Stargate
Stargate: Cross-chain infrastructure resource for asset transfers and interoperability.
ABAB Structured Brief
Stargate is indexed in ABAB Crypto Map under Cross-chain & Bridges. This page keeps the official site, category, tags, and related ABAB coverage together as a searchable crypto project profile. Official domain: stargate.finance.
Related News & Analysis
OpenAI: The Valuation Logic of a Technology Platform, Strategic Capital, and Compute Empire
Core view. OpenAI is no longer best understood as “just an AI model company.” By mid-2026, it looks more like an emerging full-stack intelligence platform: consumer products such as ChatGPT and Codex on top, enterprise agent infrastructure such as Frontier in the middle, and multi-cloud, multi-chip, data center, power, and custom-chip strategy at the bottom. OpenAI itself describes this as a flywheel in which more compute enables better models, better models enable better products, better products drive adoption and revenue, and revenue pays for more compute. Organizational evolution. OpenAI began in 2015 as a nonprofit with the mission of ensuring AGI benefits all of humanity. In 2019 it created a for-profit subsidiary, while keeping nonprofit control, and Microsoft invested $1 billion and partnered on Azure supercomputing. After the 2023 ChatGPT and GPT-4 breakout, governance pressure intensified. In 2025, OpenAI completed a recapitalization: the nonprofit became the OpenAI Foundation, the operating company became OpenAI Group PBC, and the Foundation remained in control. As of closing, the Foundation held 26% of equity, Microsoft about 27%, and the remaining 47% was held by current and former employees and investors. In June 2026, OpenAI said it had confidentially filed for a U.S. IPO. Technology platform. OpenAI now operates across at least four layers: consumer AI products, enterprise workplace products, developer APIs and tools, and agent distribution/execution products. Frontier is explicitly positioned as a platform for building, deploying, and managing AI agents that can do real work inside companies. On the developer side, OpenAI has moved toward agent-native infrastructure through the Responses API, built-in search and file tools, computer use, the Agents SDK, and observability. OpenAI said at DevDay 2025 that 4 million developers had built with OpenAI, ChatGPT had 800M+ weekly users, and the API platform processed 6B tokens per minute; by March 2026, the company said API throughput had risen above 15B tokens per minute. Strategic capital and partner network. Microsoft remains the historically central investor and infrastructure partner, with total funding commitments of $13 billion disclosed in Microsoft’s 2025 annual report. But OpenAI has clearly shifted beyond a single-cloud dependency. By 2025, Microsoft’s exclusivity on new capacity had been relaxed into a right of first refusal, while OpenAI gained flexibility to build elsewhere. In 2026, OpenAI announced $110B of new investment at a $730B pre-money valuation, including $30B from SoftBank, $30B from NVIDIA, and $50B from Amazon; by March 2026 it said the round closed at $122B of committed capital and an $852B post-money valuation. Oracle, SoftBank, NVIDIA, Amazon, CoreWeave, Google Cloud, Broadcom, AMD, AWS Trainium, and Cerebras all now appear in OpenAI’s disclosed infrastructure map. This is not just financing; it is a supply-chain and compute-alliance strategy. Compute and cost logic. OpenAI’s own disclosure says available compute grew from about 0.2GW in 2023 to 0.6GW in 2024 and roughly 1.9GW in 2025, while ARR grew from $2B to $6B to $20B+. That makes compute the clearest operating bottleneck in the business. External estimates from Epoch AI suggest OpenAI spent about $7B on cloud compute in 2024, with roughly $5B for R&D-related compute and about $2B for inference, and that much of the R&D spend went to experiments and unreleased work rather than just final training runs of released models. In 2025, Reuters, citing the Financial Times, reported about $34B in total spending, including roughly $19B in R&D; Reuters also reported, citing The Information, that OpenAI burned $3.7B in Q1 2026 against $5.7B in revenue. Public information is still incomplete, but the broad conclusion is clear: OpenAI has proven demand faster than it has proven durable cash generation. Infrastructure strategy and unit economics. OpenAI is moving from renting compute to locking in compute. Stargate is central to that shift. In July 2025, OpenAI said Oracle and OpenAI would develop an additional 4.5GW of Stargate data center capacity, taking total capacity under development above 5GW and more than 2 million chips; by September 2025, it said Stargate had nearly 7GW of planned capacity and more than $400B in investment over three years. In June 2026, OpenAI unveiled its first custom inference chip, Jalapeño, co-developed with Broadcom, saying early testing showed substantially better performance per watt than current state-of-the-art. Reuters reported the design is aimed at lowering inference costs and reducing OpenAI’s dependence on Nvidia GPUs. This is strong evidence that OpenAI is trying to become not only a model company, but also an infrastructure-design company. Business model and valuation expectations. OpenAI now monetizes across consumer subscriptions, enterprise seats, API usage, tool calls, and early-stage advertising. Public pricing shows Go at $8/month in the U.S., Plus at $20/month, and Pro at $200/month, while the free and Go tiers may include ads. On the API side, pricing spans from low-cost nano models to flagship models such as GPT-5.5 at $5 per 1M input tokens and $30 per 1M output tokens, with additional pricing for web search, file search, and containers. OpenAI also said in March 2026 that enterprise revenue was already more than 40% of total revenue and on track to approach parity with consumer revenue by the end of 2026. Valuation has risen accordingly: Reuters reported about $157B in October 2024, about $840B in February 2026, and OpenAI itself disclosed an $852B post-money valuation in March 2026. Based on the company’s statement that it was then generating about $2B in monthly revenue, that implies roughly a 35.5x annualized revenue multiple; using Reuters’ report of a $25B annualized run rate at the end of February 2026 against an $840B valuation implies roughly 33.6x. These are not normal SaaS multiples; they reflect a market belief that OpenAI may become a dominant intelligence platform plus scarce compute allocator. Controversies and unresolved questions. The biggest controversy is whether OpenAI can still credibly claim mission-first governance while operating at infrastructure scale and pursuing an IPO. Related issues include the 2023 board crisis, ongoing copyright litigation from The New York Times and multiple authors, and Elon Musk’s long-running allegation that OpenAI abandoned its nonprofit mission—though Reuters reported Musk lost that case in May 2026. There is also real competitive pressure: OpenAI remains the dominant consumer-facing AI brand by its own disclosure, but Ramp AI Index reported that Anthropic overtook OpenAI in business adoption in May 2026 among businesses in Ramp’s U.S. payments sample. The deepest open question is still economic, not technical: can OpenAI convert extraordinary usage and infrastructure scale into durable free cash flow before its capital commitments become too heavy? Public data is still limited, and some high-profile financial figures remain based on reporting rather than fully public audited filings.
