OpenAI IPO Slips, Anthropic Leads.

Mazen Al Sabbagh

— 7 min read
Share

TL;DR

  • OpenAI IPO Delay: OpenAI now targets a 2027 IPO, per NYT, citing market volatility and a $1 trillion valuation goal. SoftBank shares dropped 12%, and Bloomberg suggests Anthropic may list first.
  • Government Gating: The Trump administration mandated a staggered, customer-by-customer release for OpenAI's GPT-5.6, mirroring the control applied to Anthropic's Mythos, establishing a de facto government permission layer for frontier model launches.
  • Samsung's $648B AI Bet: Samsung Group plans a 1,000 trillion won ($648 billion) decade-long investment, with a significant portion allocated to new chip fabs in South Korea, driven by insatiable AI memory demand.
  • Global Tech Selloff: Friday saw an intensified tech market downturn: KOSPI triggered a circuit breaker, Samsung and SK Hynix fell ~10%, SoftBank dropped 12%, and Micron surrendered most of its post-earnings rally.
  • End of Tokenmaxxing: Enterprise clients from Uber to Lindy are abandoning "tokenmaxxing," shifting from maximum AI usage to cost-efficient open-source alternatives, anticipating both OpenAI and Anthropic's IPOs.

Lead Story: OpenAI IPO Slips, Anthropic Leads.

OpenAI is deferring its initial public offering until 2027, as reported by the New York Times and Bloomberg, reorienting one of the most anticipated technology listings. The company had been targeting a fall 2026 debut at a near-$1 trillion valuation, having filed confidentially with the SEC on June 9. Advisers presented CEO Sam Altman with a strategic choice: either accept a reduced valuation for an immediate listing or wait to secure the trillion-dollar mark. The company chose patience.

The underlying driver is market volatility. SpaceX's recent IPO, despite raising $85 billion at a $1.77 trillion valuation, saw its stock price plummet from $225 to approximately $151 by Friday morning, unsettling retail investor sentiment around mega-cap technology debuts. Investment bankers warned that SpaceX's whipsaw performance could temper enthusiasm for another prominent AI entity entering public markets.

The immediate repercussion was felt acutely by SoftBank. The Japanese conglomerate, with a projected $65 billion commitment to OpenAI by October, experienced a 12% decline on Friday, marking its steepest single-day drop since August 2024. The anticipated financial upside from OpenAI's public debut had propelled SoftBank's market capitalization past Toyota's last month; this valuation premium has now eroded. Bloomberg additionally reported that OpenAI's leadership now anticipates Anthropic will initiate its public offering first, despite both firms having filed confidential S-1s within a week of each other in early June. Kalshi traders currently assign a 73% probability to OpenAI announcing its IPO by June 2027.

This delay precipitated a broader market selloff on Friday. In Asia, the KOSPI index triggered another circuit breaker, with both Samsung and SK Hynix experiencing approximately 10% declines. In New York, Micron, which had rallied significantly post-Q3 earnings on Thursday, ceded over 5% of its gains. Apple's 6% drop on Thursday, attributed to increased MacBook and iPad pricing due to memory shortages, continued to pressure the Nasdaq, contributing to its anticipated weekly loss.

In Other News

Samsung's $648 Billion AI Infrastructure Play. Samsung Group is poised to unveil a comprehensive decade-long investment strategy on Monday, committing 1,000 trillion won ($648 billion) to bolster South Korea's next economic cycle. This initiative includes a potential 300 trillion won allocation for new chip fabrication facilities in the country's southwest, a politically nuanced effort to decentralize production from the concentrated capital region. Key executives from Samsung Electronics and SK Hynix will present this roadmap to President Lee Jae Myung. A presidential advisor indicated that projects initially scheduled for the 2040s might be accelerated to the mid-2030s, given the unexpectedly rapid escalation of AI-driven memory demand. Samsung holds 38% of global DRAM market share; this $648 billion commitment surpasses even the $500 billion Stargate project, ranking among the largest corporate capital outlays in history.

The Demise of Tokenmaxxing. A CNBC investigation published Friday confirmed a rapid pivot from "spend-at-all-costs" AI adoption to aggressive cost-containment among enterprise clients of OpenAI and Anthropic. Uber, for instance, exhausted its entire 2026 AI tooling budget by April after deploying Claude Code to 5,000 engineers, subsequently imposing spending tiers. Lindy CEO Flo Crivello stated a full migration of his company's traffic from Claude to DeepSeek, yielding millions in savings. Glean CEO Arvind Jain estimates 95% of enterprise AI usage still relies on frontier models, a ratio he deems unsustainable long-term. D.A. Davidson analyst Gil Luria cautioned that OpenAI and Anthropic's largest enterprise customers are likely to cap their unchecked token expenditure. This timing is critical: both companies near potentially landmark IPOs, and revenue growth predicated on unconstrained token consumption may prove less robust than their S-1 filings require.

