A 60-day US-Iran memorandum of understanding expired Monday without a replacement, and by Tuesday the breakdown was visible across asset prices. Gold's December futures slipped from a $4,473.40 open to $4,447.20 per troy ounce by mid-morning ET, and silver's September contract eased from $65.90 to $65.12, as negotiations with Iran "appear to have stalled."
The clearer signal came from rates. The 10-year Treasury benchmark reached 4.72%, near a one-year high, after special envoy Jared Kushner indicated Iran was unwilling to meet US demands. The British military reported a vessel struck by "an unknown projectile" in the Strait of Hormuz, and higher oil prices pushed yields further. The forward tension is that a Hormuz supply shock reads as inflationary, which can lift yields even as safe-haven flows into Treasuries raise the opportunity cost of holding non-yielding gold - which is why bullion fell rather than rallied. Silver sits between the same forces, with inflation a headwind and any growth slowdown a potential source of investment demand.
Why this matters: The forward watch-item is the long end of the yield curve. If Hormuz tension persists and feeds through to oil, a 10-year pinned near one-year highs would keep financial conditions tight regardless of central-bank intentions, complicating any easing narrative. Three things are worth tracking: whether Gulf shipping insurance and freight costs climb, whether oil holds its bid, and whether real yields keep rising. Note that gold's reaction function has flipped for now - in a yields-up, inflation-risk regime it can trade as a rate-sensitive asset rather than a pure geopolitical hedge.
Etched raised $700 million at a $21 billion valuation on Tuesday, led by quant firm Jane Street after it tested, bought, and installed the startup's AI hardware. The valuation has roughly doubled in a month, from $5 billion in December and $10.3 billion in July.
Etched sells full systems it calls "frontier inference clusters," competing with the system-level products Nvidia markets as "AI factories." Per co-founder and COO Robert Wachen, the company engineered two purpose-built components: a low-voltage "prefill" chip that adds transistors without the usual heat penalty, and a "cluster-scale memory" interconnect that pools fast, low-latency memory across many chips for the "decode" stage - aimed at higher throughput at lower cost, and now able to run any frontier model. Jane Street said it was "pleased with the early results" and is running a rack in its own datacenter. The forward signal is the customer-investor overlap: a latency-sensitive buyer validating the hardware in production before leading the round.
Why this matters: The move worth watching is whether more demanding enterprise buyers follow Jane Street from pilot to production, which would validate that purpose-built inference silicon can undercut general-purpose GPUs on cost per token. If that holds, the competitive frontier in AI hardware shifts toward inference efficiency, and the assumption of a single-vendor stack weakens. The cautionary counterpoint is pace: an $11 billion valuation step in four weeks reflects momentum-driven private pricing, and subsequent rounds will test whether shipped-system economics justify the mark.
Sources: TechCrunch
Unitree Robotics, described as the world's biggest humanoid-robot maker, saw its shares jump more than 600% at the open of its Shanghai listing on Wednesday before giving back some gains. The stock opened at 1,100 yuan ($163.12) and traded near 900 yuan on the Star Market, marking the first humanoid-robot IPO in mainland China.
Founded in 2016 in Hangzhou and officially named Yushu Technology Co Ltd, Unitree shipped more than 5,500 humanoid robots last year and posted a 278 million yuan net profit in 2025 - a rarity in a cash-burning field. It competes with US developers on features while pricing lower, benefiting from a Hangzhou robotics "golden cluster" that helped Chinese robotics firms more than triple between 2020 and 2024, per state media. The listing arrives amid an intensifying US-China contest over robots and the AI models that control them, with Tesla, BYD, and Amazon all developing humanoids.
Why this matters: The forward question is not the first-day pop, which reflects domestic liquidity as much as conviction, but whether profitable, low-cost manufacturing becomes the decisive variable in humanoids. If Unitree can scale unit volumes at prices Western rivals cannot match, the sector's competitive axis shifts from demonstration quality to cost and supply chain, areas where clustered Chinese manufacturing holds an edge. Worth watching: shipment growth, gross margins as it scales, and export access, since trade and "foreign entity" restrictions could shape which markets it can reach.
Sources: BBC
Reach Capital closed an oversubscribed $265 million Fund V on Tuesday, raised in under six months, to back AI applications across learning, health, and work with $1 million to $10 million checks from pre-seed through Series A. The 11-year-old San Francisco firm plans roughly 50 investments from the fund over the next three years.
