## The Week in Creative AI & Media

The Week in Creative AI & Media

The infrastructure wars heated into new territory this week. <cite index="8-1,8-2">Anthropic signed an agreement with SpaceX to use all of the compute capacity at their Colossus 1 data center, giving Anthropic access to more than 300 megawatts of new capacity (over 220,000 NVIDIA GPUs) within the month.</cite>^[1] Industry observers estimate <cite index="9-3">the deal at roughly $5B/year, with Claude ARR growth at 8000% annualized</cite>^[2] following the announcement. <cite index="13-21">Anthropic are getting Colossus 1, but xAI are keeping their larger Colossus 2 data center for their own work.</cite>^[3]

In voice AI, <cite index="51-5">OpenAI introduced GPT-Realtime-2, its first voice model with GPT-5-class reasoning that can handle harder requests and carry the conversation forward naturally.</cite>^[4] <cite index="53-19">The context window is now 128K, up from 32K, which makes longer sessions and complex agentic flows feasible without external state stitching.</cite>^[5] OpenAI also released <cite index="51-6">GPT-Realtime-Translate, a new live translation model that translates speech from 70+ input languages into 13 output languages while keeping pace with the speaker</cite>^[6] and <cite index="51-7">GPT-Realtime-Whisper, a new streaming speech-to-text that transcribes speech live as the speaker talks.</cite>^[7]

Movements

• Anthropic secures SpaceX compute megadeal. <cite index="8-4,8-5,8-6">Anthropic signed an agreement with SpaceX to use all of the compute capacity at their Colossus 1 data center. This gives Anthropic access to more than 300 megawatts of new capacity (over 220,000 NVIDIA GPUs) within the month. This additional capacity will directly improve capacity for Claude Pro and Claude Max subscribers.</cite>^[8]

• Anthropic ARR surges to $30B annualized. <cite index="72-2,72-3">Anthropic is now at $30 billion annualized run-rate. OpenAI is at $24 billion ($2B per month, confirmed by the company).</cite>^[9] <cite index="72-14,72-15">$30 billion in April. That last sequence — $14B to $30B in roughly 8 weeks — is hard to make sense of in traditional software terms.</cite>^[10]

• xAI dissolves into SpaceX, becoming infrastructure provider. <cite index="15-15">Musk said that xAI "will be dissolved as a separate company" and will be called SpaceXAI.</cite>^[11] The deal signals a strategic pivot: rather than competing as a frontier lab, xAI is now providing compute to a competitor.

Breakthroughs & Demos

• OpenAI launches GPT-Realtime-2 voice model. <cite index="51-5">GPT-Realtime-2, OpenAI's first voice model with GPT-5-class reasoning that can handle harder requests and carry the conversation forward naturally.</cite>^[12] <cite index="51-21">The model can call multiple tools at once and make those actions audible with phrases like "checking your calendar" or "looking that up now," helping agents stay responsive while completing tasks.</cite>^[13]

• Anthropic launches dreaming for Claude Managed Agents. <cite index="23-7">Dreaming is a scheduled process that reviews your agent sessions and memory stores, extracts patterns, and curates memories so your agents improve over time.</cite>^[14] <cite index="30-7">Legal AI company Harvey saw task completion rates increase roughly 6x after implementing dreaming.</cite>^[15]

• Anthropic ships outcomes and multiagent orchestration to public beta. <cite index="23-28,23-29,23-30">With outcomes, you write a rubric describing what success looks like and the agent works toward it. A separate grader evaluates the output against your criteria in its own context window, so it isn't influenced by the agent's reasoning. When something isn't right, the grader pinpoints what needs to change and the agent takes another pass.</cite>^[16]

• Mozilla fixes 271 Firefox vulnerabilities via Claude Mythos. <cite index="43-3,43-4">Anthropic applied an early version of Claude Mythos Preview to Firefox. This week's release of Firefox 150 includes fixes for 271 vulnerabilities identified during this initial evaluation.</cite>^[17] <cite index="47-25">271 real vulnerabilities fixed in Firefox 150, with, as distinguished engineer Brian Grinstead put it, "almost no false positives."</cite>^[18] <cite index="49-5,49-6">Anthropic's Claude Mythos Preview found 271 unknown vulnerabilities in Firefox 150, some up to 20 years old. Mozilla's agentic pipeline lets the AI write and run its own test cases to verify findings, and will soon check every new code commit automatically.</cite>^[19]

