Investing $5 Billion: Cerebras Uses a 'Redemption Bond' to Enter Sam Altman's Hall of Gods

marsbitDipublikasikan tanggal 2026-05-14Terakhir diperbarui pada 2026-05-14

Abstrak

Cerebras Systems, the chipmaker known for its massive wafer-scale WSE-3 processor, is making a dramatic IPO push with a $48 billion valuation. This move follows a crucial deal with OpenAI, where Cerebras will provide 750 megawatts of computing power over three years. In exchange, OpenAI receives warrants for a 10% stake in Cerebras, valued at around $5 billion. This agreement, termed a "computing ransom note," secures Cerebras a place within Sam Altman's influential network while sacrificing a significant portion of its potential profits from the deal. The partnership has supercharged Cerebras's market prospects, granting it a premium valuation and positioning it as a core OpenAI partner. The article frames this as part of "Altman Economics," where OpenAI leverages its model sovereignty to extract concessions from hardware suppliers, reshaping the AI hardware landscape into a hierarchy of sovereigns and vassals.

On Wall Street's monitors, the flames of the digital oracle burn ever brighter.

This week, the chip industry's 'maverick', Cerebras Systems, charged towards an IPO with a valuation of $48 billion

This is the largest listing of 2026, a commercial ritual imbued with both sanctity and brutality.

To squeeze into Sam Altman's 'Hall of Gods', Cerebras personally cut away 10% of its own 'flesh and blood'—stock warrants worth approximately $5 billion.

This is no longer a simple IPO. It is a tribute from underlying hardware to model sovereignty, paid in the twilight of an era.

And standing behind all this prosperity is the man known as the 'new lord of Silicon Valley'.

The Return of Brutalist Aesthetics

It Forged a 'Silicon-based Leviathan'

If Nvidia proudly calls its B200 chip an exquisite postage stamp, then Cerebras's flagship product, WSE-3, is a massive dinner plate.

For the past three decades, the semiconductor industry's creed has been 'miniaturization'—dancing on a pinhead, stacking transistors at the nanoscale.

But Cerebras took a counterintuitive path: Since memory bandwidth is the Achilles' heel of AI inference, why not just make the entire wafer into a single chip.

This 'silicon-based leviathan' is nearly half a square foot in size, featuring over 4 trillion transistors (19 times that of Nvidia's B200), 900,000 AI-optimized cores, and 125 petaflops of AI compute—making its total computing power 28 times that of the B200.

Its design is fundamentally about 'trading space for time': etching cores directly onto a single 12-inch wafer, flattening the 'memory wall' where data shuttles between the chip and external storage.

This brutalist aesthetic shows terrifying dominance in the 'Age of Inference':

When running top-tier models from OpenAI, Cognition, Meta, etc., it can achieve a throughput speed of up to 3000 tokens per second, outpacing traditional GPUs by 70 times.

However, in 2026, the hardest technology, if it can't be converted into the fastest Tokens, is just expensive sand.

Eighteen months ago, Cerebras was a forlorn geek.

In 2025, its revenue was only $510 million, almost entirely from a single customer, the UAE's G42. It operated at a loss of $145.9 million, forcing the postponement of its IPO plans.

Despite possessing performance surpassing Nvidia's, its fate hung on the checkbook of a single client, like an orphan at constant risk of drowning.

Until it knocked on the door leading to Sam Altman's inner circle.

The $5 Billion 'Redemption Bond'

The Golden Ticket to the Future

In the world of AI, Moore's Law is giving way to the 'Law of the Circle'.

In exchange for survival and explosive growth, Cerebras signed an agreement that left outsiders dumbfounded: providing OpenAI with 750 megawatts of computing capacity over the next three years.

Calculated at current market prices (750 MW is roughly worth $9 billion annually), this deal could bring about $27 billion in revenue and $10 billion in gross profit.

For a company still in the red last year, this was a lifeline.

More tellingly, these 750 megawatts constitute just one-fortieth of the total computing power OpenAI believes it will need by 2030. This leftover order is enough to catapult Cerebras from a 'marginal player' to a market focal point.

But there is no free lunch, only expensive endorsements.