VeChain & Sunny Lu: The Real Architecture of an Enterprise Blockchain Empire
VeChain is not merely a token project. It is better understood as a hybrid of enterprise-grade public blockchain infrastructure, a foundation-led operating body, a protocol economy, and a product suite serving both enterprises and end users. Its narrative has clearly evolved from anti-counterfeiting, traceability, and supply chains into sustainability, digital product passports, Web3 applications, and, by 2026, infrastructure for the agentic economy. Public records indicate at least two core co-founders. The most visible public founder is Sunny Lu, now CEO; in the MiCAR white paper, his legal name appears as Yang LU. The other crucial figure is Jay Zhang, who appears in current records as Jie Zhang. He was originally the co-founder and CFO and is now the chairman and legal representative of VeChain Foundation San Marino S.R.L. In practical terms, Sunny has been the strategic narrator, product-direction leader, and public representative, while Jay has been the architect of governance, finance, and risk structure. Sunny Lu’s edge is not that of a pure academic or protocol researcher. His public biographies consistently emphasize his enterprise IT and luxury-industry background, especially his role as CIO / IS&T Director of Louis Vuitton China. He has also explicitly framed VeChain as a project driven by applications first and technology second. That background explains why VeChain has always prioritized enterprise requirements, governance design, predictable operating costs, compliance dialogue, and integration tooling over maximalist decentralization narratives. VeChain’s growth path is unusually clear. It began with enterprise anti-counterfeiting and traceability, moved into public-chain infrastructure and dual-token economics, then into enterprise onboarding products such as ToolChain, and later pivoted toward sustainability, consumer-facing participation, digital product passports, and AI/agent infrastructure. It has not grown in a straight explosive line; instead, it has repeatedly changed gears and still remained alive—something few early enterprise-blockchain projects managed to do. Its most important assets are not only code and tokens, but also institutional networks. The hardest assets are the VeChainThor network, the VET/VTHO economy, the foundation treasury, and the wallet/governance/developer product stack. Its most important influence assets are long-term associations with DNV, PwC, Walmart China, BCG, UFC/Dana White, and more recently BitGo, Keyrock, Franklin Templeton, and Rekord. VeChain as an Organization VeChain’s official starting point is 2015. Official materials state that VeChain was founded in 2015 and describe VeChainThor as the smart-contract platform it created. Messari adds that the project initially moved under BitSE-supported private/consortium-chain conditions before becoming a more independent public-chain ecosystem. Official sources emphasize the 2015 project establishment and the 2018 VeChainThor launch; third-party research helps fill in the transition from a private-chain model to a public-chain model. The real organizational core is not a single company, but a layered structure. Whitepaper 1.0 presented the VeChain Foundation as the central operating body for daily development, community growth, business engagement, and technical maintenance of VeChainThor. Whitepaper 2.0 made it even clearer that the Steering Committee was the highest governance body, responsible for strategy, finance, protocol parameters, VTHO economics, and major votes. VeChain was therefore never designed as a purely leaderless blockchain. Its early governance model was defined by identifiable, vetted authority. Official documents repeatedly stressed that VeChainThor did not allow anonymous block producers. Authority Masternodes were approved by the Foundation/Steering Committee and required KYC; Whitepaper 2.0 even stated that, in the trial phase, public disclosure of an Authority Masternode’s status could be left to the node holder’s discretion. That design was enterprise- and compliance-friendly, but it naturally invited criticism from decentralization purists. VeChain has since tried to become more open. Messari in 2025 described its governance as semi-centralized, while also noting that the VeChain Renaissance roadmap aimed to reduce the top-layer role of the Steering Committee and increase protocol- and community-level governance. Official 2025 materials show that Hayabusa shifted core consensus from KYC-based PoA toward DPoS, while StarGate became the new staking and delegation gateway. In short, VeChain has been moving from “enterprise-friendly with strong central coordination” toward “more open, but still operationally pragmatic.” The Founders Sunny Lu’s family background is largely unavailable in reliable public English sources. His date of birth, place of birth, parents’ occupations, and family-class background cannot be firmly confirmed from high-quality public records. What can be stably confirmed is his education, enterprise IT career, senior Louis Vuitton China role, and the fact that he started VeChain in 2015. His educational profile is consistent across public sources. He is widely described as having graduated from Shanghai Jiao Tong University in Electronics and Communication Engineering. LinkedIn also shows CCIE and CBCP certifications. Official whitepaper material likewise confirms his Shanghai Jiao Tong University background in electronics and communication engineering. This makes him look like a genuine engineering-and-enterprise-IT operator rather than a pure crypto promoter. His strongest publicly verified career credential is Louis Vuitton China. Whitepaper 1.0 states that his most important role before co-founding VeChain was CIO / IS&T Director for Louis Vuitton China. Several conference bios also describe him as having spent nearly two decades as an IT executive in Fortune 500 firms. Third-party profiles often add earlier roles at Bacardi China and 3M China, but those are not laid out in the same detail in every official document, so the safest statement is that Louis Vuitton China and long Fortune 500 IT leadership are the most firmly verifiable parts of his pre-VeChain career. Sunny Lu entered blockchain through enterprise pain points, not protocol idealism. In an official VeChain article summarizing a Fenbushi interview, he explicitly argued that the industry had many technical people but too few product-minded people. He