NVIDIA just unveiled Vera, a CPU purpose-built for AI agent workloads, and it's already picked up OpenAI, Anthropic, Oracle, and Perplexity as early customers. Bitcoin is trading around $62,000 to $63,000, holding above $60k even as the US-Iran ceasefire collapsed and oil spiked past $75. Spot Bitcoin ETFs snapped a three-day inflow streak with an $84 million outflow on Wednesday. The SEC is drafting Regulation Crypto, its first real safe harbor framework, and quietly dropped its MetaMask case against ConsenSys with no fines and no admissions. And humanoid robots are hitting real factory floors, with Agibot running eight units for 64 hours straight to inspect 17,625 tablets at 99.99% success. Let's dig in.
NVIDIA is no longer just a GPU company. This week they unveiled Vera, a data center CPU designed specifically for AI agent workloads, and the numbers are aggressive. 88 cores built around NVIDIA's new Olympus core, 1.2 terabytes per second of memory bandwidth, and 3.4 terabytes per second of core-to-core bandwidth. That last figure is roughly three times what typical data center CPUs deliver. NVIDIA claims 1.8x sustained per-core performance versus x86 on agent-style workloads.
Here's why this matters. AI agents aren't like traditional server workloads. They're doing tool use, code execution, sandbox creation, result checking, memory management. That work is latency-sensitive and single-thread-heavy. Traditional x86 chips have been optimized around core count and cloud economics, which means per-core speed has stagnated and memory becomes a bottleneck when agents are hammering it.
Perplexity is already running Vera in production. They report 1.5x faster coding workflows and 1.9x faster sandbox startup times versus x86. Partners are citing up to 3x faster large-scale SQL analytics and 6x lower latency in real-time streaming. NVIDIA is targeting $20 billion in Vera CPU sales by year-end, which would be a serious dent in Intel and AMD's data center franchise.
And there's a roadmap. The next chip, Rosa, targets a 2028 launch using TSMC's A16 process with back-side power delivery, 128-plus Rigel cores on Arm v9.2, and IPC gains potentially north of 50%. This is NVIDIA extending its architectural grip from GPUs into CPUs and storage processors, aiming for a unified AI infrastructure stack.
The competitive picture isn't one-sided though. DeepSeek is reportedly a year into building its own inference chip, and Chinese domestic AI hardware is consolidating fast. Huawei already controls roughly half of China's $50 billion AI chip market. NVIDIA stock is testing its 200-day moving average and three customers alone account for 54% of last quarter's revenue. Concentration risk is real. But if Vera lands the way early benchmarks suggest, NVIDIA just widened its moat from silicon into platform.
Humanoid robots stopped being a demo this year. They're on active production lines.
BMW is deploying Figure AI's Figure 03 at its Spartanburg, South Carolina plant for logistics sequencing. This follows Figure 02, which already helped produce over 30,000 X3s in ten months in BMW's body shop. The new model runs Helix 02, a vision-language-action model coordinating arms, torso, and legs to move heavy carts while walking.
But the more striking story is coming out of China. Agibot ran eight humanoid robots continuously for 64 hours on a Longcheer Technology tablet production line in Nanchang, producing and inspecting 17,625 tablets with a 99.99% task success rate. That was on livestream. Agibot has already produced 15,000 humanoid robots by June and is guiding for 40,000 to 50,000 shipments this year. Deployment time to bring a new workstation live has dropped from 4 months to 1 to 2 weeks.
Another Chinese startup, Simplexity Robotics, shipped 100 i7 Pro units in 11 months to Leaderdrive's CNC facility and other AI infrastructure factories. They're claiming an out-of-the-box deployment window of about an hour per unit and just crossed a unicorn valuation. And in India, Agni Robotics started a pilot at a Chennai auto plant targeting a 40% reduction in operational costs versus human labor over six months, with subscription pricing to lower the upfront barrier.
