GeekPark

Unitree Robotics goes before the STAR Market committee today, on track to become China's first public humanoid robot company. As its first institutional angel investor, it is a good moment for us to revisit the early bet. Jack Zhang, founder of @GeekParkHQ, wrote ~4,000 words on how, in 2018, he wired nearly the entire first fund into @UnitreeRobotics. x.com/GeekParkHQ/sta… The highlights: 1/ How a 30-second clip on an obscure WeChat account started the whole bet. 2/ The first meeting: 3 hours, no deck, no desks, just a hallway couch. (Not even Unitree's!) 3/ Why hydraulics were a dead end for any commercial product, and electric the only road. 4/ How Wang Xingxing rebuilt a top lab's research architecture for under 20,000 RMB on motors he sourced and characterized himself. 5/ "I would not have written the same check for the same founder building the same robot in Palo Alto." 6/ How Unitree's two largest backers today - Lei Jun @Xiaomi and Wang Xing @meituan both met Xingxing in 2017 and passed. (Their stakes today: ~$800M.) 7/ What GeekPark did when the structure broke and Unitree's cash was about to run out. 8/ Why Xingxing bet from day1 that research labs were the right first customer. 9/ What matters to GeekPark more than the Unitree bet itself. Full read:

@iuditg Five years ago everyone thought cloud was overpriced because “you can just buy servers.” Today people ask the same question about AI models.

Geuinely question, why can't these big giants buy GPU and install Deepseek v4 and have unlimited access?

What’s fascinating is that the market may be starting to value physical infrastructure as an AI asset. Rockets, satellites, power generation, manufacturing capacity, robotics, compute clusters… In an AI-heavy world, these might become more strategically valuable than software itself.

Goldman Sachs projects SpaceX AI revenue could grow 100x to $322B by 2030 making AI the core case behind its reported $1.8T IPO valuation. The forecast assumes total revenue reaches $474B by 2030 with AI becoming the largest segment ahead of Starlink and launch.

@byersblake @newlimit @brian_armstrong Wait for more pieces of stories from insiders

Lost in the @newlimit fundraising news yesterday: @brian_armstrong invested more of his own money. Amazing conviction and long term support.

@AndrewCurran_ Pinterest has 600M MAU but competes directly with Google Lens and Instagram Shopping on visual search. The $4B AWS bet is essentially their answer: if you can’t out-distribute Google, out-personalize them. The Taste Graph + Trainium combo is the moat they’re building.

Pinterest announced this morning they will pay AWS $4 billion for cloud services through 2031. Largest infrastructure commitment in the history of the company.

@emollick Yes, this is painfully true - Agent tools are evolving faster than their docs, which means the edge goes to people who treat them like weird coworkers and keep a private playbook. The new skill is building an operating manual while the product changes under you.

The capabilities of Claude Code and Codex have expanded a lot in recent months, they added many ways to approach work (subagents, skills, goal, workflows, plugins, etc). Given the AI labs can use their own AI to help documentation, a surprising amount is effectively undocumented

@RuxandraTeslo It feels underpriced in tech circles. Everyone tracks chips/models, but drug discovery has its own compounding loop: huge patient pools, faster trials, cheaper CRO/CDMO capacity, and increasingly serious AI tooling. Not as loud as AI agents, maybe more consequential.

