🎙️ I-ElevenLabs ifinyelela inani lamaRandi ayizigidi eziyi-11 ngemuva komdlalo omusha wamaRandi ayizigidi eziyi-500 ↗
I-ElevenLabs isanda kungena esigabeni esithi “lokhu kuya ngokuya kuba nzima” - kuqoqwe u-$500M, inani lika-$11B. Lokho kuyigxathu elikhulu kunenombolo yayo yokugcina exoxwe ngayo esidlangalaleni, futhi kugcizelela ukuthi abatshalizimali basabona kangakanani izwi le-AI njengeplatifomu, hhayi iqhinga le-parlor.
Iphimbo: inkulumo engokoqobo kakhudlwana, izilimi eziningi, izwi lengxoxo "elithinta imizwa" kakhudlwana, kanye nokukhuluma ngendlela ehlekisayo - ngokuyisisekelo kuhloswe ukuhlala ngaphansi kwenqwaba yezindlela zokusebenza zemidiya kanye nezama-ejenti... kungakhathaliseki ukuthi kuhle noma kubi kangakanani.
🧠 I-Cerebras ithole u-$1B ngaphezulu kanye nenani lika-$23.1B emjahweni we-AI chip ↗
AmaCerebras athole u-$1B ekuxhasweni kwezimali okusesigabeni sokugcina, futhi inani lentengo likhulu: $23.1B. Uma ubulokhu uzwa ukuthi “iNvidia ayikwazi ukuba yimpendulo kuphela” izinyanga eziningi, nakhu lokho kuzwakala ngendlela yokubhala isheke.
Babheja ukuthi ihadiwe esezingeni eliphezulu - ama-chips amakhulu okuqeqesha kanye nokucabanga - ingaqhubeka nokuthola isidingo esihlala isikhathi eside njengoba wonke umuntu ezama ukubala. Ingxenye iwukuhlukahluka, ingxenye yokuphelelwa yithemba, ingxenye ethi "ngicela ungavumeli ukuhlinzekwa kwe-GPU kulawule yonke imephu yami," konke ngesikhathi esisodwa.
💸 Amacebo e-Alphabet's AI capex ayakhanga amehlo - futhi inkinga akuyona imali kuphela ↗
Uhlamvu lwezinhlamvu luhlele izinhlelo zokusebenzisa ingqalasizinda ezi... ezingenangqondo ngobukhulu. Isimo sengqondo yilesi: qhubeka uthela ukhonkolo, qhubeka uthenga ama-chips, qhubeka wandisa izikhungo zedatha - ngoba i-AI ayisebenzi kuma-vibes, isebenza ngamandla kanye ne-silicon.
Kukhona okuthile okuqinisekisayo kancane - futhi okwesabisayo: ngisho nangalolo hlobo lwesabelomali, imikhawulo yokuhlinzeka isabalulekile. Imali iyasiza, impela - kodwa awukwazi ukuthola ngokushesha ama-transformer, umthamo wegridi, noma izikhungo zedatha ezintsha eziyinkulungwane ezivela emoyeni ongenalutho.
🎓 I-Adaption Labs kaSara Hooker ithole imbewu engu-$50M yokwakha amamodeli “okufunda ngokushesha” ↗
Ama-Adaption Labs aphume phambili ethola imali engu-$50M, eholwa ngumqondo wokuthi amamodeli amancane, ahlakaniphile ajwayela ngokushesha angase anqobe izinga eliphezulu ezindaweni eziningi zangempela.
Ukubheja okuyisisekelo kubukhali: esikhundleni sokuqeqeshwa kusengaphambili okukhulu kuze kube phakade, gxila ezinhlelweni eziqhubeka nokufunda kahle. Kungaba yisigaba esilandelayo esinengqondo… noma umzamo wesibindi wokuhamba umjaho wezikhali ze-GPU, kuye ngesimo sakho sengqondo.
🧾 Isivumelwano sekhompyutha se-OpenAI seMicrosoft siphenduka indaba eyingozi kubatshalizimali ↗
Umbono kaBloomberg: abatshalizimali sebeqala ukuhlanganisa ubudlelwano beMicrosoft ne-OpenAI njenge-jackpot eqinisekisiwe kodwa njengobuso bengozi - izindleko, izibopho, ukubusa, yonke inqwaba ephithene.
Lokhu akukhona ukuthi “ubambiswano lubi,” impela - kufana nokuthi, lapho izindleko sezikhula ngokwanele, ngisho nenzuzo yamasu ingaqala ukuzwakala sengathi iwumthwalo wemfanelo. Kufana nokuba nehhashi lomjaho eliqhubeka liwina… ngenkathi lidla indlu yakho.
📜 Umfutho woMthetho we-EU AI - ikhodi yokucaca ewuhlaka yezindawo zokuqukethwe ezikhiqizwe yi-AI ↗
Ikhodi Yokusebenza Esalungiswa mayelana nokucaca kokuqukethwe okukhiqizwe yi-AI noma okuguquliwe ijikeleza, ixhumene nendlela umkhiqizo we-AI okufanele ubhalwe futhi uphathwe ngayo. Akuyona isihloko esikhanga kakhulu, kodwa uhlobo "lwesendlalelo samaphepha" olugcina lubumba izinqumo zomkhiqizo ngokushesha.