In exchange, Cerebras must gradually grant stock warrants to OpenAI. Upon exercise, OpenAI would hold 10% of Cerebras's shares—worth about $5 billion at the offering price midpoint of $155.

This means Cerebras generously offered up half the potential future profits from this mega-deal as a 'tribute' directly to OpenAI.

This is not an equal exchange. This is the AI era's 'computing power redemption bond'.

OpenAI is the 'lord' holding the power to allocate traffic and demand, while Cerebras is a heavily armored knight providing advanced tools of production—to enter the lord's domain, the knight must pay homage and share half his spoils.

Once branded as OpenAI's core blood ally, Cerebras's price-to-sales (P/S) ratio instantly skyrocketed to 15 times, a level even Nvidia must respectfully acknowledge.

'Altmanomics'

He Is Cashing Out the Hardware Giants

Cerebras is not the only computing power enterprise bowing to Sam Altman.

Last October, AMD's stock doubled after reaching a similar agreement with OpenAI; earlier, Nvidia was rumored to have deepened its ties with OpenAI through a $30 billion financing round.

A clear map of power is emerging: OpenAI is evolving from a software company into a global 'tax authority' over the AI hardware landscape.

Sam Altman's true ambition is not merely to acquire data centers for training and running its models, but to quietly expand the group of companies and investors who have a 'vested interest in OpenAI's success.'

He is weaving a vast web of an 'interest community'.

OpenAI doesn't buy chips; it only rents the future—and in the process, binds all players to the same ship.

For hardware vendors: Even with profits halved, obtaining the 'OpenAI certification' and a 15x P/S valuation is a worthwhile deal.

For Sam Altman: He not only secures the computational foundation for the next decade but also becomes a shareholder in all potential competitors.

This is the essence of 'Altmanomics': In the second half of the AI race, whoever defines the demand for models holds the 'power to tax' the underlying supply chain.

Can $48 Billion Buy a Seat at the Power Table?

This Thursday, Cerebras's IPO received 20 times oversubscription. The pricing range was raised to $150-$160 per share, implying a valuation of $48 billion, coinciding with semiconductor stocks rising 53% cumulatively from their March lows.

The capital market's frenzy is almost blind. They no longer care whether Cerebras lost $140 million last year or if its P/S ratio has already surpassed Nvidia's.

They are buying a kind of 'certainty'.

In the uncertain landscape of 2026, what identity tag is more certain than being a 'Core Partner of OpenAI'?

This is precisely the most intriguing aspect of this IPO. Cerebras's high valuation is essentially a market premium for the 'Altman endorsement'.

Conclusion: Sovereignty or Vassalage?

This 2026 listing marks the semiconductor industry's entry into a brutal era of stratification.

Future chip companies have only two paths:

Either, like Nvidia, rely on first-mover advantage to build their own ecological empire and become the rule-maker; or, like Cerebras, offer up half their soul in a high-stakes gamble for a ticket into the core circle.

If you cannot become a god, then become the most expensive offering.

This is not just a technological battle; it's a battle for sovereignty.

In every megawatt of electricity flows the 'tribute' from chipmakers to Silicon Valley's new overlord.

References:

https://x.com/FT/status/2054141078710768006

https://www.ft.com/content/3f77f8ad-16b8-4f97-ae55-0bd2e31122fa?syn-25a6b1a6=1

https://www.theinformation.com/newsletters/ai-agenda/startup-helping-openai-optimize-ai-cerebras-chips

https://www.morningstar.com/stocks/why-ai-chip-designer-cerebras-is-2026s-hottest-ipo-yet

This article is from the WeChat public account "Xin Zhi Yuan," author: Xin Zhi Yuan, editor: KingHZ

Pertanyaan Terkait

QWhat is the core technology and design philosophy behind Cerebras Systems' WSE-3 chip, and how does it differ from traditional chip designs?

ACerebras Systems' WSE-3 chip employs a radical design philosophy that contrasts with the traditional industry focus on miniaturization. Instead of shrinking components, it uses an entire 12-inch silicon wafer to create a single, massive chip. This 'brute-force' approach, described as 'trading space for time,' is aimed at eliminating the 'memory wall' bottleneck by providing immense on-chip memory bandwidth. It contains over 4 trillion transistors (19x more than NVIDIA's B200), 900,000 AI-optimized cores, and delivers 125 petaflops of AI computing power. Its key difference is its monolithic wafer-scale design versus the traditional multi-chiplet or discrete GPU designs, prioritizing high-speed data access for AI inference.

