←
orbitant5,974 posts
orbitant avatar

orbitant

@orbitant

🇸🇬 | observer + crafter... music nonstop. NFA cc @marginx_io | fdr @SPARQIO

◷ Joined January 2008▧ 1,716 Photos and videos
orbitant avatar
orbitant@orbitant·3h

Na na na nananana, nannana …

We’ve priced our IPO at $83 per share, with trading expected to begin on Nasdaq tomorrow. anthropic.com/news/anthropic… Legal disclaimer: A registration statement relating to the Class A common stock was declared effective by the Securities and Exchange Commission on November 18, 2026. The offering is being made only by means of a prospectus. Copies of the final prospectus may be obtained, when available, from: Goldman Sachs & Co. LLC, Attention: Prospectus Department, 200 West Street, New York, NY 10282, by telephone at (866) 471-2526, or by email at prospectus-ny@ny.email.gs.com; or Morgan Stanley & Co. LLC, Attention: Prospectus Department, 180 Varick Street, 2nd Floor, New York, NY 10014, or by email at prospectus@morganstanley.com

orbitant avatar
orbitant@orbitant·6h

The suits are coming for AI through Anthropic and OpenAI, just as they came for bitcoin through stablecoins and ETFs. We’ve seen this before. Railroads. Radio. The internet. Open-source software. Crypto. Now AI. The technology starts with innovators. Then comes institutional money, consolidation and control over infrastructure and distribution. They don’t need to own the technology. They just need to own the means of accessing it and collecting the fees. Filter out the noise. Who’s financing AI? Who owns the infrastructure? Who controls the models? Who collects the recurring revenue? Are you paying attention?

They removed Jensen. Now this - the forces in the US are determined to force centralised AI down in everyone’s throat. Build US OSS, or use what’s available, resist. ✊ WAKE UP, PEOPLE!

orbitant avatar
orbitant@orbitant·9h

Singapore launched three national artificial intelligence initiatives on November 18 at the Singapore FinTech Festival, targeting financial crime, small business financing, and autonomous agent transactions. Five banks are already testing a shared data pool to detect illicit fund movements that cross multiple institutions within minutes. Isolated banking silos fail to catch modern scams because stolen capital bounces across different ledgers before anyone blinks. Pooling data allows machine learning models to track patterns invisible to single lenders, shifting compliance from reactive reporting to collective defense. Regulatory oversight is also tightening around hardware imports, with authorities pledging strict enforcement against firms attempting to circumvent chip export rules to neighboring regions. Infrastructure scaling requires both computational access and ironclad compliance to maintain cross-border trust.

Post attachment
orbitant avatar
orbitant@orbitant·10h

Subscribers to paid tiers on X secure a 6.00% annual percentage yield on cash balances up to $150,000 through the newly outlined financial services app. Cross River Bank holds the underlying deposit accounts with Federal Deposit Insurance Corporation protection up to the standard $250,000 threshold. The product structure relies on tier gating to distribute yield and fee waivers. Users on free plans receive a 4.00% annual percentage yield, which scales to 6.00% upon meeting a qualifying direct deposit requirement of $1,000 or more within a trailing 34 day period from sources like creator payouts or automated clearing house transfers. Cross River Bank operates as the backend chartered institution because technology platforms lack direct banking licenses. Software companies handle user acquisition and the front end interface while regulated partner banks provide the balance sheet capacity and compliance infrastructure.

Post attachment
orbitant avatar
orbitant@orbitant·10h

Jensen Huang's real MOAT is becoming increasingly obvious. TSMC $TSM is reportedly getting stricter with its ASIC customers before allocating production capacity. - Checking balance sheets, payment history and creditworthiness. - Monitoring actual data centre construction progress. - Setting credit limits for individual customers. - Requiring cash payments for orders exceeding credit limits and upfront payments to secure production capacity. With all the concerns over circular financing and mounting AI infrastructure debt, it's no longer just about how many chips you want, but whether you can afford them. While the reported measures are targeted at ASIC customers, let's not forget that $NVDA relies heavily on TSMC to manufacture its chips. This is where Jensen's decades of trust and relationships across the global semiconductor and hardware supply chain matter. You can inherit his job, but you can't inherit 33 years of relationships. Nvidia's next CEO had better start working on that Guānxì (关系/關係) in Taiwan, South Korea, Japan, China, Singapore, Malaysia and the rest of Nvidia's global ecosystem. Source: news.mydrivers.com/1/1166/11…

Post attachment

ICYMI... This is Jensen Huang's real MOAT - his Fujian Gang. Not just $NVDA CUDA, GPUs or AI models, but decades of relationships with Taiwan's semiconductor and hardware supply chain. Chips can be designed. Software can be replicated. But trust, execution and relationships built over decades are much harder to replace.