described VeChain’s philosophy as application-driven rather than technology-driven. That framing is one of the clearest windows into why VeChain was built the way it was. There were at least two major triggers behind his move into blockchain. One was his long exposure to authenticity, traceability, and trust problems in luxury and branded-product environments. The second was early exposure to Bitcoin. In a 2025 Cointelegraph profile, he discussed being scammed while trying to buy 100 BTC in 2012; rather than pushing him away from crypto, that experience deepened his interest in trust infrastructure. Sunny Lu’s actual role inside VeChain is best described as a product strategist, business architect, and chief public face. He is not primarily known for original consensus research. He is known for setting direction, reframing the narrative, pushing products into real markets, and assembling cross-industry networks. VeChain’s shift from supply chains to sustainability and then to DPP/AI-agent infrastructure still follows this same application-first logic. Jay Zhang is less visible but structurally essential. Official whitepaper material describes him as co-founder and CFO with more than 14 years of PwC and Deloitte senior-manager experience. He joined in 2015 to lead blockchain governance framework design and digital-asset management structure. In the current MiCAR record, as Jie Zhang, he is chairman and legal representative of the San Marino entity. That makes him much more than a finance manager; he is one of the core designers of VeChain’s governance, finance, and compliance backbone. Jay Zhang’s family background is also publicly limited. What can be consistently confirmed is his Shanghai Jiao Tong University education in electrical and electronic engineering and his long work in IT assurance, governance, and risk at PwC and Deloitte. Beyond that: public information is limited / cannot be confirmed for now. Brands, Assets, Capital, and Business Model VeChain’s real assets sit on four layers. The first is the VeChainThor public chain itself. The second is the VET/VTHO dual-token economic system. The third is the foundation treasury. The fourth is the product and access layer: VeWorld, VeBetter, StarGate, VeVote, VORJ, MaaS, and PoP. Current official product pages and docs make that structure explicit. Those products play different roles in a broader ecosystem design. VeWorld is the user wallet/super-app entry point; VeBetter is the consumer participation and incentive layer; StarGate is the staking/security participation layer; VeVote is the governance layer; VORJ is the no-code Web3-as-a-Service layer; MaaS is closer to a brand-facing marketplace layer; and PoP serves event verification and attendance-proof functions. They are not random additions—they help complete a wallet–incentive–governance–developer–brand-interaction loop. The treasury is one of VeChain’s most important hard assets. The official Q1 2024 Treasury Report put VeChain Foundation’s treasury at roughly $551 million at the end of Q1 2024. Later official 2024 financial reports showed the treasury’s dollar value falling with market conditions to roughly $305 million at the end of Q2 2024 and roughly $288 million at the end of Q3 2024. So VeChain is not a cashless shell, but its balance-sheet strength is still highly sensitive to crypto-market pricing. In terms of influence assets, DNV and PwC have been the two most important early institutional lines. DNV first partnered with VeChain in 2018, later acquired a minority stake, and became an Authority Masternode while jointly expanding products such as My Story. PwC served as both a client-network bridge and a legitimacy anchor in enterprise risk/compliance circles. In VeChain’s own 2021 interview summary, Sunny Lu said DNV and PwC upgraded their involvement from partners to investors. The DNV equity stake is clearly disclosed in DNV’s own announcement; PwC’s investor role is less transparently disclosed and should therefore be treated with more caution. Earlier capital support came from Fenbushi Capital and a broader advisory network. VeChain’s own 2021 article explicitly called Fenbushi an angel-round investor. Whitepaper 1.0 also listed figures such as Jim Breyer and Bo Shen in its advisory structure, showing that VeChain was embedded early in a cross-network that included Chinese blockchain capital, Silicon Valley capital, and enterprise consulting/certification circles. VeChain’s business model has never been based on fees alone. Whitepaper 2.0 explicitly listed revenue sources such as asset management and investment, consulting/development services for enterprises, professional training, and VTHO-supported service/solution packages. That means VeChain historically operated as a combination of industry enabler, protocol infrastructure provider, and treasury-backed ecosystem builder. Turning Points, Achievements, and Controversies The key turning points are easy to identify. 2015 was the project’s establishment; 2017–2018 was the move into foundation/governance/public-chain infrastructure; 2019 was the ToolChain/Walmart China platformization phase; 2021 brought the San Marino digital COVID certificate use case; 2023–2024 marked the sustainability and VeBetter pivot; and 2025–2026 brought Galactica, Hayabusa, StarGate, Rekord, and the AI/agent roadmap. VeChain’s most representative success is that it produced named, repeatable, real-world cases earlier and more persistently than most enterprise-blockchain projects. DNV’s My Story is explicitly built on VeChain’s public ledger; Walmart China’s traceability platform is cited in official case material; San Marino’s digital COVID certificate used VeChainThor for verifiable digital authenticity; and VeChain still reuses Walmart, BMW, and DNV as flagship proof points in its current materials. It is remembered less for radical theory and more for the persistence of enterprise-oriented deployment. Its biggest long-term criticism has been centralization. Official materials themselves make clear that early authority nodes were approved by the Foundation/Steering Committee, required KYC, and were not all necessarily public in identity. Messari’s 2025 research explicitly described the model as semi-centralized. This has been the most durable structural criticism of VeChain from the broader crypto world. The second major controversy is the 2019 security incident and the blocklist response. CoinDesk reported the theft of roughly 1.1 billion VET; Messari said VeChain attributed it to staff