it takes to win, while others are more restrained. Which referenced and reminded me of a great video of Steve Jobs from the 80s: “computers are bicycles of the mind”. Watch it here.
Robotics
Filics (warehouse AMRs) – raised $15.8M from Sandwater, Alven, F-Log, Amazon Industrial Innovation Fund, Bayern Kapital, Capnamic, 10x Founders.
Cybersecurity / digital risk
Gradient Labs (AI fraud/risk/compliance for fintechs) – $13M Series A led by Redpoint; LocalGlobe, Puzzle, Liquid 2, Exceptional also in.
Pimloc (automated video redaction) – $5M from Amadeus, Edge; Zetta, MD One, Symvan – helps meet privacy rules by blurring faces/plates/sensitive objects.
Energy / Climate
Vinod Khosla argued rolling back mature wind/solar subsidies could free capital for riskier bets like fusion, geothermal and low-carbon steel. Once viable, subsidies distort markets by propping up incumbents.
Chloris Geospatial (satellite-based carbon measurement) $8.5M Series A led by Future Energy Ventures, with AXA IM Alts, At One, Cisco Foundation, Counteract, Orbia Ventures.
US Department of Energy warned blackout risk could rise 100-fold by 2030 due to surging AI power demand. Urgent need for +100GW peak supply. Renewables + storage are already proving 24/7 capability (e.g. CrusoeAI demo), but US policy still clings to coal. The bigger challenge is grid flexibility and investment, not the viability of clean power.
Azeem Azhar wrote about the “weird nostalgia for coal”, arguing the US should be building 21st-century grids rather than subsidising 19th-century fuels.
Space
Varda Space (pharma manufacturing in orbit) raised $187M to crystallise HIV drugs in microgravity; planning monthly launches. If scalable, expect pharma licensing deals to follow.