Post attachment
orbitant avatar
orbitant@orbitant·10h

Good read ...

Datacenter Economics: Scale, Costs, Financing, and the Path to Equilibrium To build a data center, you need four things: compute (chips or GPUs), land, water, and electricity. In other words, it is capital intensive and takes about two years to build. If you have land and water but no chips or no electricity, you have a problem. America is where it is happening. The United States has land (especially in states like Texas), natural resources such as natural gas to power cheap electricity, and, under Trump before the Blue Wave, pretty willing investors and regulators. It also has chips (Nvidia and others). We will talk about outside of America soon, particularly Asia. People who talk about data centers talk in billions of dollars. A gigawatt is a measure of electricity—think of one nuclear plant, or the power used by 650,000 homes, or an American city. That is a lot of power, so the price of power matters. Cheaper is better for data center economics and competitiveness. You need water too, to cool the systems. It is about the water usage of 100,000 households per year, or 10.5 billion gallons of water. Land requirements are roughly 700 to 1,100 acres. For a visualization, think of Central Park, which is 843 acres—so a one-gigawatt facility is likely bigger than that. In short, one gigawatt of data center capacity equals the electricity of 650,000 homes, the water of 100,000 homes, and roughly the land of Central Park. You need resources, capital, land, water, electricity, and the willingness to build. Roughly, the facility itself costs about $15 billion, give or take, and this is without the compute or GPUs. An average one-gigawatt data center will cost about $15 billion without the compute. The compute will cost about $35 billion depending on the chips you use. Amazon uses cheaper chips so it is around $35 billion, while Meta uses more expensive ones so it is closer to $50 billion. This is just a ballpark. Chips, compute, and GPUs are about 70 percent of the cost of a one-gigawatt data center. Hardware exporters take the lion’s share of the profit here. Nvidia does not make the chips itself; it hires companies such as SK Hynix and TSMC to make them. Hence export earnings have been amazing for Asian-linked hardware exporters. When people talk about the winners of the AI and data center build-out, they are talking about South Korea, Taiwan, Japan, and others. Look at the price of DRAM and NAND. DRAM is the memory chip you need to use with power; without power it does not work, and it is a key part of Korea’s exports. NAND is more like the chips used for storage and does not need continuous power. Prices for NAND are rolling over, but not yet for DRAM, which is more important for companies like SK Hynix and Samsung Electronics. Concepts like chokepoint, bottleneck, or moat explain why you have higher prices. The question regarding these super-high prices and amazing exports is: how long can this last? If you look at the basic economics of a data center, assuming roughly $50 billion (it could be cheaper), and 70 percent of it is GPU, then you can see why people have an incentive to lower that cost. It is the biggest component, and it has been going up. For every parabolically higher level of exports, there is the other side of the equation: the importers, or the people who have to pay for it. Ultimately that includes you too. To keep things simple, we use round numbers. To build one gigawatt you need about $50 billion, and 70 percent of that is in the GPU portion. Note that acquiring land, electricity, and water can also be an issue, and one that is increasingly very political, because these are finite resources and inflationary before they raise productivity. As Mark Twain said, “Buy land. They don’t make land anymore.” Because GPUs are the most expensive part, two scenarios are both happening. Hyperscalers are coming up with their own versions of chips to reduce costs—something Jensen Huang was worried about. Google, Microsoft, and others all have their own development underway. Competitors to companies like SK Hynix are motivated to make more, because high prices create high motivation to make money (think China). And of course a company like SK Hynix has to ask itself: high prices mean demand destruction and more competition, so we need to increase supply as well. Whatever bottleneck or moat exists will not last forever. This is why the KOSPI sits at valuations that look very attractive because of sky-high earnings, but the question is whether those earnings will sustain in growth and performance forever. The answer is no; the question is when, not whether. To come up with $50 billion to build a data center, you have to fund it. Hyperscalers used to fund this with free cash flow—their own money. That is running out, so they borrow in various ways. The key concept is interest rates, or the price of money. Capex by hyperscalers is about $1 trillion this year, and the on-balance-sheet portion is basically about 10 percent of investment-grade bond issuance. Next year, capex will likely rise and will be funded via borrowing as cash flow goes negative. That will likely rise to 16–19 percent of global investment-grade bonds. This is where the Federal Reserve, the ECB, and the Bank of Japan come in. Central banks set the price of money. When the price of money goes up, the cost of capital goes higher. The impact is very widespread and complicated. When the cost of capital rises, it affects the economics of data centers and of everyone else. Money is not free and has high opportunity costs. By doing nothing and taking what we call a “risk-free” decision, you get 4.8 percent for the two-year U.S. Treasury and 5.44 percent for the 30-year, just by lending to the U.S. government. If yields rise you face mark-to-market losses, but you still get a decent coupon by not doing much. Assuming inflation stays below 4 percent, this would be a positive real