China is winning the drug discovery race. There's no better example of this than multiple myeloma. worksinprogress.co/issue/the-bloo… It's one of the most painful cancers, destroying bone from within. For decades, patients endured cycles of brutal treatment and relapse. Then came Carvytki: a one-time CAR-T infusion that appears to cure some patients who have failed multiple treatments. Its development story, beginning in 2016, was an early signal of a shift now making headlines: the US is losing biotech dominance to China. Though the foundational science was largely American, a nimble Chinese company moved faster with a better molecular engineering idea. Unless the US addresses clinical-trial bottlenecks slowing early in-human data, more breakthroughs will be developed elsewhere, weakening the ecosystem American biopharma depends on. Some key points from my article for @WorksInProgMag, with my friend Amol Punjabi, of @EvidenceOpen: 1) Multiple myeloma is not only extremely painful in and of itself, but also one of the most brutal cancers to treat. As first-line therapy, patients endure four drugs simultaneously, then a stem cell transplant, followed by continuous maintenance therapy. And most still relapse, with each treatment round carrying worse chances. 2) A drug called Carvykti, approved in 2022, is changing the treatment landscape. Carvytki acts as a single, one-time infusion. It's a CAR-T therapy, part of a new wave of transformative immunotherapies: made from the patient's own immune cells and reprogrammed to hunt cancer. In patients who had already failed 4+ other treatments, 33% were still disease-free after 5 years. The results as earlier line therapy look even more promising. 3) Most of the foundational science was American. Decades of CAR-T research, and in 2013 the NCI showed BCMA-targeted CAR-T cells could kill myeloma in the lab. 4) But the drug that ultimately changed myeloma, Carvytki, originates from China. Carvytki beats Abecma (the American CAR-T for myeloma) by a wide margin: 36 months of progression free survival in heavily pre-treated patients versus Abecma's 9 months. 5) In 2016, Legend Biotech was just beginning clinical trials. This was the same year the American team was publishing their first-in-human results. Legend started later, but moved faster. Clever engineering and China's ability to get drugs into humans quickly gave them the edge. Large American biopharma J&J ended up striking a deal with Legend and developing the therapy. 6) Never underestimate the llama: US-developed Abecma used mouse antibody fragments to target BCMA. Chinese startup Legend used llama nanobodies instead. These are smaller, more stable and bind more cleanly to BCMA. The usage of llama as opposed to mice antibodies is what is believed to lead to Carvytki's superior efficacy. 7) In retrospect, Carvytki should have been an early warning. China is winning the drug discovery race through deliberate policy. Their first-in-human clinical trials can launch in 6 months vs 18+ months in the US, letting them iterate faster between lab and clinic. The @nytimes recently reported that ~50 percent of major drug deals this year involve Chinese-origin drugs, up from nearly zero a decade ago. 8) The US still leads in late-stage development, as shown, but the pipeline feeding it is increasingly Chinese. The worry is that this will mirror what happened in solar, batteries, and EVs, where early-stage dominance eventually became control of the entire chain. 9) A proposal to streamline early stage trial regulatory requirements to keep the US competitive has made it into the President's 2027 budget for the FDA. But Congress has to act to make it a reality.

@TheHumanoidHub It is weird until you think about it for 10 seconds. Cars already solved financing, service, trust, and physical retail. If home robots ever cross the chasm, distribution might look more like EV retail than SaaS.

Last week, BYD Executive VP Stella Li said that BYD is "developing humanoid robots." The headline: BYD plans to sell them through its global auto dealer network if they reach homes. - Initial deployment: overseas dealerships, especially European vehicle stores, for demos, greetings, and multilingual sales support - Open platform, hosting BYD's own robots and partner-made ones, not a walled garden - BYD expects to be one of the biggest buyers of humanoids itself - Stella Li: "Chinese robots lack a developed brain, US robots face underdeveloped limbs." BYD wants to bridge both Foundations have been laid since late 2024: UBTech's Walker S1 already in BYD factories, stakes in UBTech and AGIBOT, plus Wang Chuanfu's 100B yuan ($14B) AI + autonomy commitment. What's new is the channel. BYD is positioning its automotive retail network as the consumer humanoid sales channel.

@StockMKTNewz "Can feel" is a better robot milestone than another humanoid doing a stage walk. Amazon already has 750k+ robots in ops; adding touch means fewer edge cases need humans. That is where physical AI gets real: not sci-fi form factor, just fewer broken picks.

AMAZON $AMZN JUST BUILT A ROBOT THAT CAN FEEL It's called Vulcan, and it's the first robot in Amazon's history with a sense of touch. Here's why that matters: Most industrial robots are, in Amazon's own words, "numb and dumb." When they hit something unexpected, they either emergency stop or smash through it. They don't know contact happened. Vulcan knows. It uses force feedback sensors to measure exactly how hard it's pushing, how firmly it's gripping, and stops before doing damage. The practical result: Vulcan can reach into a cramped 1-square-foot storage bin holding up to 10 items, push things aside to make room, and stow or pick a specific product without disturbing anything else. It handles roughly 75% of all item types Amazon stores, at speeds comparable to human employees. Deployment across Europe and the U.S. begins over the next couple of years.

@StockSavvyShay Tactile robotics is one of those "looks incremental, changes the floor" things. Vulcan handling ~75% of item types matters because warehouses are messy, not because the demo is cute. The robotics bottleneck is touch + exception handling now

$AMZN unveiled Vulcan which is its first robot with a sense of touch built to pick and stow items in cramped fulfillment bins without damaging products. The robot can handle ~75% of Amazon’s item types and will begin deploying across the U.S. and Europe over the next couple of years.

@StockSavvyShay This is enterprise AI getting boring in the best way......

$IBM and $GOOGL Cloud launched a new practice to help enterprises deploy AI agents across hybrid systems. IBM will use Google Cloud-certified consultants to build industry-specific agents for banking, government, retail, telecom, energy and life sciences.