negligence and an improper wallet-creation process. VeChain’s later official financial report confirmed that token holders voted to permanently introduce a blocklist tied to the theft and permanently remove roughly 727.6 million VET from supply. Supporters see this as responsible loss containment; critics see it as proof that VeChain governance can intervene too heavily in the ledger. Today VeChain sits in a very unusual middle position. It is not a universal base-layer standard like Bitcoin or Ethereum, nor is it the hottest consumer chain of the moment. Yet it remains one of the few older public-chain projects that still combines a real legal entity, long-running enterprise case studies, a meaningful treasury, a structured product stack, and repeated narrative reinvention capacity. Official records place its headquarters in San Marino, with teams/offices across Asia, Europe, and the US; VET also remains actively traded, with CoinMarketCap showing a market cap of roughly $544 million and a rank around #80 at the time of retrieval. Sunny Lu’s current real-world position is that of an active veteran founder still directly shaping the project. He remains the CEO, public narrator, and external interface for VeChain. The official roadmap, partnership announcements, media appearances, and high-profile branding moves such as Dana White joining as advisor continue to revolve around him. VeChain has not yet become a founder-agnostic organization in the strong sense.
The OpenAI Empire: Ideals, Capital, Power, and the Founders Who Changed the World with AI
OpenAI did not begin as an ordinary startup, but as a nonprofit research organization with a strong public-mission narrative. It was launched in December 2015 as a nonprofit AI research company whose stated goal was to advance digital intelligence in the way most likely to benefit humanity as a whole. The launch materials publicly identified a core founding group that included Sam Altman, Elon Musk, Greg Brockman, Ilya Sutskever, Trevor Blackwell, Vicki Cheung, Andrej Karpathy, Durk Kingma, John Schulman, Pamela Vagata, and Wojciech Zaremba. OpenAI also disclosed that Sam, Greg, Elon, Reid Hoffman, Jessica Livingston, Peter Thiel, AWS, Infosys, and YC Research had collectively committed $1 billion in support. The real story of OpenAI is not simply that “a lab got bigger,” but that a mission-constrained organization was repeatedly reshaped by compute needs, product expansion, and capital structure. In 2019, OpenAI shifted to a capped-profit hybrid structure because frontier model development, talent competition, and supercomputing required funding on the scale of billions of dollars. In 2025 it reaffirmed that the nonprofit would retain control, while transitioning the for-profit arm into a Public Benefit Corporation. After the October 2025 recapitalization, the OpenAI Foundation held about 26% equity, Microsoft about 27%, and current/former employees plus other investors about 47%, while the Foundation retained the power to appoint and remove the for-profit board. If reduced to one sentence, OpenAI is no longer just an AI company. It is simultaneously a research lab, a mass consumer platform, an enterprise software vendor, a compute infrastructure organization, a policy actor, and a capital-intensive quasi-platform company. That position was not created by a single “genius founder” narrative, but by a combination of roles: Sam Altman as organizer, fundraiser, external operator, and political narrator; Greg Brockman as engineering systems builder and recruiter; Ilya Sutskever as scientific and safety-minded research leader; and figures like Schulman, Karpathy, Zaremba, and Kingma as builders of reinforcement learning, generative modeling, alignment, and technical systems. OpenAI’s greatest success is not merely that it produced strong models, but that it stacked frontier models, global distribution, commercialization, and public influence into one machine. In 2020 it released the API as its first commercial product and explicitly argued that commercialization would fund continued research, safety, and policy work. In November 2022 it released ChatGPT as a research preview and pushed itself from the research world into mainstream culture and everyday work. By late 2025, OpenAI said it had more than 1 million direct business customers, and by January 2026 it said ChatGPT had over 700 million weekly active users and that 2025 ARR had surpassed $20 billion. Sam Altman is the OpenAI founder most recognizable as a power integrator. Public reporting says he was born in Chicago in 1985, grew up in affluent suburban St. Louis, had a dermatologist mother and a father involved in real estate and housing issues, attended the private John Burroughs School, and got early exposure to computing through a Macintosh. He studied computer science at Stanford but left without completing the degree, founded Loopt, later sold it, moved into investing, and eventually became president of Y Combinator. That path helps explain why he differs from a purely academic founder: his strength is in combining talent, capital, narrative, and organizational scale. Greg Brockman is the clearest example of a systems builder inside OpenAI’s founding group. He grew up in North Dakota and has publicly emphasized the importance of the local community and educational flexibility that let him take large numbers of University of North Dakota classes while still in high school. In his own writing he said he seriously started programming during a gap year after reading Turing’s “Computing Machinery and Intelligence,” and that he became fascinated by the idea of building code that could understand something its own author did not fully understand. He later attended Harvard and MIT without completing a degree, joined Stripe in 2010, worked deeply on backend infrastructure, became its first CTO, and then left to help form OpenAI. Ilya Sutskever was the closest thing OpenAI had to a scientific soul. Public biographical sources describe him as born in Russia, raised in Israel, captivated by computers from the age of five, and later relocated with his family to Canada. He completed his PhD at the University of Toronto, became a central co-inventor of AlexNet and sequence-to-sequence learning, worked at Google Brain, and then left Google in late 2015 to cofound OpenAI, where he served for years as chief scientist. His significance is unusual: he was both one of the hardest-core researchers of the modern deep learning era and later one of the strongest