yield. With the global liquidity pool shrinking as central banks tighten, and with demand for money rising because of more capex, investors have to choose between funding sovereigns such as the United States, France, Germany, or Japan, or funding hyperscalers. They want risk-adjusted returns. We already see spread differentiation among Microsoft, Amazon, Google, Meta, and Oracle, which pays the highest spread among them. Hyperscalers are Amazon (AWS), Microsoft Azure, Google Cloud, Meta, and Oracle Cloud Infrastructure. They are not all the same. There is differentiation in equity valuations and in the credit spreads they pay versus U.S. Treasuries (or German Bunds in euros). Hyperscalers have scale: the bigger ones have 35 gigawatts of capacity—equivalent to 35 Central Parks of land, the electricity of 35 medium-sized cities, and the water of 35 smaller cities. In dollar terms that is ballpark $1.7 trillion of data center investment (using $50 billion per gigawatt), and they are building more. They have breadth of income streams, which markets value. They have financial strength and strong ratings. Utilization is better because of a diversified workload mix, and they can deploy faster. Neo-clouds—think CoreWeave, Nebius, Lambda Labs, and Crusoe—are specialized providers of GPU compute for AI training and inference. They are smaller, accounting for about 3.6 gigawatts or 11 percent of total demand. They can be faster in deploying AI clusters and price lower than hyperscalers. Utilization is about 90 percent. Financially they are weaker because they need leverage, rely heavily on debt, and run lower-margin businesses renting GPU servers. Their focus is narrower. In markets this means neo-clouds pay double-digit interest rates, while hyperscalers pay a smaller spread over Treasuries. Within the hyperscalers there is further differentiation. Microsoft pays the cheapest because investors believe its return on investment is positive above the cost of capital. As rates rise, the more vulnerable players become more exposed and have to pay a higher risk premium. When companies report, they give expectations of future earnings and capex. Markets also have expectations. Capex growth for data centers means demand for chips, capital, and resources. Higher capex growth means a higher cost of capital, more revenue for Asian hardware exporters, and tighter natural resources that require more electricity build-out. The high-conviction view is that we are approaching an equilibrium level of capex and long-term structural interest rates. There are about 40 gigawatts of operational data center capacity in the United States. Asia has about 32 gigawatts by the end of 2026, Europe about 11 gigawatts, and not much in Latin America. The Middle East is also important because it has energy. Global operational total is ballpark 90–100 gigawatts by the end of 2026. Estimates for future build-out vary widely. A lower-end figure for the United States is an additional 50 gigawatts by 2030 and a higher-end figure is an additional 100 gigawatts. Global capacity figures from sources such as BNEF are around 173 gigawatts, with some estimates much higher. These estimates are going to have to be scaled down, and that is a good thing. There are two types of builders. Neo-clouds are very interest-rate sensitive and concentrated in revenue source. Hyperscalers are also sensitive but less so. If the AI data-center build-out continues as planned in a structurally higher cost-of-capital environment, these companies will issue more debt and push costs higher. Can they monetize even more to afford the higher cost of capital? Probably not. So they will scale plans down a bit, which makes their business models more sustainable. As they demand less funding from the market, that also helps keep the cost of capital more sustainable. We need a lot of data centers, but we do not need infinite capacity. Capital and resources are finite, and regulation matters. The United States has added more restrictions on how top-end chips get exported and who the end user will be, including limits on renting GPUs to China. For neo-clouds that is an important consideration; they need to rent out compute and find customers. China has been electrifying its economy. Land is not abundant but they can go to the deserts, so electricity and land are more manageable, though water remains an issue and chips are another. China can build out its own chips, but that takes time. Meanwhile this remains a chokepoint. China also has its own chokepoints it can impose on others, notably rare earths. Regulation and resources are pointing toward data-center capex by both hyperscalers and neo-clouds coming down to earth a bit. That has an impact on Asian economies, corporates, and households. We live in a structurally higher interest-rate environment, driven by triple-D dynamics: data centers, debt, and decoupling. These three forces are strong undercurrents in the global landscape and create divergence in performance across sectors and sovereigns. The good news is that we are approaching an equilibrium level of interest rates and capex that will make global growth more sustainable. For some, it will fade the parabolic rise of exports (think Korea) by 2028; for others, it provides breathing space (think India). As we approach 2027, while decoupling trends remain stronger for chips and data centers, geopolitical headwinds regarding energy and commodity shocks will fade, providing tailwinds for the most vulnerable Asian economies. For some central banks we are approaching the end of the hiking cycle (think Australia), leading to rates steadying or even providing some relief. Central banks do not have much wiggle room right now, but hiking cycles are coming to an end for some, leaving us in a structurally higher rates environment—one we can live with and thrive in more sustainably. Hopefully, now when you hear people talk about gigawatts and data centers, you have a visualization of what they mean.