QWhat was the crucial strategic deal Cerebras made with OpenAI to enable its 2026 IPO, and what were the key terms?

AThe crucial deal was a multi-year agreement where Cerebras committed to supply 750 megawatts of computing capacity to OpenAI. In exchange for this massive, revenue-guaranteeing contract—valued at approximately $27 billion in revenue and $10 billion in gross profit over three years—Cerebras granted OpenAI warrants. Upon exercise, these warrants would give OpenAI a 10% equity stake in Cerebras, valued at around $5 billion based on the IPO's mid-range share price. This deal, framed as a 'ransom note' or 'tribute,' provided Cerebras with the financial credibility and market validation needed for a successful high-valuation IPO.

QHow does the article characterize the evolving power dynamic in the AI industry, particularly regarding OpenAI's role?

AThe article characterizes the power dynamic as a shift where OpenAI is evolving from a software company into a dominant 'tax authority' or 'feudal lord' over the global AI hardware landscape. By securing large, long-term compute capacity deals from hardware makers like Cerebras, AMD, and NVIDIA—often in exchange for equity or favorable terms—OpenAI is locking in its future compute foundation. More importantly, it is weaving a vast 'community of interest,' binding these suppliers to its success. This grants OpenAI 'taxation rights' over the hardware supply chain, as it defines model demand and controls access to the most lucrative market segments.

QWhat does the term 'Altmanomics' (Altman Economics) refer to in the context of this article?

AIn this article, 'Altmanomics' refers to the strategic and economic model pioneered by OpenAI's Sam Altman. Its essence is that in the later stages of the AI race, the entity that defines the demand for AI models gains the power to 'tax' the underlying hardware supply chain. Instead of merely purchasing chips, OpenAI 'rents the future' by making long-term capacity deals that often include equity stakes. This strategy allows OpenAI to secure its computational base at favorable rates while simultaneously making shareholders and partners out of potential hardware competitors, thereby consolidating its ecosystem power and ensuring alignment of interests.

QAccording to the article's conclusion, what two paths are available for semiconductor companies in the new AI era, as exemplified by NVIDIA and Cerebras?

AThe article concludes that semiconductor companies now face two distinct paths: 1) Become a rule-maker like NVIDIA, which uses first-mover advantage and ecosystem strength to establish its own empire and dictate distribution rules. 2) Become a vassal or 'high-value offering' like Cerebras, which trades a significant portion of its potential sovereignty and profits (through equity and deals) to a dominant AI platform (OpenAI) in exchange for a guaranteed, high-value ticket into the core industry circle. The choice is between maintaining sovereign control or becoming a strategically important, yet subordinate, partner to the new 'lords' of silicon.

Bacaan Terkait

Mampukah DeepSeek Menghemat 1 Triliun Dolar untuk China?