Washington's Deepening Permission Layer: GPT-5.6 Gated, Fable 5 Deadline Missed. The Trump administration requested OpenAI to stagger the release of GPT-5.6 due to security concerns, as reported by The Information. CEO Sam Altman informed staff that the government would grant "customer by customer" approval during an initial preview period, with broader availability anticipated "a couple of weeks later." The Office of the National Cyber Director and Office of Science and Technology Policy initiated this request. Altman characterized the arrangement as "not our preferred long-term model." The pattern is now unambiguous: Fable 5 was shut down on June 12; GPT-5.6 was gated on June 25. Both frontier labs now operate under federal approval for model deployments. Concurrently, Commerce Secretary Lutnick's congressional deadline to articulate the legal basis for the Fable 5 export controls passed today without a public statement. Meanwhile, Rep. Nathaniel Moran (R-TX) introduced the AI Incident Reporting Act, mandating frontier developers to report dangerous capabilities to Commerce within seven days, marking the first targeted federal AI safety legislation directly influenced by recent events.

X / Social Pulse

Friday's discourse swiftly transitioned from "Anthropic's regulatory issue" to the broader implication that government is now gating all frontier AI. Dean Ball, a former Trump AI policy advisor now at OpenAI, asserted that requirements "change constantly and stay secret even to the officials applying them," a statement widely interpreted as evidence of an ad hoc, rather than principled, permission layer. The OpenAI IPO delay prompted split commentary, balancing prudent capital discipline against valuation anxiety; SoftBank's 12% drop underscored the extent to which the IPO windfall thesis had been baked into its valuation. "Tokenmaxxing" entered mainstream business lexicon following simultaneous features in CNBC and Forbes, with Lindy's full migration to DeepSeek highlighted as a signal that frontier pricing power might be weaker than presumed. Regarding Fable 5, Secretary Lutnick's congressional deadline passed by late afternoon without public comment, initiating a flurry of position-taking on Polymarket. TechCrunch's editorial perspective—"it's not about Anthropic vs. OpenAI anymore"—captured the evolving consensus that the permission layer is industry-wide.

One to Watch

Adobe Acquires Topaz Labs: Consolidating On-Device AI. Adobe announced a definitive agreement to acquire Topaz Labs, the Emmy-winning AI firm specializing in industry-standard video upscaling (Astra) and image enhancement (Wonder) tools. This acquisition integrates Topaz's proprietary Neurostream technology, which facilitates local execution of large AI models on consumer GPUs rather than cloud infrastructure, into Adobe's Firefly and Creative Cloud ecosystem. TechCrunch and Variety covered the deal, anticipated to close in H2 2026, with no disclosed price. Adobe is strategically consolidating AI creative tools amidst escalating competition from Canva and Blackmagic. Topaz's on-device AI capability is the key asset, positioning Adobe to deliver rapid, cost-efficient AI processing independent of cloud reliance—an essential efficiency imperative within the post-tokenmaxxing enterprise landscape.

Quick Hits

  • Amazon India Investment: CEO Andy Jassy announced an additional $13 billion investment in India for AI and cloud infrastructure, bringing Amazon's total commitments there to $48 billion for 2026-2030, following a meeting with PM Modi.
  • General Intuition Funding: General Intuition secured $320 million at a $2.3 billion valuation (TechCrunch), based on its premise that video game action data can train AI agents for real-world applications, bringing total funding to $454 million.
  • GPT-4.5 Retirement: GPT-4.5 will be retired from ChatGPT tomorrow, June 27, completing its 30-day sunset. Conversations will automatically migrate to GPT-5.5.
  • US-India Fable Talks: US Special Envoy Helberg informed Indian media of "sensitive national security" discussions with India regarding Anthropic Fable model access for "trusted partners," emphasizing a "gradual, measured approach."
  • AI Bills Vetoed/Passed: Arizona Governor Hobbs vetoed all three AI bills passed by the legislature, while California's ban on AI public school teachers advanced to Governor Newsom's desk.

The market's narrative earlier this week questioned whether Micron's $100 billion in contracts quelled doubts about an AI bubble. Friday's response arrived on two fronts. Financially, one data point did not establish a trend: Micron's post-rally gains largely evaporated, SoftBank's 12% decline exposed the fragility of IPO-driven investor confidence, and Samsung's $648 billion commitment failed to insulate its stock from the broader selloff. Yet, the more salient revelation emanated from Washington. The government's decision to gate GPT-5.6 mirroring the Fable 5 shutdown—customer-by-customer approval by federal officials with opaque criteria—signals that frontier AI model releases are no longer merely product decisions. They are now policy events. Both OpenAI and Anthropic confront an identical structural constraint: their advanced models cannot reach public deployment without governmental authorization. Concurrently, enterprise spending data reinforces a distinct trend: companies are not abandoning AI, but they are definitively discarding the premise that maximized usage inherently equates to optimal value. This "tokenmaxxing" reversal, timed precisely as both labs prepare for potential IPOs, introduces a fundamental question their S-1 filings will struggle to answer: how much of the reported revenue growth constituted genuine demand, and how much was merely an unsustainable spending surge now subject to fiscal correction?

Sources

Lock in. M. mazen@thorterminal.com

Read more