orbitant avatar
orbitant@orbitant·11h

I did what many retail investors would probably do before buying an IPO. Asked ChatGPT to read the filing. In this case, Anthropic $ANT “Find all the red flags.” Here’s what it flagged, including updated figures from subsequent amendments. Some are existing issues. Others are risks disclosed by the company. 1. Massive spending commitments. $900.9 billion over multiple years, including $451.6 billion before end-2030. 2. Heavy near-term payments. Approximately $70.7 billion due within twelve months of September 2026. 3. Pay regardless of usage. Some compute contracts require payment even if the capacity sits unused. 4. Revenue can disappear faster than costs. Customers can reduce spending quickly. Infrastructure commitments stretch for years. 5. Defaults can spread. A commercial default could trigger debt defaults, accelerated repayments and loss of infrastructure. 6. Financing comes with conditions. Some funding depends on revenue targets and other requirements. Undrawn commitments aren't cash in the bank. 7. Paying before capacity arrives. Infrastructure payments can begin before usable compute becomes available. 8. Hardware can become outdated. Long contracts could leave Anthropic paying for older infrastructure while competitors adopt cheaper, better systems. 9. Less visibility into margins. Cost of revenue and gross profit are no longer presented separately. 10. Adjusted profits exclude economic costs. Stock compensation and charitable contributions still affect shareholders. 11. Complicated headline losses. The first-half $80.7 billion loss includes $72.4 billion of non-cash financing revaluations. It isn't equivalent to cash burn. 12. Cash-flow timing matters. Prepaid-asset drawdowns, increased liabilities and deferred revenue supported first-half operating cash flow. 13. Run-rate revenue isn't guaranteed revenue. It annualises recent activity, including a 28-day consumption window. 14. Retention figures exclude some subscribers. Pro and Max subscribers aren't included in net revenue retention. 15. Customer concentration. The top five generated approximately 25% of nine-month revenue. Many large customers can cut spending quickly. 16. Cloud distribution dependence. Marketplaces generated 48% of first-half revenue, creating reliance on intermediaries for sales and collections. 17. Constant pressure to deliver. Over 80% of third-quarter revenue came from model versions released within six months. 18. Partners have competing interests. Some are simultaneously investors, suppliers, distributors and competitors. 19. Rivals can undercut pricing. Larger companies can subsidise AI or bundle it into existing products. Cheaper open models add pressure. 20. Customers face their own risks. Startup exposure increases sensitivity to funding downturns. Larger customers can develop alternatives internally. 21. Distribution is expensive. Nine-month sales, marketing and partnerships expense reached $5.5 billion, including a $4.8 billion increase in partner fees. 22. Money doesn't solve every bottleneck. Chips, memory, electricity, water and construction timelines can constrain growth. 23. Energy costs can rise. Exposure includes fuel-price pass-throughs and commitments to cover electricity-cost increases caused by its data centres. 24. Safety controls have already failed. The company discloses failures, misconfigurations and controls being inadvertently disabled. 25. Technical failures can compound. Model errors, prompt injection and misuse create liability. Using Claude to develop successors can carry existing flaws forward. 26. Copyright exposure continues. The $1.5 billion settlement doesn't resolve remaining litigation. Further damages, licensing costs or restrictions remain possible. 27. Government disputes remain unsettled. Favourable rulings on some claims haven't removed the broader uncertainty. 28. Policy can interrupt products. Export restrictions previously forced Anthropic to disable Fable 5 and Mythos 5 until restrictions lifted. 29. Public shareholders have little control. One founder-controlled Class F share carries 50.1% of general voting power. 30. Board control and financial exposure are separated. The LTBT generally elects a board majority while holding no economic interest. 31. Control can survive substantial selling. Original founders need retain only 2% of their IPO-era holdings to satisfy the ownership condition. Other conditions apply. 32. No simple expiry date. Founder control has conditional sunset and succession rules. 33. Shareholder returns aren't the sole objective. Public-benefit obligations permit decisions that sacrifice financial returns to advance the mission. 34. More dilution is possible. The board can issue additional public-benefit shares at repeated share-price milestones. 35. Fewer governance protections. The nominating committee won't be entirely independent under controlled-company exemptions. Audit and compensation committees are independent. 36. Family concentration warrants scrutiny. The CEO and president/chair are siblings. The president's husband is also employed. 37. Future share supply and entrenched control. Options, RSUs and lock-up releases can create selling pressure. Anti-takeover provisions make changing control harder. 38. Founders selling while retaining control. Dario and Daniela each propose selling approximately $250 million at the assumed $74 IPO price. About 1.4% of their disclosed holdings each. To be fair, operating cash flow was positive $6.6 billion for the nine months to September 2026. The $900.9 billion also represents multiyear commitments, not debt due tomorrow. The concern is whether future demand, margins and financing can support those commitments. Using AI to check the risks of investing in AI. 😅 Very 2026. NFA. src: sec.gov/Archives/edgar/data/…