voices around superintelligence safety and alignment. Elon Musk and the broader technical cohort gave OpenAI both funding glamour and unusually high research density from the start. Musk was a co-chair and early funder but left the board in 2018 and later became adversarial. John Schulman represented the reinforcement learning and RLHF line; Andrej Karpathy embodied computer vision, deep-learning pedagogy, and later product intuition; Wojciech Zaremba represented algorithms, robotics, code models, and later resilience work; Durk Kingma represented generative-model methodology; Trevor Blackwell, Vicki Cheung, and Pamela Vagata were more closely tied to early engineering foundations, robotics, and systems building. For several of these figures, the best-documented public information concerns education and career rather than family background, so public information on private family history is limited. OpenAI’s institutional core was never simply “open source”; it was always about how to pursue AGI while claiming that mission should outrank shareholder interests. The 2018 Charter states that OpenAI’s mission is to ensure AGI benefits all of humanity, that its primary fiduciary duty is to humanity, that it is committed to long-term safety, and even that if another aligned, safety-conscious project came close to building AGI first, OpenAI would stop competing and start assisting. That Charter matters because it later became the baseline against which every commercialization and governance controversy was judged. But the Charter quickly collided with compute reality. In 2019 OpenAI said it had learned firsthand that the most dramatic AI systems required enormous computational power and that it would need to invest billions in large-scale cloud compute, talent, and AI supercomputers. Its capped-profit solution was designed so that employees and investors could receive capped returns, while value beyond the cap would flow back to the nonprofit. The nonprofit board still controlled the entity, and employees and investors formally agreed that the Charter came first, even above their financial stake. OpenAI’s business model was not a sudden betrayal of its mission; it was a progressively constructed response to frontier-scale costs. When OpenAI launched the API in 2020, it explicitly gave three reasons: commercialization funds research, safety, and policy work; APIs let smaller organizations benefit without training giant models; and APIs are easier to control than openly released weights when misuse is a concern. That logic already revealed a major shift: OpenAI was becoming a controlled deployment platform, not a purely open lab. ChatGPT’s 2022 launch then brought that platform into the consumer world. Its revenue logic is best understood as “frontier model capability × distribution × enterprise packaging × steady compute supply.” OpenAI’s own 2026 materials say its product layer spans text, images, voice, code, and APIs and is moving toward agents and workflow automation. OpenAI also says 2025 ARR exceeded $20 billion, compute capacity grew from roughly 0.2 GW in 2023 to about 1.9 GW in 2025, business customers surpassed 1 million, and ChatGPT weekly active users surpassed 700 million. By that point, OpenAI was no longer simply selling model calls; it was selling an intelligence layer increasingly embedded in both personal and organizational workflows. Microsoft is the single most important outside organization in OpenAI’s transition from lab to industrial-scale platform. In 2019 Microsoft invested $1 billion and became OpenAI’s exclusive cloud provider, with the two sides jointly developing Azure AI supercomputing technologies. In the 2025 structure, Microsoft held roughly 27% of OpenAI Group PBC, and the companies continued renegotiating key terms as OpenAI moved toward a new structure and a possible future IPO. Microsoft is therefore not just a funder, but also a compute provider, a distribution partner, and a long-term strategic counterpart. By 2025–2026, OpenAI’s assets fell into two different categories. The first are directly capitalizable assets: ChatGPT distribution, enterprise relationships, model IP, compute and chip contracts, PBC equity structure, infrastructure bets like Stargate, and hardware expansion via the Jony Ive deal. The second are influence assets: the Charter, AGI narrative authority, safety discourse, public-policy access, and regulatory visibility. The first category supports valuation; the second supports the organization’s ability to keep scaling under political and social pressure. The first decisive turning point was that OpenAI defined itself from day one as an AGI organization, not a narrow AI startup. That choice meant it would require the best researchers, huge compute, and long-duration capital from the beginning. The original launch post explicitly positioned Ilya as research director, Greg as CTO, and the broader group as world-class research engineers and scientists. This was not a conventional startup team; it was effectively a founding research institute. The second decisive turning point was the 2019 acceptance that mission would have to be amplified through capital. OpenAI LP was not a side detail; it was the most important structural break in company history. It marked the moment OpenAI stopped treating pure nonprofit research as a sustainable end-state and accepted that frontier AI competition required financing, equity incentives, and platform-scale compute. Almost every later dispute—Microsoft dependence, Musk litigation, nonprofit control, PBC conversion, IPO talk—flows from this decision. The third decisive turning point was the choice of controlled deployment over full openness. The Charter already suggested that safety and security concerns would reduce traditional publication over time, and the API page explicitly argued that API access was safer than releasing open weights. Product launches, enterprise sales, subscription tiers, and workflow integration later confirmed that OpenAI had fused safety, business, and distribution into one institutional choice. That is why “how open is OpenAI really?” has remained a persistent argument. The fourth decisive turning point was ChatGPT’s release in 2022. Its importance was not simply that a chatbot became famous, but that OpenAI gained direct user distribution, brand control, and enterprise lead generation at planetary scale. By 2025–2026, OpenAI itself described the company as simultaneously operating Research, Compute, Applications, and a very large