McKinsey pegs annual robotics funding at around $40.7 billion by 2025. China accounts for over 2 million industrial robot installations. Paradigm, the crypto VC firm, just closed a $1.2 billion AI fund partly targeting robotics. What was theoretical in 2024 is being ordered by the pallet in 2026. The gating factor isn't the hardware anymore. It's identifying which specific factory workflows produce a real closed-loop business case.
Bitcoin is doing something interesting right now. It's trading around $62,000 to $63,000, holding above the critical $60,000 level, and it's up about 9% since the end of June. That's happening while the US-Iran ceasefire collapsed, oil surged past $75 on Hormuz blockade fears, and gold slid to roughly $4,060 an ounce.
Read that again. Gold sold off on a geopolitical shock. Bitcoin held. The market is treating this Middle East escalation as a rates event rather than a pure risk-off event. Front-end Treasury yields rose, which hurts non-yielding assets like gold, and Bitcoin is increasingly tracking the front end of the curve instead of oil or gold. If that pattern holds, Bitcoin is behaving less like digital gold and more like a long-duration bet on future Fed easing.
The ETF picture is fragile though. Spot Bitcoin ETFs pulled in $265.7 million over July 6 and 7, led by BlackRock's IBIT with $209 million in a single day. Then Wednesday brought an $84 million net outflow, ending the three-day streak. Weekly net flows are positive around $257 million, but the pattern is choppy, not committed.
Glassnode's on-chain view is more sobering. Long-term holders are now realizing 43% of losses, up from 15% in February. Entity-adjusted long-term holder realized losses spiked to $280 million daily, the highest since December 2022. Many of these are people who bought near the October 2025 peak around $126,000 and are finally capitulating. Bitcoin trades below True Market Mean at $76,600 and short-term holder cost basis at $72,200. Until those levels get reclaimed on sustained ETF inflows, every rally is suspect.
One more signal worth noting. Coindesk ran a piece on pricing houses in Bitcoin exposing the dollar's loss of value. In BTC terms, US home prices keep grinding lower year after year. The monetary framing hasn't changed even as the spot chart wobbles.
The regulatory picture in Washington got clearer this week, and it's coming from an unexpected direction. The SEC officially placed its Regulation Crypto proposal on the 2026 Regulatory Agenda as economically significant, with rulemaking targeted for July.
What's in it. Three exemption pathways for token fundraising. A startup exemption up to $5 million over four years with light disclosure. A larger fundraising exemption up to $75 million per 12 months requiring audited financials. And critically, an investment-contract safe harbor that codifies when a token stops being a security once the network is sufficiently decentralized and managerial control has ended. That third piece is what the industry has been asking for since 2018.
The timing is deliberate. The CLARITY Act missed its July target and now faces an August 7 deadline in the Senate before the recess. Bitcoin's rally has essentially four working weeks to get a legislative catalyst before the clock runs out. If CLARITY stalls, Regulation Crypto becomes the primary US framework by default, and it's an SEC rule, not a statute, which means it's more durable than guidance but still vulnerable to future administrations.
Meanwhile, the SEC quietly closed its enforcement case against ConsenSys over MetaMask Swaps and MetaMask Staking. No fines, no admissions of wrongdoing. The original theory was that non-custodial wallet interfaces could be broker-dealer activity. That theory just got shelved without a court ruling, which means it can come back, but for now wallet developers and DeFi front-ends have breathing room.
A few other data points worth flagging. Swift launched a blockchain ledger with 17 major banks including HSBC, UBS, Wells Fargo, and Citi to pilot tokenized deposits. Sony Bank got conditional US approval to launch a stablecoin trust with $40 million in starting capital. Tokenized stock transfers surged 105% in a month to $8.4 billion. The rails are getting built quickly. Whether they get built under CLARITY or under Regulation Crypto is the fight of the next four weeks.
Watch what NVIDIA does over the next six months. If Vera lands anywhere near its performance claims and Rosa arrives on schedule in 2028, the AI compute stack becomes a NVIDIA monopoly from silicon to software. That is either the greatest investment thesis of the decade or the single biggest concentration risk in modern infrastructure. Probably both.