Post attachment
orbitant avatar
orbitant@orbitant·12h

.$ANT hype building up... FOMO retail investors will jump in 😅

BREAKING: Anthropic's IPO has now attracted nearly $500 billion in investor orders, per CNBC. Institutional demand has reached ~$400 billion, while retail orders are approaching $100 billion. The company is now considering expanding its IPO and pricing shares above $80, above the initial range of $70 to $78 per share. Anthropic is also considering raising more than the initially planned $64.4 billion.

orbitant avatar
orbitant@orbitant·14h

Is the 🇺🇸 US risking losing ground to 🇨🇳 China in ASEAN, a market of over 700 million people, roughly twice the population of America? Here's the latest on Washington's AI chip restrictions: 🇸🇬 Singapore: PM Lawrence Wong is prepared to facilitate US inspections of AI chip usage. Singapore will not allow companies to circumvent US export controls and is working towards maintaining trusted access to advanced American technology. 🇲🇾 Malaysia: Seeking clarification from Washington, investigating alleged chip diversion and urging data centre operators to comply with US restrictions. Nvidia partner YTL Power fell nearly 10% on Tuesday. 🇮🇩 Indonesia: Assessing the potential impact on data centre investments, semiconductor supply chains and domestic technology development. 🇹🇭 Thailand: Acknowledges that the restrictions could undermine its ambitions to become a regional data centre hub. Meanwhile, 🇺🇸 US, 🇦🇺 Australia, 🇮🇳 India and 🇯🇵 Japan announced the Quad Super Intelligence Accelerator, with over US$85 million in planned support for trusted AI systems across the Indo-Pacific. Washington wants deeper economic and technological cooperation with ASEAN while tightening access to its most advanced AI technology. Tighter licensing requirements and rising compliance costs could give US hyperscalers such as AWS, Microsoft Azure and Google Cloud a competitive advantage. Instead of building their own AI computing infrastructure, more ASEAN businesses may end up renting computing power from American cloud providers, increasing the region's dependence on US technology. However, these restrictions could also accelerate the adoption of Chinese chips, cloud infrastructure and open-source AI models across the region. ASEAN wants access to technology and investment from both the US and China, not to be forced to choose sides. Washington may strengthen America's cloud dominance in the short term, but at the risk of creating stronger Chinese competitors in the long term. src: CNA, Straits Times, Reuters, Bloomberg and Business Times.