nonprofit footprint. The addition of Fidji Simo showed that applications and execution were no longer side functions orbiting research, but pillars in their own right. The fifth decisive turning point was the 2023 board coup and the rapid reversal that followed. In November 2023, OpenAI’s board said it no longer had confidence in Sam Altman and that he had not been consistently candid in his communications. Days later, under pressure from employees, Microsoft, and investors, Altman returned, Greg Brockman returned as president, and by March 2024 a WilmerHale-backed review led the board to declare full confidence in Sam and Greg’s leadership. The long-term consequence was profound: instead of weakening Altman, the episode ultimately strengthened his position. The sixth turning point was the diaspora of founders and senior researchers. In May 2024 Ilya left OpenAI and soon launched Safe Superintelligence, explicitly stating that its only goal and product would be safe superintelligence. John Schulman left for Anthropic in 2024 and then moved to Thinking Machines in 2025 as chief scientist. Andrej Karpathy moved into Eureka Labs and then joined Anthropic’s pretraining effort in 2026. Wojciech Zaremba moved into the OpenAI Foundation’s AI Resilience work in 2026. The result is that the “OpenAI founding network” no longer exists inside one firm; it has spread across the frontier AI sector. The seventh turning point was the 2025–2026 escalation into compute and interface control. The SoftBank-led $40 billion round valued OpenAI at $300 billion; Stargate tied it to an infrastructure plan of up to $500 billion; and the $6.5 billion all-stock io deal with Jony Ive moved it beyond software and models toward the device layer. This showed that OpenAI’s ambition had expanded from model leadership to control over compute, distribution, and future AI-native hardware interfaces. The biggest controversy remains whether OpenAI betrayed its founding nonprofit mission. Musk’s legal case crystallized that accusation, claiming OpenAI, Altman, and Brockman had departed from nonprofit principles and enriched themselves through commercialization. OpenAI responded publicly that Musk had understood the need for a for-profit structure and had at one point wanted control himself. In May 2026, Musk lost in court on timeliness grounds, which gave OpenAI an important legal victory, yet Reuters also reported that the case reopened serious questions about Altman’s leadership style and credibility. Legally, OpenAI won; reputationally, it did not emerge untouched. The second major controversy is governance credibility. When Altman was removed in 2023, the public explanation was only that he had not been consistently candid with the board. By March 2024, the company had completed a review and restored confidence in him. That sharp reversal has left a durable question: was the crisis fundamentally about AI safety, leadership style, or control? Because the full internal record has never been made public, the underlying reasons remain disputed and cannot be fully verified from public sources. The third major controversy is whether safety has been subordinated to speed. In 2024 OpenAI created a Safety and Security Committee as it began training its next frontier model, yet former safety figures such as Jan Leike argued that safety culture and processes had taken a back seat to “shiny products.” So the issue is not whether safety structures exist—they do—but whether they are independent and powerful enough to constrain the pace of expansion. The fourth major controversy concerns copyright, consent, and training data. In 2025, multiple suits by authors and news organizations against OpenAI and Microsoft were consolidated in New York. In 2024, the Scarlett Johansson “Sky” voice dispute raised another kind of consent problem, after Johansson said the voice was eerily similar to hers and OpenAI paused it. Like other frontier AI labs, OpenAI does not merely face controversy; it faces controversy at a scale that turns each dispute into an industry-wide stress test. The fifth major controversy is that attrition has come to symbolize perceived value drift. As founders and top researchers moved to Anthropic, Thinking Machines, SSI, and elsewhere, critics increasingly argued that OpenAI had shifted from a research-and-safety-first lab into a product, finance, and expansion-first supercompany. That claim is not settled as fact, but it has become strong enough that every major departure revives it. At the same time, OpenAI has not completely abandoned mission constraints; it has tried to re-embed them in the new structure. Officially, the nonprofit still controls the company; the Safety and Security Committee remains at the Foundation level; Zaremba was put in charge of AI Resilience; and the Foundation said it expected to deploy at least $1 billion over the next year across life sciences, jobs and economic impact, AI resilience, and community programs. The dispute is not over whether these arrangements exist, but whether critics believe they are strong enough to counter the logic of hyper-capitalized expansion. As of 2026, OpenAI’s identity is now highly layered. Sam Altman remains CEO and says he directly oversees Research, Compute, and Applications while also working with the board on the nonprofit. Fidji Simo is being brought in to lead Applications. Greg Brockman continues as president and publicly represents the company in large infrastructure efforts. The Foundation board comprises Bret Taylor, Adam D’Angelo, Sue Desmond-Hellmann, Zico Kolter, Paul Nakasone, Adebayo Ogunlesi, Nicole Seligman, and Sam Altman. OpenAI therefore looks less like a classic startup and more like a multi-pillar super-organization coordinated by Altman. In industry terms, OpenAI remains one of the very few organizations able to combine frontier training, global user distribution, enterprise sales, and infrastructure finance at full scale. With more than 700 million weekly active users, more than 1 million business customers, more than $20 billion ARR, Stargate-level infrastructure ambitions, and a continually revised long-term relationship with Microsoft, it increasingly resembles a next-generation computing platform company rather than a conventional SaaS vendor. That is an inference, but it is a well-supported one based on public data about usage, revenue, structure, and compute strategy. Who inherits, respects, criticizes, and extends OpenAI today? Almost the entire frontier AI sector does all of those things