Uma wakha noma uthumela izinto ezikhiqizayo, lokhu kukufuqa ukuthi uqondise ekuqondisweni okwengeziwe kokumaka/ukulebula - futhi mhlawumbe ukuhlolwa kwezimali kanye nokubhala imibhalo okuningi kunalokho umuntu akufunayo ngoLwesihlanu. (Kodwa... yebo, kuyeza.)
Imibuzo Evame Ukubuzwa
Intengo ye-ElevenLabs engu-$11B ithini mayelana nokuthi izwi le-AI liya kuphi?
Kusikisela ukuthi abatshalizimali babona izwi le-AI njengengqalasizinda eyinhloko yemikhiqizo yemidiya neyesitayela sama-ejenti, hhayi isici esisha. Kugxilwe enkulumweni engokoqobo, yezilimi eziningi, eveza imizwa efinyelela kalula ekusebenzeni okuphindaphindwayo kanye nokuxoxa. Ezindleleni eziningi, lokho kwenza izwi libe ungqimba olungasetshenziswa kabusha kuzo zonke izinhlelo zokusebenza, kunokuba libe ikhono lokubonisa kanye kuphela.
Kufanele ngicabange kanjani ngokwanda kokuxhaswa ngezimali kwe-AI njenge-ElevenLabs kanye ne-Cerebras ngokwezinto ezingokoqobo?
Izigameko ezinkulu zivame ukubonisa ukuthi imakethe ilindele ukuthi imali eningi neqhubekayo ekubaleni, idatha, kanye nokusabalalisa iwine. Kwabakhi, lokho kuvame ukuhunyushwa njengokuphindaphinda okusheshayo komkhiqizo okuvela kubathengisi abaxhaswe kahle, kanye nokuncintisana okubukhali ngentengo kanye nokusebenza. Kungabonisa futhi ukuthi izigaba "zesikhulumi" - izwi, ama-chip, ingqalasizinda - yilapho kwakhiwa khona izikhundla ezivikelekile.
Iyini indlela kaCerebras yokusebenzisa isikali se-wafer, futhi kungani abantu bebheja kuyo manje?
I-Cerebras ibeka ama-chip amakhulu, anesilinganiso se-wafer sokuqeqesha kanye nokucabanga njengendlela ehlukile yokuhlangabezana nesidingo sokubala. Ukubheja ukuthi ihadiwe ekhethekile ingakha izindawo eziqinile ngenkathi amaqembu efuna izinketho ngale kochungechunge olulodwa oluphezulu lokuhlinzeka nge-GPU. Empeleni, kuyingxenye yesu lokwehlukahluka kanye nokuphuthuma kwengxenye yokuqinisekisa amandla athembekile.
Kungani i-Alphabet ingachitha imali eningi kwingqalasizinda ye-AI kodwa isabhekene nemikhawulo yokuhlinzekwa?
Ngoba ukukalwa kwe-AI kunqunyelwe yizithiyo ezibonakalayo, hhayi isabelomali kuphela. Ukutholakala kwamandla, ukwakheka kwesikhungo sedatha, kanye nokufinyelela kuma-chips nezingxenye kungathatha isikhathi ukunwebeka. Ngisho noma une-capex enamandla, awukwazi ukwengeza ngokushesha umthamo wegridi noma ukusheshisa yonke ingxenye yehadiwe kanye nepayipi yokwakha ngesikhathi esisodwa.
Ziyini izinhlobo “zokufunda ngokushesha”, futhi zingase zinqobe nini izinhlobo ezinkulu eziqeqeshwe kusengaphambili?
Ziyizinhlelo eziklanyelwe ukuzivumelanisa nezimo ngempumelelo ngemva kokusetshenziswa, kunokuthembela kuphela ekuqeqeshweni kwangaphambi kwesikhathi okukhulu. Ezimweni eziningi zokukhiqiza, ukuzivumelanisa nezimo okusheshayo kungabaluleka kakhulu kunesilinganiso esingavuthiwe, ikakhulukazi lapho idatha ishintsha noma imisebenzi ishintsha. Indlela evamile ukugcina amamodeli emancane nokwenza ukufunda noma ukubuyekeza kusebenze kahle ekukhiqizeni.
Imizamo yokubonisa izinto obala ye-EU AI Act ithinta kanjani okuqukethwe okukhiqizayo kwamaqembu okuthumela?
Basunduzela imikhiqizo ekulebheni okucacile nasekuphatheni imiphumela ekhiqizwe yi-AI noma eguquliwe. Ezinhlanganweni eziningi, lokho kuholela ekuqondisweni okwengeziwe kokumaka noma ukudalula, kanye nemikhuba eqinile yokubhala imibhalo kanye nokuhlola. Uma usebenzisa imithombo yezindaba ekhiqizayo, kuwukuhlakanipha ukuhlela ukulandelela imvelaphi nokwakha imisebenzi yokuthobela imithetho elula kusenesikhathi.