• NightCafe launches Consistent Character feature. <cite index="31-11,31-12">NightCafe's Consistent Character feature changes that in a way that feels surprisingly approachable. Instead of wrestling with complicated setups or uploading dozens of training images, creators can now build recognizable characters using just a handful of photos.</cite>^[20]

• Google releases Gemma 4 Multi-Token Prediction drafters. <cite index="61-2">By using a specialized speculative decoding architecture, these drafters deliver up to a 3x speedup without any degradation in output quality or reasoning logic.</cite>^[21] <cite index="62-4">MTP drafters use a speculative decoding architecture that pairs a lightweight drafter model with a heavy target model — the drafter proposes several tokens at once, and the target model verifies them all in a single forward pass, breaking the one-token-at-a-time bottleneck.</cite>^[22]

Blockers & Open Questions

• What does xAI's pivot to neocloud status signal about frontier model competitiveness? <cite index="14-4">This also suggests xAI isn't doing much when it comes to training its own frontier AI models, and it's harder for the company to position itself as a "forward-looking, innovative" business when that's the case.</cite>^[23] The deal underscores whether smaller labs can sustain model development against incumbents flush with capital.

• Can Mythos access controls contain unrestricted security-capability deployment? <cite index="48-17">On the same day Anthropic announced Glasswing, a group of unauthorised users gained access to Mythos Preview by guessing the model's URL through a third-party vendor environment, an incident Anthropic said it is investigating.</cite>^[24] Early unauthorized access raises questions about how long gating mechanisms can hold.

• Does Colossus 1's environmental record complicate Anthropic's scaling narrative? <cite index="13-2,13-3">That's the one with the particularly bad environmental record. The gas turbines installed to power the facility initially ran without Clean Air Act permits or pollution control devices, which they got away with by classifying them as "temporary".</cite>^[25] Anthropic now owns the environmental and political liabilities alongside the computational gains.

• How does the Musk v. Altman trial reshape OpenAI's governance after ruling? <cite index="29161b62-0e96-4b60-862a-bd510c927ef9">Musk sues OpenAI over alleged mission drift from nonprofit AI research to profit-driven ChatGPT development; legal battle threatens leadership and operational structure.</cite>^[26] Testimony from Mira Murati and others continues to expose OpenAI's board dynamics and the 2023 ouster.

What to Watch Next Week

• <cite index="21-2">Anthropic launched Claude Managed Agents last month</cite>^[27], and enterprise adoption curves will signal whether agent pricing sticks above current capacity constraints post-Colossus 1 ramp.

• OpenAI's next earnings cycle or investor update will clarify whether Anthropic's $30B ARR claim triggers competitive revaluation. Early enterprise accounts are multi-homing; switching costs remain low.

• The Trump administration's AI safety testing agreements (signed after Anthropic withheld Mythos) will reveal how quickly government agencies operationalize frontier model security evaluation, and whether classified testing changes threat assessment.

• Final rulings in the Musk v. Altman trial could force structural changes to OpenAI's charter or trigger board reconfiguration, with implications for enterprise confidence and talent retention.

The Numbers

• Disclosed funding: $0 (no new VC rounds this week)
• News stories tracked: 28 distinct items across partnership, product launch, security, litigation, and corporate structure categories
• Most-discussed entities: Anthropic (10 mentions), OpenAI (8), xAI/SpaceX (7), Mozilla (4), NightCafe (3)
• Executive moves: Mira Murati deposition testimony; Elon Musk's on-the-record Anthropic team meetings; potential Claude Code leadership announcements at Code w/ Claude event
• Product launches: 6 (dreaming, outcomes, multiagent orchestration, Consistent Character, GPT-Realtime family, Gemma 4 MTP)

Sources

This brief was generated with AI assistance, drawing on the sources cited above. See /disclosures for methodology.