at once. Anthropic has taken in Schulman, Karpathy, and Durk; Ilya continues the safe-superintelligence thesis through SSI; Thinking Machines built around former OpenAI talent including Schulman; the Foundation gave Zaremba a resilience role; and figures like Pamela Vagata, Vicki Cheung, and Trevor Blackwell continue to extend influence through investing, founding, and robotics. OpenAI’s deepest legacy, then, is not only what remains inside the company, but the way it has seeded a generation of frontier AI institutions. The reason the world remembers OpenAI is not only ChatGPT. More fundamentally, OpenAI fused four things that had rarely been fused inside one institution before: frontier research, commercial deployment, AGI safety language, and enormous capital-organizing capacity. Academia had the science, investors had the money, internet companies had distribution—but OpenAI brought all four into one organizational template. Even the firms that criticize it are, in different ways, building within a template it helped define. The clearest bottom-line summary is this: OpenAI was built by a coalition of researchers, systems engineers, and one especially strong capital-and-organization founder. It formed influence through frontier model capability, deep alignment with Microsoft and later capital networks, ChatGPT-scale distribution, an enterprise platform, and a persistent AGI narrative. It owns powerful brand, platform, and infrastructure assets, as well as unusually strong influence assets. Its success lies in turning AI from a research frontier into an everyday global tool; its controversy lies in mission, governance, safety, and copyright; and its real-world position is now close to that of a new general computing infrastructure. There are still hard limits in the available public record. For several lesser-known founders—especially Trevor Blackwell, Vicki Cheung, Pamela Vagata, and some phases of Wojciech Zaremba’s career—public information is rich on education and work history but thin on family background, parents, and childhood resources. Those aspects therefore cannot be reconstructed with high confidence from public sources alone. Figures about valuation, future IPO, ad revenue, and 2030-scale forecasts should be handled carefully. Many of them are company statements, media reports, annualized run-rate metrics, or forward-looking investor presentations rather than audited historical accounts. They are useful for understanding direction and capital expectations, but they should not be read as fixed accounting facts. The single most important unresolved question remains the true internal cause of Altman’s 2023 removal. Public sources still do not fully unify the story of what the board knew, what evidence it had, and whether the deepest conflict was primarily about safety, candor, or control. That remains the sharpest evidentiary limit in this entire subject.
Dialogue with Sam Altman: OpenAI's Strategic Choices, Computing Bets, and Underlying Thoughts on Commercialization
1. Future Trends and the Rhythm of AI Implementation: Why is Social Evolution Slower than Technology? 1. Industry Pioneer Case: Shopify CEO Tobi Lütke's Extreme Sensitivity • Deeply Involved in Code and Products: Tobi Lütke shows strong sensitivity in the AI field, always ahead of the industry by 6-8 months; as a large enterprise CEO, he personally writes code, restructures workflows, and provides extremely precise product details to OpenAI. • Actively Reshaping Company Form: He insists that "Shopify will never be a passive recipient (NPC Company)" and advocates for actively embracing agents, even attempting to rewrite Shopify with a new AI-native architecture at night. 2. Technological Disruption vs Economic and Social Inertia • Actual Delay in Disruption Cycle: Sam Altman once believed that the software industry would rapidly reshape after the launch of GPT-4, but reality shows that the speed of social evolution is much slower than pure technological development. • Barriers of Habit and Switching Costs: History repeatedly proves (e.g., Larry Ellison's discovery that installing software is easy, but changing user habits is difficult; Netflix had already mailed DVDs, but the public still preferred Blockbuster), the economic system has significant inertia, and the public tends to stick with familiar ways of working and suppliers. • Anti-Cyclical Nature of Non-AI Native Experiences: The more high-tech becomes prevalent, the more fields with real interpersonal connections, physical experiences, or emotional recognition (such as offline experiences, sports events) possess a solid moat. 3. Psychological Resistance of Habits and the Absence of "iPhone-Level Interaction" • Contradiction in Founders' Own Behavioral Inertia: Sam Altman admits to having 20 years of traditional computer operation habits (mechanically checking emails, copying and pasting, listing to-do items), and even with powerful Codex agent tools, he often finds it difficult to fully switch to a pure AI workflow due to psychological coding habits. • Currently in the "Palm PC Era Before the iPhone's Birth": Similar to the Palm Treo or Sidekick in 2003/2004, the underlying technology modules are ready, but a truly revolutionary super product that fundamentally changes user interaction interfaces has yet to emerge. 2. OpenAI's Strategic Positioning: Extreme Focus and Abandoning "Good Ideas" 1. Core Strategy: Transitioning from a "Product Company" to "Platform Infrastructure" • Integrating Core Entry Points: Deeply merging ChatGPT and Codex to create a unified personal and enterprise AGI interaction entry point, supported by powerful underlying APIs. • Covering the Full Cost-Performance Curve: • High-End Scenarios: Providing cutting-edge super intelligence for frontier scientific discoveries and complex reasoning. • Low-End Scenarios: Offering cheap, efficient, high-throughput computing power to support massive daily tasks. • Not Competing with Ecological Customers: OpenAI's goal is to become the underlying platform supporting 100 million startups and 8 billion users, rather than extending its reach into all vertical application tracks. 