The raise is notable against a bifurcated market. PitchBook and the National Venture Capital Association found established firms captured more than 90% of the roughly $62 billion raised across US VC funds through May, squeezing first-time and mid-sized managers. General partner Jomayra Herrera said most existing LPs "doubled down" alongside a few new names, attributing the smooth process to "LP interest in sector-focused boutique funds that focus on conviction-based investments." Limited partners include Capricorn Investment Group, the Los Angeles Fire and Police Pensions, the LEGO Foundation, and College Board. Prior funds were $215 million in 2023 and $165 million in 2021; a June exit saw Superhuman, formerly Grammarly, acquire portfolio company GPTZero.
Why this matters: The forward implication for fund managers is that a defensible, narrow thesis is increasingly the price of entry rather than a differentiator. Reach's quick close, driven largely by re-upping institutional LPs, suggests pensions and foundations will still commit to specialists with a clear conviction story even in a tight market. Worth watching: whether more sub-$300 million specialist funds can close on similar timelines, and whether the 90% concentration at the top persists or eases as distributions back to LPs recover.
Sources: TechCrunch
Amylyx surged, a move typical of clinical-stage biotech reacting to trial or pipeline developments. Such outsized swings reflect binary sentiment around drug-development milestones. The stock warrants attention as a high-beta biotech gauge.
Haemonetics rose meaningfully, putting the blood-management and medical-technology company in focus for healthcare-sector watchers. Moves of this size usually follow earnings or guidance shifts. It offers a read on demand across hospital and plasma-collection end markets.
Targa Resources gained, reflecting positive sentiment in the midstream energy space and natural-gas infrastructure demand. The name tracks volume growth across gathering and processing assets. It provides diversification context outside the tech-heavy watchlist.
Credo Technology dropped sharply, a notable move for a supplier of high-speed connectivity products tied to AI data-center buildouts. Such declines can reflect shifting expectations around AI-hardware demand or guidance. The name is a sensitive gauge of the AI-networking supply chain.
Applied Optoelectronics fell steeply, keeping the optical-networking supplier in focus amid volatile sentiment around data-center and AI-transceiver demand. High-beta moves like this are common for the name. It reflects investor positioning across the optical-components space.
Aehr Test Systems declined significantly, drawing attention to semiconductor test-equipment demand and its exposure to the silicon-carbide and reliability-testing markets. The stock is highly sensitive to order-flow and customer-concentration dynamics. It offers a read on niche chip-equipment sentiment.
MaxLinear dropped sharply, putting the communications-chip designer in focus as investors reassess broadband and connectivity-semiconductor demand. Moves of this magnitude reflect the cyclical nature of its end markets. It serves as a data point on the broader chip-demand cycle.
AXT Inc (AXTI) traded at 82.31, -14.23% on the session.
Geopolitics reasserted itself as the dominant macro driver this week, with the lapse of a 60-day U.S.-Iran memorandum of understanding pushing safe-haven flows into Treasuries rather than the usual precious-metals bid. The 10-year yield climbed to roughly 4.72%, near a one-year high, after special envoy Jared Kushner signaled Iran was unwilling to compromise and the British military reported a vessel struck by an unknown projectile in the Strait of Hormuz. Higher yields and firmer oil are keeping an inflation tail-risk alive even as recent data point to softening consumer spending, a tension that left gold and silver drifting lower and mortgage rates broadly firmer. On the trade front, a short reprieve for Canada tempered an otherwise risk-off tone.
The week's tech tape captured the two poles of the AI cycle: euphoric capital formation and mounting governance friction. China's humanoid-robot champion Unitree delivered a blockbuster Shanghai debut, underscoring how the US-China race now extends from models to the machines they animate, while OpenAI simultaneously tightened both child-safety guardrails and internal security in the wake of a training-environment breach. Underneath the headlines, the harder economic question of who gets paid when generative models train on human work is moving from principle to litigation, with AI music as the test case.
Switzerland's ecosystem this week spanned its full stack - from startups and life sciences to large-cap corporates, financial regulation and energy policy. Fresh clinical milestones across three biotechs and a reverse-merger route to a SIX listing sat alongside a landmark FINMA warning on AI and cyber risk, a major Sandoz biosimilars deal, a Russian state-takeover threat to Nestlé, a roughly CHF 4 billion power-security cost estimate, and a structural upgrade to the Twint payment rails.
Venture capital continues to bifurcate along a barbell. PitchBook and NVCA data cited this week show established firms captured more than 90% of the roughly $62 billion raised across U.S. VC funds through May. That dynamic played out in real time, with AI-inference hardware commanding an eye-watering valuation step-up, a conviction-led boutique closing an oversubscribed fund, and hardware and health founders drawing large checks against a still-selective backdrop.