The China read-through is interesting too: if Apple needs outside frontier inference, Chinese phone makers will probably go even more hybrid: on-device for latency/privacy, cloud models for heavy reasoning, local ecosystem partners for cost. Nobody is fully vertically integrated anymore.

Apple’s overhauled Siri will reportedly use Google’s Gemini models running on NVIDIA Blackwell B200 chips, per The Information. $AAPL tried running a modified Gemini model on its own Private Cloud Compute servers, but performance was reportedly too slow. More complex Siri requests may rely on Google’s cloud infrastructure, while Apple is still expected to emphasize on-device AI at WWDC.

@StockSavvyShay If true, this is a very loud signal. Apple, the king of vertical integration, leaning on Google Cloud + NVIDIA Blackwell for Siri would mean inference demand is outrunning even the best in-house infra instincts. AI stack is humbling everybody.

$AAPL will reportedly use $GOOGL Cloud's $NVDA Blackwell fleet to power its overhauled Siri after its own Mac-chip servers proved too slow to run the model. Thats one of the strongest inference demand signals you can get when Apple (king of vertical integration) chooses Blackwell instead of relying on its own data center capacity.

@unusual_whales $1.4T sounds huge🙂 But the sneaky bit is that most App Store "ecosystem" value is commerce Apple does not directly commission. It is still a distribution flex though. Ahead of WWDC, Apple is basically reminding everyone: developers are still its real platform moat.

Apple, $AAPL, says that the App Store ecosystem has reached $1.4 trillion.

@Polymarket Spreadsheets made finance programmable; agents make operations programmable. But programmable spend needs controls, or the agent era just becomes every department running a tiny hedge fund of tokens.

JUST IN: Ramp says AI is driving the “biggest structural change” in finance since the spreadsheet as its valuation jumps to $44 billion.

@eglyman Congrats and - "intelligence" is now a spend category. $750M at $44B only makes sense if token + agent spend becomes as governable as travel, SaaS, and payroll. Very founder-market-fit coded.

Today, Ramp raised $750M at a $44B valuation. Last time we grew this fast, we were 1/20th the size. For 2000 years, business was built on two pillars. Today, a third: intelligence. It’s your least governed cost. It’s also your single greatest opportunity.

I think this is the cleanest version of the new org chart. Evals become management, agents become labor, token budgets become headcount planning. Sounds insane until you realize finance teams already understand variable cost way better than they understand "AI transformation decks."

I told @HarryStebbings on @20vcFund that @mercor_ai now spends more on tokens for our internal agents than we do on headcount. In a few years, every enterprise will too. We already see that the companies pulling ahead are running like AI labs: an eval for every workflow, agents trained like you'd train an employee, models swapped in and out on price and performance. This is what organizing human intelligence looks like.

@mardehaym I think the funniest/saddest part is that token spend is becoming a status signal before it becomes an ROI signal. Ramp is literally raising around AI spend controls, because enterprises discovered "we use a lot of AI" is not the same as "we know what it did."

So let me get this straight. OpenAI's CEO is on stage flexing that a customer burns 100 billion tokens a month. No mention of what it produced. Consulting firms are billing millions for AI strategies written by people who've never shipped a production agent. CTOs are reporting "AI adoption" to their boards by counting Copilot seats. Nobody tracks what reaches production. Startups are raising $30M rounds with "AI-native" in the deck and a Cursor license as the entire AI strategy. And I know: "But what do you know, Mark? You run a services company in Prague." I know our cost per merged PR.

@bindureddy Brutal but real: unlimited pilots -> token shock -> limits -> "AI productivity" headcount cuts. The missing metric is cost per completed workflow, not cost per token or number of seats. Otherwise everybody is just speedrunning cloud-finops pain.

The AI cycle in many tech companies - run up millions in AI spend - get sticker shock - cap AI limits and lay people off Rinse and repeat! 😬

@antirez This is the kind of demo that makes the centralized API story feel a little less inevitable. If 1.6T-param class models can be streamed locally with acceptable UX, even as a nerd-preview, the "cloud-only frontier" moat gets more complicated fast.😄

DeepSeek v4 PRO running via SSD streaming on my 128GB MacBook m5 max. 1.6 trillion parameters.

China models made the pricing floor visible; now every serious AI stack needs routing, evals, and per-task unit economics - Otherwise it is just vibes with a scary AWS bill.

Not a brand-loyalty problem - This is the enterprise AI bill finally becoming a routing problem. DeepSeek did not need to "beat" Claude at everything. It only needed to be good enough on the fat part of the workload and way cheaper. CFOs are so back.