**Ringkasan Artikel: Apakah DeepSeek Dapat Menghemat 1 Triliun Dolar AS untuk China?** Artikel ini membahas bagaimana DeepSeek, melalui serangkaian inovasi teknologinya, berpotensi secara drastis menekan biaya infrastruktur AI China dan berpotensi menghemat investasi hingga triliunan dolar AS. **Tantangan: Biaya AI yang Melonjak** Biaya perangkat keras AI, terutama memori berbandwidth tinggi (HBM), sedang meroket. Misalnya, pada sistem AI canggih NVIDIA Vera Rubin, biaya komponen memori saja mencapai sekitar $2 juta dan telah naik 435% dalam setahun. **Solusi DeepSeek: Tiga Inovasi Utama** DeepSeek mendekati masalah ini dengan tiga terobosan teknis utama untuk meningkatkan efisiensi token perangkat keras hingga 4 kali lipat: 1. **Kompresi Memori (KV Cache):** Melalui **Multi-head Latent Attention (MLA)**, DeepSeek mengompres "memori" konteks panjang model secara signifikan (hingga 90%+ pengurangan) tanpa mengorbankan kualitas, mengurangi ketergantungan pada HBM yang mahal. 2. **Aktivasi "Tubuh" Model Sesuai Kebutuhan:** Menggunakan arsitektur **Mixture of Experts (MoE)**, model hanya mengaktifkan sebagian kecil dari total parameternya untuk setiap tugas. Ini seperti hanya memanggil dokter spesialis yang relevan, bukan seluruh rumah sakit. 3. **Cache dan Penggunaan Ulang:** Hasil komputasi yang sudah dihitung disimpan dan digunakan kembali (cache hit), menghindari perhitungan berulang yang mahal. DeepSeek menerapkan ini bahkan dalam struktur harga layanannya. **Dampak dan Potensi Penghematan** Efisiensi ini mengubah ekonomi infrastruktur AI: * **Pengurangan Biaya Operasional:** Untuk tugas tertentu (misalnya, membaca ulang kode panjang), biaya menggunakan DeepSeek V4-Pro bisa **100 kali lebih murah** dibandingkan model seperti GPT-5.5 atau Claude Opus. * **Penghematan Infrastruktur Berskala Besar:** Artikel memperkirakan, dengan peningkatan efisiensi 4x, untuk memenuhi permintaan token harian masa depan China yang mencapai ribuan triliun, pendekatan DeepSeek berpotensi menghemat investasi setara dengan **puluhan ribu pusat komputasi cerdas**, yang nilainya bisa mendekati **$1 triliun**. * **Pergeseran Strategis:** Alih-alih hanya bergantung pada chip komputasi canggih (di mana China tertinggal), DeepSeek mengalihkan beban ke manajemen memori, cache, dan rekayasa sistem — area di mana industri China seperti produsen memori DRAM (contoh: CXMT) semakin kompetitif. Ini meningkatkan keamanan rantai pasokan. **Kesimpulan** DeepSeek tidak menghilangkan kebutuhan akan perangkat keras canggih, tetapi secara radikal mengurangi ketergantungan marjinal padanya. Dengan membuat AI lebih terjangkau melalui efisiensi teknis mendalam, DeepSeek berpotensi membentuk ulang buku besar infrastruktur AI China, menghemat investasi besar, dan yang terpenting, membuat daya AI dapat diakses oleh industri luas di China. *(Catatan: Artikel ini didasarkan pada analisis industri dan proyeksi, bukan fakta yang sudah pasti.)*

marsbit41m yang lalu

Mampukah DeepSeek Menghemat 1 Triliun Dolar untuk China?

marsbit41m yang lalu

Menggugat Gagasan Utama Menangani Halusinasi: Metakognisi, Solusi Baru untuk Menghancurkan Halusinasi Model Besar

Peneliti Google mengusulkan pendekatan baru untuk mengatasi "halusinasi" AI dalam makalahnya, **"Hallucinations Undermine Trust; Metacognition is a Way Forward."** Alih-alih berusaha membuat AI mengetahui segalanya atau menolak menjawab banyak pertanyaan (yang menimbulkan "pajak utilitas"), makalah ini mendefinisikan ulang halusinasi sebagai **AI memberikan informasi yang salah dengan keyakinan tinggi padahal ia tidak yakin.** Solusi intinya adalah **"ketidakpastian yang setia" (faithful uncertainty)** atau **metakognisi** – kemampuan AI untuk merasakan tingkat keyakinan internalnya dan mengekspresikannya dengan jujur dalam bahasa. Misalnya, jika AI ragu, ia harus mengatakan "Saya tidak terlalu yakin." Pendekatan ini lebih realistis karena hanya mengharuskan keselarasan antara sinyal internal dan output AI, bukan kebenaran mutlak yang mustahil. Metakognisi menjadi sangat penting di era **Agent AI** yang menggunakan alat eksternal seperti mesin pencari. Tanpanya, Agent tidak dapat memutuskan kapan harus mencari, kapan harus berhenti, atau bagaimana mengevaluasi informasi yang ditemukan. Namun, implementasinya menghadapi tantangan seperti **"paradoks bootstrap"** (data pelatihan yang statis untuk kemampuan dinamis), **"sinyak perusak keselarasan"** dari pelatihan RLHF yang mendorong AI selalu tampak yakin, dan kesulitan menilai apakah AI benar-benar memiliki metakognisi atau hanya berpura-pura. Makalah ini menyerukan perubahan paradigma: fokus pada pengembangan AI yang jujur tentang batas pengetahuannya, bukan AI yang sempurna tanpa kesalahan. Kepercayaan datang dari kejujuran tentang ketidakpastian, bukan dari klaim kepastian yang salah.

