Three separate threads converged this week into one story: Anthropic, not OpenAI, is currently setting the terms of competition in frontier AI. Claude Fable 5, restored to global access on July 1 after a 19-day export-control suspension, now ranks first on independent Intelligence Index benchmarks with a score of 60 -- ahead of Anthropic's own Opus 4.8 at 56 and OpenAI's flagship GPT-5.5 at 55. Four of the top five spots on that leaderboard belong to Anthropic models, bracketing a single OpenAI entry.
That benchmark lead is showing up on the chip side too. The Information reported on July 2 that Anthropic has opened early-stage talks with Samsung Electronics about manufacturing a custom AI accelerator on Samsung's 2-nanometer process, using its advanced packaging to place compute closer to memory and cut data-transfer bottlenecks. The project is nascent -- no design work has started, and Anthropic hasn't decided what the chip needs to do -- but the signal matters: Anthropic recently hired Clive Chan, an early member of OpenAI's own custom-chip team, as part of a deliberate silicon buildout, echoing the paths Google (TPU), Amazon (Trainium) and OpenAI itself (the in-house 'Jalapeño' chip) have already taken.
The Samsung conversation isn't happening in isolation from Anthropic's balance sheet. Samsung, SK Hynix and Micron all participated in Anthropic's $65 billion fundraising round in May, giving the memory giants a direct stake in seeing Anthropic's compute roadmap succeed -- a tighter alignment between a foundry partner and a customer than OpenAI has disclosed with any single chip supplier.
“IPO valuations are also just that -- implied, not priced -- until an S-1 actually goes public and a roadshow tests real investor demand.”
Meanwhile OpenAI's posture looks defensive by comparison. Sam Altman has reportedly floated giving the US government a 5% equity stake in OpenAI, worth roughly $42.6 billion off the company's $852 billion March valuation, as part of a broader idea -- discussed with Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick -- that leading AI labs each hand Washington a slice modeled on Alaska's oil-wealth permanent fund. The move follows the government's 10% stake in Intel taken in August for $8.9 billion, but where that deal came with an explicit CHIPS Act rationale, OpenAI's overture reads more like an attempt to pre-empt political blowback than a negotiated industrial-policy win.
The revenue numbers reinforce the gap in momentum. Anthropic's annualized revenue reportedly climbed from $9 billion at the end of 2025 to more than $44 billion by May 2026, a run rate Anthropic says puts it on track for $47 billion and profitability by 2029 -- a year ahead of OpenAI's own targets. That growth is a direct input into Anthropic's confidential S-1, filed June 1 and targeting an October Nasdaq listing led by Goldman Sachs, JPMorgan and Morgan Stanley, seeking to raise more than $60 billion. Secondary markets already imply a valuation of $1.05 trillion to $1.15 trillion, a premium over the $965 billion mark from Anthropic's May round -- and ahead of where OpenAI's own IPO chatter (targeting $852 billion to $1 trillion) stood as of its confidential filing in May.
None of this means OpenAI is losing the war -- it still has by far the largest consumer user base and a first-mover brand advantage that Anthropic can't easily replicate. Fortune reported July 2 that Altman is pushing a 'new world order' framing for AI governance precisely because OpenAI feels the ground shifting under Google and Anthropic on technical benchmarks, and a five-percentage-point equity offer to Washington is a small price if it buys OpenAI room to operate without a heavier-handed regulatory response.
The bear case for reading too much into Anthropic's momentum: benchmark leadership is notoriously fleeting in this industry -- GPT-5.5 or a Google Gemini update could reclaim the top Intelligence Index slot within weeks, and Anthropic's Samsung chip talks are explicitly preliminary, with a real chance they never progress past exploratory conversations. IPO valuations are also just that -- implied, not priced -- until an S-1 actually goes public and a roadshow tests real investor demand.
What to watch: whether Anthropic's Samsung talks produce an actual design agreement or quietly stall the way many 'custom silicon' explorations do; whether OpenAI's 5% stake offer becomes a template other frontier labs (xAI, Google DeepMind) are pressured to match; and whether Anthropic's October IPO timeline holds or slips the way SpaceX's and OpenAI's own listings have shifted before.