2. Abandoning Good Ideas, Fully Betting on Ultimate Goals (Focus & Trade-offs) • Cutting Sora and Atlas Browsers: • Sora Video Generation: Although it has innovation and entertainment value, it consumes an enormous amount of computing resources, and after weighing against the core strategy, it was decided to make way for key intelligent reasoning. • Atlas Web Browser: The product experience was excellent, but it was decisively terminated to avoid distracting top R&D talent. • First Principles Focus: Given the limited reality of computing power, talent, and resources, OpenAI will focus all its efforts on "general intelligence leading to knowledge work and scientific discovery," covering self-developed chips, infrastructure software, self-built data centers, and model pre-training. 3. Personal Energy Allocation and Computing Infrastructure Challenges • Focusing on Research and Compute: Products are built by excellent teams, while Sam Altman invests most of his energy in model research and building the computing supply chain. • The Most Expensive Infrastructure Project in Human History: The expansion of computing power crosses complex geopolitical policies, chip design, wafer foundry capacity, rack manufacturing, power and energy system scheduling, and large-scale financing structures. 3. From Investor to Research Operator: Non-Consensus Betting and Research Management 1. The Underlying Commonality of Venture Capital and Cutting-Edge Research • Dominated by the Power Law: In AI research and investment, a few non-consensus breakthroughs (such as early bets on LLM and AGI) create value that can completely overshadow all conventional projects combined. • Identifying Non-Standard Extreme Talent: Rejecting mediocre entrepreneurs/researchers who follow popular concepts, focusing on selecting "outliers" with strong independent thinking, the courage to adhere to obscure non-consensus hypotheses, and extreme conviction. 2. Breaking Conventional Startup Paths: The Darkest Moment of Not Releasing Commercial Products for 4.5 Years • Research Exploration Against YC Conventional Rules: From its establishment at the end of 2015 to around 2020 when the first commercial product was launched, the team had no real user feedback signals for 4.5 years. • Building Internal Signal Replacement Mechanisms: • Using Dota 2 Reinforcement Learning Ranking Leaderboard to build objective measurement standards. • Introducing high-standard external expert demo presentations to drive R&D breakthroughs. • The Confusion of the 2016 Apartment Cold Start: At its inception, 12 people were in Greg Brockman's apartment without even a whiteboard, gradually establishing the research rhythm from unsupervised sentiment analysis, GPT-1 to Scaling Laws. 3. The Cognitive Weight of Success and Failure Experiences • The Value of Learning from Success Far Exceeds That from Failure: • The reasons for failure are varied, and often only generalized conclusions about "perseverance" can be extracted from them. • Deeply understanding the core elements of success (such as the key grips that YC and OpenAI got right) and amplifying them through compounding is key to driving business leaps. 4. Think Tanks and Cognitive External Brains: The Influence of Peter Thiel and Paul Graham 1. Extremely Non-Linear Thinking Inspiration • When facing extremely tricky strategic bottlenecks, the main think tanks that can provide counterintuitive perspectives are Paul Graham and Peter Thiel, whose non-linear thinking can directly break through mental deadlocks. 2. Peter Thiel: Doubling Down on the "Blank Input Box" • Key Guidance During ChatGPT's Initial Confusion: Within two months of ChatGPT's launch, although there was growth, due to the lack of the information flow (Feeds), network effects, and user lock-in mechanisms that Silicon Valley valued at the time, the team considered shifting to 5-6 other directions. • The Power of Minimal Essence: Peter Thiel pointed out that this was the most powerful "blank search box" form since Google, capable of inputting everything and outputting correct results, and it was essential to double down on this core interface, which completely ended internal wavering. 3. Paul Graham: Rapid Iteration and Early Release (Iterative Deployment) • "You should release when the product makes you feel embarrassed": Pushing the initial version to market early to face real-world feedback and iterate quickly is the core gene that determines startup success rates. 5. AI Safety, Social Governance, and Human-Centricism 1. Agile Iteration is the Optimal Path to Achieve Safety (Iterative Safety) • Stepping Out of the Ivory Tower: True AI safety cannot be achieved through pure theoretical deduction in closed laboratories; models must be pushed to hundreds of millions of real users to discover hallucinations, alignment failures, and vulnerabilities in real edge scenarios, and establish a transparent review and improvement system similar to civil aviation accident investigations (FAA). • Co-evolution of Models and Society: Society needs time to adapt to technology, and technology also needs to establish robust boundaries through real interactions. 2. Beware of "Power Rent-Seeking and Centralization" Under the Guise of Safety • Firmly Opposing Anti-Human Governance Views: • Be wary of arguments that deprive the public of technology usage rights in the name of protecting humanity, concentrating superintelligence control in the hands of a few oligarchs or a single AI decision-making body. • Firmly oppose the authoritarian technological concept of "exchanging freedom and decision-making power for the elimination of diseases and cheap materials." • Empowering the Public with More Autonomy and Leverage: The ultimate significance of AI is to grant ordinary people greater creativity and freedom of action, and the future will witness the largest wave of small and micro enterprises and individual entrepreneurship in human history. 3. The Next Stage of AI: Reconstruction of Decision-Making Through Ultra-Long Context • From Model IQ to Individual Context Empowerment: Current models have significantly improved intelligence, and the next breakthrough point lies in AI's ability to digest and refine ultra-massive context (internal documents, communication records, vast papers) in seconds, becoming an indispensable high-dimensional think tank for humans when making significant decisions. • The Fundamental Connection of Humanity is Irreplaceable: No matter how advanced superintelligence develops, the human desire for real interpersonal connections, emotional resonance, and physical interactions remains a core foundation.
Paxos Stablecoin USDG Natively Launches on Arbitrum
... Maple. Kraken will support deposits and withdrawals, while Stargate will support transfers between other chains. CoinDesk's list is longer, also including Li.Fi, Gauntlet, Steakhouse, and LayerZero, with Uniswap and Fhe...