marsbit47m yang lalu

Menggugat Gagasan Utama Menangani Halusinasi: Metakognisi, Solusi Baru untuk Menghancurkan Halusinasi Model Besar

marsbit47m yang lalu

Belilah Emas dan Minyak untuk Hedging, AI untuk Pertumbuhan Cepat, Bitcoin yang 'Ketinggalan Zaman' Masuki Pasar Beruang

Bitcoin terus mengalami penurunan, mencapai level terendah dalam dua bulan di sekitar $66.123. Menurut analis, alasan di balik penurunan ini lebih dalam daripada sekadar aliran keluar dana ETF atau penjualan oleh Strategy. Inti masalahnya adalah Bitcoin kini kalah bersaing di tiga bidang utama. Pertama, sebagai lindung nilai inflasi, emas dan saham energi kini lebih disukai karena memiliki dukungan fisik dan logika yang lebih langsung. Kedua, untuk pertumbuhan tinggi, investor beralih ke perusahaan AI yang memiliki pendapatan dan laba nyata, di mana Bitcoin tidak menghasilkan arus kas. Ketiga, bahkan di dalam ekosistem kripto, eksposur kini dapat diperoleh melalui stablecoin, pertukaran, atau infrastruktur yang terkait dengan adopsi nyata, bukan hanya Bitcoin. Narasi "emas digital" Bitcoin tidak lagi mendapatkan reaksi positif otomatis terhadap kekhawatiran inflasi. Aliran keluar ETF dan penjualan oleh institusi mencerminkan gejala dari realitas baru: investor kini lebih selektif. Mereka memiliki lebih banyak pilihan dan mempertanyakan nilai tambah Bitcoin. Logika bearish baru untuk Bitcoin bukan lagi tentang penipuan atau kegagalan teknologi, melainkan bahwa kelangkaan saja tidak lagi cukup untuk menarik modal. Bitcoin terjepit di zona tengah yang canggung, bukan sebagai aset lindung nilai terbaik, aset pertumbuhan terbaik, atau satu-satunya pilihan kripto.

华尔街日报51m yang lalu

Belilah Emas dan Minyak untuk Hedging, AI untuk Pertumbuhan Cepat, Bitcoin yang 'Ketinggalan Zaman' Masuki Pasar Beruang

华尔街日报51m yang lalu

Terlewat Gelombang Kenaikan Saham AI, Sekarang Harus Bagaimana?