Post attachmentPost attachmentPost attachment
2:26

Lawrence: "Let's get the chips over after we sort out the rules." Trump: "There's a lot of money here." With Washington's latest AI chip export restrictions, there's plenty for Southeast Asian leaders to discuss. 🇸🇬 Singapore: Engaging US authorities to protect access to advanced AI technology. 🇲🇾 Malaysia: Investigating suspected chip diversion and strengthening enforcement. 🇮🇩 Indonesia: Assessing the impact on investment and data centres. 🇹🇭 Thailand: Preparing tighter scrutiny of chip importers and data centres. 🇧🇷 Brazil: Also among the five countries identified as facing additional licensing requirements. The bigger question is whether these restrictions will slow Southeast Asia's AI infrastructure ambitions or give US hyperscalers an even bigger advantage. Washington and US hyperscalers will also need to consider their next moves carefully. Tighter restrictions could accelerate China's development of domestic AI chips, open-source models and alternative computing infrastructure. Restricting access may protect America's technological lead today, but it could also strengthen tomorrow's competitors.

0:31
orbitant avatar
orbitant@orbitant·Nov 17

The US Securities and Exchange Commission (SEC) has opened an investigation into Nvidia, amid the Jensen Huang situation, per WSJ. The focus of the investigation includes whether the company’s disclosures misled investors. The Justice Department is also now reportedly leading multiple investigations involving Alice Huang.

orbitant avatar
orbitant@orbitant·Nov 17

Google Threat Intelligence Group released tracking data showing agentic threat activity crossed a significant threshold on November 17, 2026. Adversaries now deploy autonomous multi-agent frameworks like ARTEX, CAIRN, CyberStrikeAI, and PentAGI to execute multi-stage intrusions at machine speed. Financially motivated groups monetize compromised enterprise environments through LLMJacking, stealing credentials and cloud resources to resell access to high-performance AI models via underground proxy services known as LLM Mixers. Attackers also use large language models for automated vulnerability research, reverse engineering patches via differential analysis to weaponize n-day vulnerabilities against internet-facing servers without human intervention. cloud.google.com/blog/topics…

Post attachment
orbitant avatar
orbitant@orbitant·Nov 17

OpenAI is rolling out two new models, GPT-6.2 Astra and GPT-6.2 Sol, designed to handle complex professional work and everyday tasks with lower token consumption. GPT-6.2 Astra focuses on demanding professional projects. It produces spreadsheets, documents, and slides that adhere closer to specific industry templates and standards in fields like law, finance, and accounting. On the Agents Last Exam benchmark evaluating multistep professional workflows, Astra scored higher than previous iterations while running at less than a third of the cost. In scientific benchmarks including Terminal-Bench-Science 0.1 and the Vals Science Index, the model handles terminal operations and biological data interpretation. API pricing for GPT-6.2 Astra features a 20 percent reduction from the previous version, alongside a 60 percent discount on cached input at eight dollars for input, forty cents for cached input, and forty dollars per million output tokens. GPT-6.2 Sol is built for efficiency in everyday and recurring workflows. It maintains the same API pricing as its predecessor while offering increased processing capability for multistep agentic tasks. On Cognition's FrontierCode 2.0 benchmark for codebase modifications, Sol approaches the performance of Astra at a fraction of the cost.

Post attachment
orbitant avatar
orbitant@orbitant·Nov 17

Open-source AI isn't just an alternative to OpenAI or Anthropic. It's an essential part of the ecosystem for AI and superintelligence (SI) to thrive. Jensen Huang understood that. Open models enable transparency, accessibility, independent scrutiny and innovation beyond a handful of powerful companies. Closed models have their place, but the future of AI shouldn't be controlled by a few corporations. If AI is going to benefit humanity, its development must remain open to competition, collaboration and independent builders. Losing Jensen as a champion of open AI would be a setback, but the ecosystem is bigger than any one company or individual. src: ft.com/content/418ca885-fbcf…

Post attachment
orbitant avatar
orbitant@orbitant·Nov 17

Better be Epic!

They haven’t flown 24 trillion miles to negotiate. The Emmy-nominated blockbuster 3 BODY PROBLEM returns January 14.

1:40