Penulis: Think AI, Aaron Saham AI telah naik selama tiga setengah tahun tanpa tanda-tanda berhenti. Mereka yang memperkirakan bubble AI atau kejatuhan pasar saham AS tahun lalu kini merenung. Pasar global menunjukkan kinerja kuat: indeks Korea telah circuit breaker 19 tahun ini, naik 4 kali lipat sejak tahun lalu. Saham SK Hynix naik 260%. Micron AS, raksasa memori, tembus valuasi $1 triliun. SoftBank Jepang, dengan taruhan besar pada AI, menjadi perusahaan bernilai tertinggi di Jepang. Di China, perusahaan seperti Yushu dan ChangXin akan masuk bursa dengan valuasi besar. Namun, situasi lain muncul: mereka yang meminjam untuk beli emas awal tahun masih rugi, saham konsumen China terus turun. Banyak investor merasa cemas karena melewatkan rally AI ini. Data menunjukkan, pada 2025, hanya 18.9% investor ritel A股 yang untung, 81.1% rugi. Saat pasar koreksi awal 2026, keyakinan pada AI goyah. Banyak institusi jual di titik terendah, seperti pemegang saham besar Zhongji Innolight yang jual RMB 4.9 miliar, lalu sahamnya naik 35%. Laporan menunjukkan kerugian potensial institusi karena jual terlalu awal di sektor komputasi AI bisa lebih dari RMB 200 miliar. Beberapa institusi yang bertahan justru salah arah, fokus pada aplikasi vertikal seperti AI pendidikan/kesehatan yang turun >20%, sementara sektor komputasi naik >50%. Di luar negeri, kesalahan serupa terjadi. Bridgewater jual besar-besaran saham seperti NVIDIA, Alphabet akhir 2025, tepat sebelum mereka rata-rata naik >80%. Posisi short di pasar AS mencapai level tertinggi sejak 2012. Warren Buffett juga banyak memegang kas, melewatkan rally AI. Namun, AI tetap dianggap sebagai peluang revolusioner paling pasti – bukan konsep jangka pendek, tapi revolusi infrastruktur seperti listrik atau internet. Bagi yang melewatkan gelombang pertama, masih ada peluang di lapisan aplikasi atau gelombang infrastruktur berikutnya. China punya ruang unik dalam pengembangan AI mandiri, penerapan, dan penyempurnaan rantai pasokan. Pemimpin industri memberikan perspektif. Ma Huateng (Tencent) menyatakan pentingnya fokus pada keunggulan sendiri di era AI, bukan sekadar mengejar tren. Jack Ma menyebut AI sebagai peluang revolusi industri setara penemuan listrik, dan era AI baru dimulai. Masih banyak peluang jangka panjang di sepanjang rantai industri, dari infrastruktur komputasi, platform model besar, hingga aplikasi AI sektoral. Peluang struktural berikutnya akan datang bagi mereka yang siap dan memiliki pola pikir stabil. Kuncinya adalah berpegang pada keunggulan sendiri dan berpikir jangka panjang.

marsbit1j yang lalu

Terlewat Gelombang Kenaikan Saham AI, Sekarang Harus Bagaimana?

marsbit1j yang lalu

Trading

Spot
Futures

Artikel Populer

Cara Membeli GODS

Selamat datang di HTX.com! Kami telah membuat pembelian Gods Unchained (GODS) menjadi mudah dan nyaman. Ikuti panduan langkah demi langkah kami untuk memulai perjalanan kripto Anda.Langkah 1: Buat Akun HTX AndaGunakan alamat email atau nomor ponsel Anda untuk mendaftar akun gratis di HTX. Rasakan perjalanan pendaftaran yang mudah dan buka semua fitur.Dapatkan Akun SayaLangkah 2: Buka Beli Kripto, lalu Pilih Metode Pembayaran AndaKartu Kredit/Debit: Gunakan Visa atau Mastercard Anda untuk membeli Gods Unchained (GODS) secara instan.Saldo: Gunakan dana dari saldo akun HTX Anda untuk melakukan trading dengan lancar.Pihak Ketiga: Kami telah menambahkan metode pembayaran populer seperti Google Pay dan Apple Pay untuk meningkatkan kenyamanan.P2P: Lakukan trading langsung dengan pengguna lain di HTX.Over-the-Counter (OTC): Kami menawarkan layanan yang dibuat khusus dan kurs yang kompetitif bagi para trader.Langkah 3: Simpan Gods Unchained (GODS) AndaSetelah melakukan pembelian, simpan Gods Unchained (GODS) di akun HTX Anda. Selain itu, Anda dapat mengirimkannya ke tempat lain melalui transfer blockchain atau menggunakannya untuk memperdagangkan mata uang kripto lainnya.Langkah 4: Lakukan trading Gods Unchained (GODS)Lakukan trading Gods Unchained (GODS) dengan mudah di pasar spot HTX. Cukup akses akun Anda, pilih pasangan perdagangan, jalankan trading, lalu pantau secara real-time. Kami menawarkan pengalaman yang ramah pengguna baik untuk pemula maupun trader berpengalaman.

184 Total TayanganDipublikasikan pada 2024.12.13Diperbarui pada 2026.06.02

Cara Membeli GODS

Diskusi

Selamat datang di Komunitas HTX. Di sini, Anda bisa terus mendapatkan informasi terbaru tentang perkembangan platform terkini dan mendapatkan akses ke wawasan pasar profesional. Pendapat pengguna mengenai harga GODS (GODS) disajikan di bawah ini.

活动图片